diff --git a/.env.exemple b/.env.exemple index 78d00cf17..480752abb 100644 --- a/.env.exemple +++ b/.env.exemple @@ -1,6 +1,6 @@ -GRAPHRAG_API_KEY="fd7763c639184c20962857cacacd1a38" -GRAPHRAG_API_BASE="https://gpt-4o-fr.openai.azure.com" #optionnal , needed if using custom endpoint -GRAPHRAG_API_VERSION="turbo-2024-04-09" -GRAPHRAG_LLM_MODEL="gpt-4" -GRAPHRAG_DEPLOYMENT_NAME="gpt-4" -GRAPHRAG_EMBEDDING_MODEL="text-embedding-ada-002" \ No newline at end of file +GRAPHRAG_API_KEY="your_api_key_here" +GRAPHRAG_API_BASE="your_api_base_url_here" +GRAPHRAG_API_VERSION="your_api_version_here" +GRAPHRAG_LLM_MODEL="your_llm_model_here" +GRAPHRAG_DEPLOYMENT_NAME="your_deployment_name_here" +GRAPHRAG_EMBEDDING_MODEL="your_embedding_model_here" diff --git a/.gitignore b/.gitignore index ec7f425ad..79c51e132 100644 --- a/.gitignore +++ b/.gitignore @@ -38,7 +38,7 @@ tests/fixtures/*/cache tests/fixtures/*/output lancedb/ graphfleet/cache - +graphfleet/libs/graphrag/ # Random .DS_Store *.log* diff --git a/.vscode/settings.json b/.vscode/settings.json index 70f371814..36a37c060 100644 --- a/.vscode/settings.json +++ b/.vscode/settings.json @@ -24,5 +24,6 @@ "**/*tsconfig.json": true, ".cache": true, "retool.config.json": true - } + }, + "python.analysis.autoImportCompletions": true } \ No newline at end of file diff --git a/README.md b/README.md index 6bb40e5c9..562624168 100644 --- a/README.md +++ b/README.md @@ -81,13 +81,16 @@ GraphFleet uses knowledge graphs to provide substantial improvements in question 1. Configuration: Environment Variables: Set up your environment variables in a .env file (refer to the .env.example file for available options). Key variables include: - ```sh - export GRAPHRAG_API_KEY="your API key" - export GRAPHRAG_API_BASE="your API endpoint if you use Azure OpenAI" - export GRAPHRAG_API_VERSION="2024-04-01-preview" - export GRAPHRAG_LLM_MODEL="gpt-4o or gpt-4" - export GRAPHRAG_DEPLOYMENT_NAME="model name" - export GRAPHRAG_EMBEDDING_MODEL="embedding model" +Fill in the .env file in the root folder and the one in the graphfleet folder. + + ```sh + export GRAPHRAG_API_KEY="your_api_key_here" + export GRAPHRAG_API_BASE="your_api_base_url_here" + export GRAPHRAG_API_VERSION="your_api_version_here" + export GRAPHRAG_LLM_MODEL="your_llm_model_here" + export GRAPHRAG_DEPLOYMENT_NAME="your_deployment_name_here" + export GRAPHRAG_EMBEDDING_MODEL="your_embedding_model_here" + ``` settings.yaml: Customize GraphFleet's behavior further by modifying the settings.yaml file within the graphfleet directory. diff --git a/api.txt b/api.txt deleted file mode 100644 index 42aa71672..000000000 --- a/api.txt +++ /dev/null @@ -1,66 +0,0 @@ -from fastapi import FastAPI, HTTPException -from pydantic import BaseModel -import subprocess -import shlex - -app = FastAPI() - -class QueryRequest(BaseModel): - query: str -class ChatMessage(BaseModel): - role: str - content: str - -class ChatHistory(BaseModel): - messages: list[ChatMessage] - -# In-memory storage for chat history -chat_history = [] - -def process_query(query: str, search_method: str) -> str: - cmd = [ - "python3", "-m", "graphrag.query", - "--root", "./graphfleet", - "--method", search_method, - ] - cmd.append(shlex.quote(query)) - - try: - result = subprocess.run(cmd, capture_output=True, text=True, check=True) - output = result.stdout - response = output.split(f"SUCCESS: {search_method.capitalize()} Search Response:", 1)[-1].strip() - return response - except subprocess.CalledProcessError as e: - raise HTTPException(status_code=500, detail=f"An error occurred: {e.stderr}") - -@app.get("/") -async def read_root(): - return {"message": "Welcome to the GraphFleet API"} - -@app.post("/process_local_query/") -async def process_local_query_endpoint(request: QueryRequest): - response = process_query(request.query, "local") - return {"response": response} - -@app.post("/process_global_query/") -async def process_global_query_endpoint(request: QueryRequest): - response = process_query(request.query, "global") - return {"response": response} - -@app.post("/add_message/") -async def add_message(message: ChatMessage): - chat_history.append(message) - return {"status": "Message added"} - -@app.get("/get_chat_history/") -async def get_chat_history(): - return {"messages": chat_history} - -@app.delete("/clear_chat_history/") -async def clear_chat_history(): - chat_history.clear() - return {"status": "Chat history cleared"} - -if __name__ == "__main__": - import uvicorn - uvicorn.run(app, host="0.0.0.0", port=8001) # Changed port to 8001 diff --git a/app.py b/app.py deleted file mode 100644 index 52255c847..000000000 --- a/app.py +++ /dev/null @@ -1,59 +0,0 @@ -import streamlit as st -import subprocess -import shlex - -def process_query(query: str, search_method: str) -> str: - cmd = [ - "python3", "-m", "graphrag.query", - "--root", "./graphfleet", - "--method", search_method, - ] - cmd.append(shlex.quote(query)) - - try: - result = subprocess.run(cmd, capture_output=True, text=True, check=True) - output = result.stdout - response = output.split(f"SUCCESS: {search_method.capitalize()} Search Response:", 1)[-1].strip() - return response - except subprocess.CalledProcessError as e: - return f"An error occurred: {e.stderr}" - -def display_chat_message(role: str, content: str): - with st.chat_message(role): - st.markdown(content) - -def clear_chat_history(): - st.session_state.messages = [] - -def main(): - st.set_page_config(page_title="GraphFleet", page_icon="šŸ¤–", layout="wide") - st.title("GraphFleet") - - st.markdown(""" - GraphFleet uses knowledge graphs to provide substantial improvements in question-and-answer performance when reasoning about complex information. It addresses limitations of traditional RAG approaches: - """) - - if "messages" not in st.session_state: - st.session_state.messages = [] - - with st.sidebar: - st.header("Settings") - search_method = st.selectbox("Search Method", ["local", "global"], index=0) - if st.button("Clear Chat History"): - clear_chat_history() - - for message in st.session_state.messages: - display_chat_message(message["role"], message["content"]) - - if prompt := st.chat_input("What's your question?"): - st.session_state.messages.append({"role": "user", "content": prompt}) - display_chat_message("user", prompt) - - with st.spinner("Thinking..."): - response = process_query(prompt, search_method) - - st.session_state.messages.append({"role": "assistant", "content": response}) - display_chat_message("assistant", response) - -if __name__ == "__main__": - main() \ No newline at end of file diff --git a/dictionary.txt b/dictionary.txt index 93290f1ca..372ae933c 100644 --- a/dictionary.txt +++ b/dictionary.txt @@ -1,9 +1,3 @@ -<<<<<<< HEAD - -======= -# Team -Alonso ->>>>>>> origin/main # Pythonisms PYTHONPATH @@ -101,15 +95,12 @@ retryer agenerate aembed dedupe -<<<<<<< HEAD GRAPHRAG graphrag GraphFleet graphfleet Qredence - -======= ->>>>>>> origin/main +pydantic # LLM Terms AOAI @@ -136,43 +127,13 @@ docstore datasource devcontainers codebases -<<<<<<< HEAD -gfleetenv - -======= - -# Microsoft -MSRC ->>>>>>> origin/main # Broken Upstream # TODO FIX IN DATASHAPER Arrary # Prompt Inputs -ABILA -Abila -ALHAMIA -Alhamia -Asadi -Aurelians -Bataglani -BATAGLANI -Bratinas -dulce -Durke -Firuzabad -Firuzabad's -FIRUZABAD -Krohaara -KROHAARA -POKRALLY -Tazbah -TIRUZIA -Tiruzia -Tiruzia's -Verdantis -Verdantis's + # English diff --git a/graphfleet/.env.exemple b/graphfleet/.env.exemple index 1dd3b487a..8dde890c3 100644 --- a/graphfleet/.env.exemple +++ b/graphfleet/.env.exemple @@ -1,6 +1,6 @@ -GRAPHRAG_API_KEY="your api key" -GRAPHRAG_API_BASE="base_url" #optionnal , needed if using custom endpoint -GRAPHRAG_API_VERSION="2023-05-13" -GRAPHRAG_LLM_MODEL="gpt-4o" -GRAPHRAG_DEPLOYMENT_NAME="gpt-4o" -GRAPHRAG_EMBEDDING_MODEL="text-embedding-ada-002" \ No newline at end of file +GRAPHRAG_API_KEY="your_api_key_here" +GRAPHRAG_API_BASE="your_api_base_url_here" +GRAPHRAG_API_VERSION="your_api_version_here" +GRAPHRAG_LLM_MODEL="your_llm_model_here" +GRAPHRAG_DEPLOYMENT_NAME="your_deployment_name_here" +GRAPHRAG_EMBEDDING_MODEL="your_embedding_model_here" \ No newline at end of file diff --git a/graphfleet/api/api.py b/graphfleet/api/api.py deleted file mode 100644 index 31ca6c7c9..000000000 --- a/graphfleet/api/api.py +++ /dev/null @@ -1,73 +0,0 @@ -from fastapi import FastAPI, HTTPException -from pydantic import BaseModel -import subprocess -import shlex - -app = FastAPI() - -class QueryRequest(BaseModel): - query: str - -class ChatMessage(BaseModel): - role: str - content: str - -class ChatHistory(BaseModel): - messages: list[ChatMessage] - -# In-memory storage for chat history -chat_history = [] - -# Allowlist of accepted search methods -ALLOWED_METHODS = ["local", "global"] - -def process_query(query: str, search_method: str) -> str: - if search_method not in ALLOWED_METHODS: - raise HTTPException(status_code=400, detail=f"Invalid search method: {search_method}") - - cmd = [ - "python3", "-m", "graphrag.query", - "--root", "./graphfleet", - "--method", search_method, - ] - cmd.append(shlex.quote(query)) - - try: - result = subprocess.run(cmd, capture_output=True, text=True, check=True) - output = result.stdout - response = output.split(f"SUCCESS: {search_method.capitalize()} Search Response:", 1)[-1].strip() - return response - except subprocess.CalledProcessError as e: - raise HTTPException(status_code=500, detail=f"An error occurred: {e.stderr}") - -@app.get("/") -async def read_root(): - return {"message": "Welcome to the GraphFleet API"} - -@app.post("/process_local_query/") -async def process_local_query_endpoint(request: QueryRequest): - response = process_query(request.query, "local") - return {"response": response} - -@app.post("/process_global_query/") -async def process_global_query_endpoint(request: QueryRequest): - response = process_query(request.query, "global") - return {"response": response} - -@app.post("/add_message/") -async def add_message(message: ChatMessage): - chat_history.append(message) - return {"status": "Message added"} - -@app.get("/get_chat_history/") -async def get_chat_history(): - return {"messages": chat_history} - -@app.delete("/clear_chat_history/") -async def clear_chat_history(): - chat_history.clear() - return {"status": "Chat history cleared"} - -if __name__ == "__main__": - import uvicorn - uvicorn.run(app, host="0.0.0.0", port=8001) # Changed port to 8001 diff --git a/graphfleet/api/api.yaml b/graphfleet/api/api.yaml deleted file mode 100644 index ebc82b4b1..000000000 --- a/graphfleet/api/api.yaml +++ /dev/null @@ -1,155 +0,0 @@ -openapi: "3.1.0" -info: - title: "GraphFleet" - version: "0.4.0" -paths: - "/": - get: - summary: "Read Root" - operationId: "read_root__get" - responses: - "200": - description: "Successful Response" - content: - application/json: - schema: {} - "/process_local_query/": - post: - summary: "Process Local Query Endpoint" - operationId: "process_local_query_endpoint_process_local_query__post" - requestBody: - content: - application/json: - schema: - $ref: "#/components/schemas/QueryRequest" - required: true - responses: - "200": - description: "Successful Response" - content: - application/json: - schema: {} - "422": - description: "Validation Error" - content: - application/json: - schema: - $ref: "#/components/schemas/HTTPValidationError" - "/process_global_query/": - post: - summary: "Process Global Query Endpoint" - operationId: "process_global_query_endpoint_process_global_query__post" - requestBody: - content: - application/json: - schema: - $ref: "#/components/schemas/QueryRequest" - required: true - responses: - "200": - description: "Successful Response" - content: - application/json: - schema: {} - "422": - description: "Validation Error" - content: - application/json: - schema: - $ref: "#/components/schemas/HTTPValidationError" - "/add_message/": - post: - summary: "Add Message" - operationId: "add_message_add_message__post" - requestBody: - content: - application/json: - schema: - $ref: "#/components/schemas/ChatMessage" - required: true - responses: - "200": - description: "Successful Response" - content: - application/json: - schema: {} - "422": - description: "Validation Error" - content: - application/json: - schema: - $ref: "#/components/schemas/HTTPValidationError" - "/get_chat_history/": - get: - summary: "Get Chat History" - operationId: "get_chat_history_get_chat_history__get" - responses: - "200": - description: "Successful Response" - content: - application/json: - schema: {} - "/clear_chat_history/": - delete: - summary: "Clear Chat History" - operationId: "clear_chat_history_clear_chat_history__delete" - responses: - "200": - description: "Successful Response" - content: - application/json: - schema: {} -components: - schemas: - ChatMessage: - type: "object" - title: "ChatMessage" - required: - - role - - content - properties: - role: - type: "string" - title: "Role" - content: - type: "string" - title: "Content" - HTTPValidationError: - type: "object" - title: "HTTPValidationError" - properties: - detail: - type: "array" - title: "Detail" - items: - $ref: "#/components/schemas/ValidationError" - QueryRequest: - type: "object" - title: "QueryRequest" - required: - - query - properties: - query: - type: "string" - title: "Query" - ValidationError: - type: "object" - title: "ValidationError" - required: - - loc - - msg - - type - properties: - loc: - type: "array" - title: "Location" - items: - anyOf: - - type: "string" - - type: "integer" - msg: - type: "string" - title: "Message" - type: - type: "string" - title: "Error Type" diff --git a/graphfleet/api/open b/graphfleet/api/open deleted file mode 100644 index e69de29bb..000000000 diff --git a/graphfleet/api/openapi.stainless.yml b/graphfleet/api/openapi.stainless.yml deleted file mode 100644 index dcf9a9f5c..000000000 --- a/graphfleet/api/openapi.stainless.yml +++ /dev/null @@ -1,61 +0,0 @@ -# yaml-language-server: $schema=https://app.stainlessapi.com/config.schema.json - -organization: - name: Qredence - docs: https://docs.qredence.ai - contact: zachary@qredence.ai - github_org: Qredence -resources: - root: - methods: - retrieve: get / - $client: - methods: - process_local_query: post /process_local_query/ - process_global_query: post /process_global_query/ - add_message: post /add_message/ - chat_history: - methods: - list: get /get_chat_history/ - delete: delete /clear_chat_history/ -targets: - node: - package_name: graphfleet - production_repo: null - publish: - npm: false - python: - package_name: graphfleet - production_repo: null - publish: - pypi: false -settings: - license: Apache-2.0 -client_settings: - opts: {} - omit_platform_headers: false - default_headers: {} - default_timeout: PT60S - default_retries: - max_retries: 2 - initial_delay_seconds: 0.5 - max_delay_seconds: 8 -environments: - production: https://localhost:8080/test-api -query_settings: - nested_format: brackets - array_format: comma -readme: - example_requests: - default: - type: request - endpoint: post /add_message/ - params: &ref_0 {} - headline: - type: request - endpoint: post /add_message/ - params: *ref_0 - pagination: - type: request - endpoint: get /get_chat_history/ - params: {} \ No newline at end of file diff --git a/graphfleet/api/openapi.yml b/graphfleet/api/openapi.yml deleted file mode 100644 index 705410381..000000000 --- a/graphfleet/api/openapi.yml +++ /dev/null @@ -1,155 +0,0 @@ -openapi: '3.1.0' -info: - title: 'GraphFleet' - version: '0.4.0' -paths: - '/': - get: - summary: 'Read Root' - operationId: 'read_root__get' - responses: - '200': - description: 'Successful Response' - content: - application/json: - schema: {} - '/process_local_query/': - post: - summary: 'Process Local Query Endpoint' - operationId: 'process_local_query_endpoint_process_local_query__post' - requestBody: - content: - application/json: - schema: - $ref: '#/components/schemas/QueryRequest' - required: true - responses: - '200': - description: 'Successful Response' - content: - application/json: - schema: {} - '422': - description: 'Validation Error' - content: - application/json: - schema: - $ref: '#/components/schemas/HTTPValidationError' - '/process_global_query/': - post: - summary: 'Process Global Query Endpoint' - operationId: 'process_global_query_endpoint_process_global_query__post' - requestBody: - content: - application/json: - schema: - $ref: '#/components/schemas/QueryRequest' - required: true - responses: - '200': - description: 'Successful Response' - content: - application/json: - schema: {} - '422': - description: 'Validation Error' - content: - application/json: - schema: - $ref: '#/components/schemas/HTTPValidationError' - '/add_message/': - post: - summary: 'Add Message' - operationId: 'add_message_add_message__post' - requestBody: - content: - application/json: - schema: - $ref: '#/components/schemas/ChatMessage' - required: true - responses: - '200': - description: 'Successful Response' - content: - application/json: - schema: {} - '422': - description: 'Validation Error' - content: - application/json: - schema: - $ref: '#/components/schemas/HTTPValidationError' - '/get_chat_history/': - get: - summary: 'Get Chat History' - operationId: 'get_chat_history_get_chat_history__get' - responses: - '200': - description: 'Successful Response' - content: - application/json: - schema: {} - '/clear_chat_history/': - delete: - summary: 'Clear Chat History' - operationId: 'clear_chat_history_clear_chat_history__delete' - responses: - '200': - description: 'Successful Response' - content: - application/json: - schema: {} -components: - schemas: - ChatMessage: - type: 'object' - title: 'ChatMessage' - required: - - role - - content - properties: - role: - type: 'string' - title: 'Role' - content: - type: 'string' - title: 'Content' - HTTPValidationError: - type: 'object' - title: 'HTTPValidationError' - properties: - detail: - type: 'array' - title: 'Detail' - items: - $ref: '#/components/schemas/ValidationError' - QueryRequest: - type: 'object' - title: 'QueryRequest' - required: - - query - properties: - query: - type: 'string' - title: 'Query' - ValidationError: - type: 'object' - title: 'ValidationError' - required: - - loc - - msg - - type - properties: - loc: - type: 'array' - title: 'Location' - items: - anyOf: - - type: 'string' - - type: 'integer' - msg: - type: 'string' - title: 'Message' - type: - type: 'string' - title: 'Error Type' diff --git a/graphfleet/in_progress_api/api.py b/graphfleet/in_progress_api/api.py new file mode 100644 index 000000000..045d9b9ea --- /dev/null +++ b/graphfleet/in_progress_api/api.py @@ -0,0 +1,146 @@ +from fastapi import FastAPI, HTTPException, Depends +from fastapi.responses import StreamingResponse +from pydantic import BaseModel +from typing import Optional, Dict, Any +import pandas as pd +from graphrag.config.models.graph_rag_config import GraphRagConfig + +# Import the search functions +from graphrag.query.api import ( + global_search, + local_search, + global_search_streaming, + local_search_streaming +) + +app = FastAPI() + +class SearchRequest(BaseModel): + query: str + index: str + +class Index: + def __init__(self, index_id: str): + self.config = load_config(index_id) + self.nodes_df = load_nodes(index_id) + self.entities_df = load_entities(index_id) + self.community_reports_df = load_community_reports(index_id) + self.text_units_df = load_text_units(index_id) + self.relationships_df = load_relationships(index_id) + self.covariates_df = load_covariates(index_id) + + if not all([self.config, self.nodes_df, self.entities_df, self.community_reports_df, + self.text_units_df, self.relationships_df, self.covariates_df]): + raise ValueError(f"Failed to load all required data for index {index_id}") + +def get_index(index: str, query: str): + try: + return Index(index) + except Exception as e: + raise HTTPException(status_code=404, detail=f"Index not found or failed to load: {str(e)}") + +@app.post("/global_search") +async def api_global_search(request: SearchRequest, index: Index = Depends(get_index)): + try: + result = await global_search( + config=index.config, + nodes=index.nodes_df, + entities=index.entities_df, + community_reports=index.community_reports_df, + community_level=0, + response_type="text", + query=request.query + ) + return result + except Exception as e: + raise HTTPException(status_code=500, detail=f"Search failed: {str(e)}") + +@app.post("/local_search") +async def api_local_search(request: SearchRequest, index: Index = Depends(get_index)): + try: + result = await local_search( + config=index.config, + nodes=index.nodes_df, + entities=index.entities_df, + community_reports=index.community_reports_df, + text_units=index.text_units_df, + relationships=index.relationships_df, + covariates=index.covariates_df, + community_level=0, + response_type="text", + query=request.query + ) + return result + except Exception as e: + raise HTTPException(status_code=500, detail=f"Search failed: {str(e)}") + +@app.post("/global_search_streaming") +async def api_global_search_streaming(request: SearchRequest, index: Index = Depends(get_index)): + try: + stream = global_search_streaming( + config=index.config, + nodes=index.nodes_df, + entities=index.entities_df, + community_reports=index.community_reports_df, + community_level=0, + response_type="text", + query=request.query + ) + return StreamingResponse(stream) + except Exception as e: + raise HTTPException(status_code=500, detail=f"Streaming search failed: {str(e)}") + +@app.post("/local_search_streaming") +async def api_local_search_streaming(request: SearchRequest, index: Index = Depends(get_index)): + try: + stream = local_search_streaming( + config=index.config, + nodes=index.nodes_df, + entities=index.entities_df, + community_reports=index.community_reports_df, + text_units=index.text_units_df, + relationships=index.relationships_df, + covariates=index.covariates_df, + community_level=0, + response_type="text", + query=request.query + ) + return StreamingResponse(stream) + except Exception as e: + raise HTTPException(status_code=500, detail=f"Streaming search failed: {str(e)}") + +# Implement these functions to load index data +def load_config(index_id: str) -> GraphRagConfig: + # Implementation to load config for the given index_id + # Return None if loading fails + pass + +def load_nodes(index_id: str) -> pd.DataFrame: + # Implementation to load nodes for the given index_id + # Return None if loading fails + pass + +def load_entities(index_id: str) -> pd.DataFrame: + # Implementation to load entities for the given index_id + # Return None if loading fails + pass + +def load_community_reports(index_id: str) -> pd.DataFrame: + # Implementation to load community reports for the given index_id + # Return None if loading fails + pass + +def load_text_units(index_id: str) -> pd.DataFrame: + # Implementation to load text units for the given index_id + # Return None if loading fails + pass + +def load_relationships(index_id: str) -> pd.DataFrame: + # Implementation to load relationships for the given index_id + # Return None if loading fails + pass + +def load_covariates(index_id: str) -> pd.DataFrame: + # Implementation to load covariates for the given index_id + # Return None if loading fails + pass \ No newline at end of file diff --git a/graphfleet/in_progress_api/query/api.py b/graphfleet/in_progress_api/query/api.py new file mode 100644 index 000000000..14cc071ed --- /dev/null +++ b/graphfleet/in_progress_api/query/api.py @@ -0,0 +1,114 @@ +from fastapi import FastAPI, HTTPException +from pydantic import BaseModel +from typing import Optional, List, Dict, Any +import pandas as pd +from graphrag.config.models.graph_rag_config import GraphRagConfig + +app = FastAPI() + + +# Pydantic models for request bodies +class SearchRequest(BaseModel): + config: Dict[ + str, Any + ] # Simplified for brevity, ideally you'd define a proper config model + nodes: List[Dict[str, Any]] + entities: List[Dict[str, Any]] + community_reports: List[Dict[str, Any]] + community_level: int + response_type: str + query: str + + +class LocalSearchRequest(SearchRequest): + text_units: List[Dict[str, Any]] + relationships: List[Dict[str, Any]] + covariates: Optional[List[Dict[str, Any]]] = None + + +# Helper function to convert dict to DataFrame +def dict_to_df(data: List[Dict[str, Any]]) -> pd.DataFrame: + return pd.DataFrame(data) + + +@app.post("/global_search") +async def api_global_search(request: SearchRequest): + try: + config = GraphRagConfig( + **request.config + ) # Assuming GraphRagConfig can be instantiated this way + result = await global_search( + config=config, + nodes=dict_to_df(request.nodes), + entities=dict_to_df(request.entities), + community_reports=dict_to_df(request.community_reports), + community_level=request.community_level, + response_type=request.response_type, + query=request.query, + ) + return {"result": result} + except Exception as e: + raise HTTPException(status_code=500, detail=str(e)) + + +@app.post("/global_search_streaming") +async def api_global_search_streaming(request: SearchRequest): + try: + config = GraphRagConfig(**request.config) + result_generator = global_search_streaming( + config=config, + nodes=dict_to_df(request.nodes), + entities=dict_to_df(request.entities), + community_reports=dict_to_df(request.community_reports), + community_level=request.community_level, + response_type=request.response_type, + query=request.query, + ) + return result_generator # FastAPI will handle the streaming + except Exception as e: + raise HTTPException(status_code=500, detail=str(e)) + + +@app.post("/local_search") +async def api_local_search(request: LocalSearchRequest): + try: + config = GraphRagConfig(**request.config) + result = await local_search( + config=config, + nodes=dict_to_df(request.nodes), + entities=dict_to_df(request.entities), + community_reports=dict_to_df(request.community_reports), + text_units=dict_to_df(request.text_units), + relationships=dict_to_df(request.relationships), + covariates=dict_to_df(request.covariates) if request.covariates else None, + community_level=request.community_level, + response_type=request.response_type, + query=request.query, + ) + return {"result": result} + except Exception as e: + raise HTTPException(status_code=500, detail=str(e)) + + +@app.post("/local_search_streaming") +async def api_local_search_streaming(request: LocalSearchRequest): + try: + config = GraphRagConfig(**request.config) + result_generator = local_search_streaming( + config=config, + nodes=dict_to_df(request.nodes), + entities=dict_to_df(request.entities), + community_reports=dict_to_df(request.community_reports), + text_units=dict_to_df(request.text_units), + relationships=dict_to_df(request.relationships), + covariates=dict_to_df(request.covariates) if request.covariates else None, + community_level=request.community_level, + response_type=request.response_type, + query=request.query, + ) + return result_generator # FastAPI will handle the streaming + except Exception as e: + raise HTTPException(status_code=500, detail=str(e)) + + +# Keep the original functions (global_search, global_search_streaming, local_search, local_search_streaming) as they are diff --git a/graphfleet/in_progress_api/query/apy.py b/graphfleet/in_progress_api/query/apy.py new file mode 100644 index 000000000..5e01dde4f --- /dev/null +++ b/graphfleet/in_progress_api/query/apy.py @@ -0,0 +1,133 @@ +from fastapi import FastAPI, HTTPException +from pydantic import BaseModel +import pandas as pd +from graphrag.config.models.graph_rag_config import GraphRagConfig +from graphrag.query.api import ( + global_search, + global_search_streaming, + local_search, + local_search_streaming, +) + +app = FastAPI() + +# Assume these are loaded or configured server-side +CONFIG = GraphRagConfig() # Load your config here +NODES_DF = pd.DataFrame() # Load your nodes data here +ENTITIES_DF = pd.DataFrame() # Load your entities data here +COMMUNITY_REPORTS_DF = pd.DataFrame() # Load your community reports data here +TEXT_UNITS_DF = pd.DataFrame() # Load your text units data here +RELATIONSHIPS_DF = pd.DataFrame() # Load your relationships data here +COVARIATES_DF = pd.DataFrame() # Load your covariates data here +COMMUNITY_LEVEL = 1 # Set your community level here +RESPONSE_TYPE = "text" # Set your default response type here + +class QueryRequest(BaseModel): + query: str + +@app.post("/global_search") +async def api_global_search(request: QueryRequest): + try: + # Debug: Print DataFrame info + print("NODES_DF:", NODES_DF.info()) + print("ENTITIES_DF:", ENTITIES_DF.info()) + print("COMMUNITY_REPORTS_DF:", COMMUNITY_REPORTS_DF.info()) + + result = await global_search( + config=CONFIG, + nodes=NODES_DF, + entities=ENTITIES_DF, + community_reports=COMMUNITY_REPORTS_DF, + community_level=COMMUNITY_LEVEL, + response_type=RESPONSE_TYPE, + query=request.query, + ) + return {"result": result} + except Exception as e: + raise HTTPException(status_code=500, detail=str(e)) + +@app.post("/global_search_streaming") +async def api_global_search_streaming(request: QueryRequest): + try: + # Debug: Print DataFrame info + print("NODES_DF:", NODES_DF.info()) + print("ENTITIES_DF:", ENTITIES_DF.info()) + print("COMMUNITY_REPORTS_DF:", COMMUNITY_REPORTS_DF.info()) + + async def stream_generator(): + async for chunk in global_search_streaming( + config=CONFIG, + nodes=NODES_DF, + entities=ENTITIES_DF, + community_reports=COMMUNITY_REPORTS_DF, + community_level=COMMUNITY_LEVEL, + response_type=RESPONSE_TYPE, + query=request.query, + ): + yield chunk + + return stream_generator() + except Exception as e: + raise HTTPException(status_code=500, detail=str(e)) + +@app.post("/local_search") +async def api_local_search(request: QueryRequest): + try: + # Debug: Print DataFrame info + print("NODES_DF:", NODES_DF.info()) + print("ENTITIES_DF:", ENTITIES_DF.info()) + print("COMMUNITY_REPORTS_DF:", COMMUNITY_REPORTS_DF.info()) + print("TEXT_UNITS_DF:", TEXT_UNITS_DF.info()) + print("RELATIONSHIPS_DF:", RELATIONSHIPS_DF.info()) + print("COVARIATES_DF:", COVARIATES_DF.info()) + + result = await local_search( + config=CONFIG, + nodes=NODES_DF, + entities=ENTITIES_DF, + community_reports=COMMUNITY_REPORTS_DF, + text_units=TEXT_UNITS_DF, + relationships=RELATIONSHIPS_DF, + covariates=COVARIATES_DF, + community_level=COMMUNITY_LEVEL, + response_type=RESPONSE_TYPE, + query=request.query, + ) + return {"result": result} + except Exception as e: + raise HTTPException(status_code=500, detail=str(e)) + +@app.post("/local_search_streaming") +async def api_local_search_streaming(request: QueryRequest): + try: + # Debug: Print DataFrame info + print("NODES_DF:", NODES_DF.info()) + print("ENTITIES_DF:", ENTITIES_DF.info()) + print("COMMUNITY_REPORTS_DF:", COMMUNITY_REPORTS_DF.info()) + print("TEXT_UNITS_DF:", TEXT_UNITS_DF.info()) + print("RELATIONSHIPS_DF:", RELATIONSHIPS_DF.info()) + print("COVARIATES_DF:", COVARIATES_DF.info()) + + async def stream_generator(): + async for chunk in local_search_streaming( + config=CONFIG, + nodes=NODES_DF, + entities=ENTITIES_DF, + community_reports=COMMUNITY_REPORTS_DF, + text_units=TEXT_UNITS_DF, + relationships=RELATIONSHIPS_DF, + covariates=COVARIATES_DF, + community_level=COMMUNITY_LEVEL, + response_type=RESPONSE_TYPE, + query=request.query, + ): + yield chunk + + return stream_generator() + except Exception as e: + raise HTTPException(status_code=500, detail=str(e)) + +if __name__ == "__main__": + import uvicorn + + uvicorn.run(app, host="0.0.0.0", port=8000) \ No newline at end of file diff --git a/graphfleet/input/FromLouvaintoLeiden.txt b/graphfleet/input/FromLouvaintoLeiden.txt deleted file mode 100644 index 00859a26e..000000000 --- a/graphfleet/input/FromLouvaintoLeiden.txt +++ /dev/null @@ -1,1687 +0,0 @@ -From Louvain to Leiden: guaranteeing well-connected communities -V.A. Traag,āˆ—L. Waltman, and N.J. van Eck -Centre for Science and Technology Studies, Leiden University, the Netherlands -(Dated: October 31, 2019) -Community detection is often used to understand the structure of large and complex networks. -One of the most popular algorithms for uncovering community structure is the so-called Louvain -algorithm. We show that this algorithm has a major defect that largely went unnoticed until -now: the Louvain algorithm may yield arbitrarily badly connected communities. In the worst case, -communities may even be disconnected, especially when running the algorithm iteratively. In our -experimental analysis, we observe that up to 25%of the communities are badly connected and up -to16%are disconnected. To address this problem, we introduce the Leiden algorithm. We prove -that the Leiden algorithm yields communities that are guaranteed to be connected. In addition, we -prove that, when the Leiden algorithm is applied iteratively, it converges to a partition in which all -subsets of all communities are locally optimally assigned. Furthermore, by relying on a fast local -move approach, the Leiden algorithm runs faster than the Louvain algorithm. We demonstrate -the performance of the Leiden algorithm for several benchmark and real-world networks. We ļ¬nd -that the Leiden algorithm is faster than the Louvain algorithm and uncovers better partitions, in -addition to providing explicit guarantees. -I. INTRODUCTION -In many complex networks, nodes cluster and form rel- -atively dense groupsā€”often called communities [1, 2]. -Such a modular structure is usually not known before- -hand. Detecting communities in a network is therefore -an important problem. One of the best-known methods -for community detection is called modularity [3]. This -method tries to maximise the diļ¬€erence between the ac- -tual number of edges in a community and the expected -number of such edges. We denote by ecthe actual num- -ber of edges in community c. The expected number of -edges can be expressed asK2 -c -2m, whereKcis the sum of the -degrees of the nodes in community candmis the total -number of edges in the network. This way of deļ¬ning -the expected number of edges is based on the so-called -conļ¬guration model. Modularity is given by -H=1 -2m/summationdisplay -c/parenleftbigg -ecāˆ’Ī³K2 -c -2m/parenrightbigg -, (1) -whereĪ³ >0is a resolution parameter [4]. Higher resolu- -tions lead to more communities, while lower resolutions -lead to fewer communities. -Optimising modularity is NP-hard [5], and consequen- -tially many heuristic algorithms have been proposed, -such as hierarchical agglomeration [6], extremal optimi- -sation [7], simulated annealing [4, 8] and spectral [9] al- -gorithms. One of the most popular algorithms to opti- -mise modularity is the so-called Louvain algorithm [10], -named after the location of its authors. It was found to -be one of the fastest and best performing algorithms in -comparative analyses [11, 12], and it is one of the most- -cited works in the community detection literature. -āˆ—v.a.traag@cwts.leidenuniv.nlAlthough originally deļ¬ned for modularity, the Lou- -vain algorithm can also be used to optimise other quality -functions. An alternative quality function is the Con- -stant Potts Model (CPM) [13], which overcomes some -limitations of modularity. CPM is deļ¬ned as -H=/summationdisplay -c/bracketleftbigg -ecāˆ’Ī³/parenleftbiggnc -2/parenrightbigg/bracketrightbigg -, (2) -wherencis the number of nodes in community c. The -interpretation of the resolution parameter Ī³is quite -straightforward. The parameter functions as a sort of -threshold: communities should have a density of at least -Ī³,whilethedensitybetweencommunitiesshouldbelower -thanĪ³. Higher resolutions lead to more communities and -lower resolutions lead to fewer communities, similarly to -the resolution parameter for modularity. -In this paper, we show that the Louvain algorithm has -a major problem, for both modularity and CPM. The -algorithm may yield arbitrarily badly connected commu- -nities, over and above the well-known issue of the reso- -lution limit [14] (Section IIA). Communities may even -be internally disconnected. To address this important -shortcoming, we introduce a new algorithm that is faster, -ļ¬nds better partitions and provides explicit guarantees -and bounds (Section III). The new algorithm integrates -several earlier improvements, incorporating a combina- -tion of smart local move [15], fast local move [16, 17] and -random neighbour move [18]. We prove that the new -algorithm is guaranteed to produce partitions in which -all communities are internally connected. In addition, -we prove that the algorithm converges to an asymptot- -ically stable partition in which all subsets of all com- -munities are locally optimally assigned. The quality of -such an asymptotically stable partition provides an up- -per bound on the quality of an optimal partition. Fi- -nally, we demonstrate the excellent performance of the -algorithm for several benchmark and real-world networks -(Section IV). To ensure readability of the paper to thearXiv:1810.08473v3 [cs.SI] 30 Oct 20192 -a) b) -c) d)Move nodes -Aggregate -Move nodesLevel 1 -Level 2 -FIG. 1.Louvain algorithm . The Louvain algorithm starts -from a singleton partition in which each node is in its own -community (a). The algorithm moves individual nodes from -one community to another to ļ¬nd a partition (b). Based on -this partition, an aggregate network is created (c). The algo- -rithm then moves individual nodes in the aggregate network -(d). These steps are repeated until the quality cannot be -increased further. -broadest possible audience, we have chosen to relegate -all technical details to appendices. The main ideas of our -algorithm are explained in an intuitive way in the main -text of the paper. We name our algorithm the Leiden -algorithm , after the location of its authors. -II. LOUVAIN ALGORITHM -The Louvain algorithm [10] is very simple and elegant. -Thealgorithmoptimisesaqualityfunctionsuchasmodu- -larityorCPMintwoelementaryphases: (1)localmoving -of nodes; and (2) aggregation of the network. In the local -movingphase, individualnodesaremovedtothecommu- -nity that yields the largest increase in the quality func- -tion. In the aggregation phase, an aggregate network is -created based on the partition obtained in the local mov- -ing phase. Each community in this partition becomes -a node in the aggregate network. The two phases are -repeated until the quality function cannot be increased -further. The Louvain algorithm is illustrated in Fig. 1 -and summarised in pseudo-code in Algorithm A.1 in Ap- -pendix A. -Usually, the Louvain algorithm starts from a singleton -partition, in which each node is in its own community. -a) b) -012 -3 -45 -6 -Rest of network012 -3 -45 -6 -Rest of networkFIG. 2.Disconnected community. Consider the partition -shownin(a). Whennode0ismovedtoadiļ¬€erentcommunity, -theredcommunitybecomesinternallydisconnected, asshown -in(b). However, nodes1ā€“6arestilllocallyoptimallyassigned, -and therefore these nodes will stay in the red community. -However, it is also possible to start the algorithm from -a diļ¬€erent partition [15]. In particular, in an attempt -to ļ¬nd better partitions, multiple consecutive iterations -of the algorithm can be performed, using the partition -identiļ¬ed in one iteration as starting point for the next -iteration. -A. Badly connected communities -We now show that the Louvain algorithm may ļ¬nd -arbitrarily badly connected communities. In particular, -we show that Louvain may identify communities that -are internally disconnected. That is, one part of such -an internally disconnected community can reach another -part only through a path going outside the community. -Importantly, the problem of disconnected communities -is not just a theoretical curiosity. As we will demon- -strate in Section IV, the problem occurs frequently in -practice when using the Louvain algorithm. Perhaps sur- -prisingly, iterating the algorithm aggravates the problem, -even though it does increase the quality function. -In the Louvain algorithm, a node may be moved to a -diļ¬€erent community while it may have acted as a bridge -between diļ¬€erent components of its old community. Re- -moving such a node from its old community disconnects -the old community. One may expect that other nodes in -the old community will then also be moved to other com- -munities. However, this is not necessarily the case, as the -other nodes may still be suļ¬ƒciently strongly connected -to their community, despite the fact that the community -has become disconnected. -To elucidate the problem, we consider the example il- -lustrated in Fig. 2. The numerical details of the example -can be found in Appendix B. The thick edges in Fig. 2 -representstrongerconnections, whiletheotheredgesrep- -resent weaker connections. At some point, the Louvain -algorithm may end up in the community structure shown3 -in Fig. 2(a). Nodes 0ā€“6 are in the same community. -Nodes 1ā€“6 have connections only within this community, -whereas node 0 also has many external connections. The -algorithm continues to move nodes in the rest of the net- -work. At some point, node 0 is considered for moving. -When a suļ¬ƒcient number of neighbours of node 0 have -formed a community in the rest of the network, it may -be optimal to move node 0 to this community, thus cre- -ating the situation depicted in Fig. 2(b). In this new -situation, nodes 2, 3, 5 and 6 have only internal connec- -tions. These nodes are therefore optimally assigned to -their current community. On the other hand, after node -0 has been moved to a diļ¬€erent community, nodes 1 and -4 have not only internal but also external connections. -Nevertheless, depending on the relative strengths of the -diļ¬€erent connections, these nodes may still be optimally -assigned to their current community. In that case, nodes -1ā€“6 are all locally optimally assigned, despite the fact -that their community has become disconnected. Clearly, -it would be better to split up the community. Nodes 1ā€“3 -should form a community and nodes 4ā€“6 should form an- -other community. However, the Louvain algorithm does -not consider this possibility, since it considers only indi- -vidual node movements. Moreover, when no more nodes -can be moved, the algorithm will aggregate the network. -When a disconnected community has become a node in -an aggregate network, there are no more possibilities to -split up the community. Hence, the community remains -disconnected, unless it is merged with another commu- -nity that happens to act as a bridge. -Obviously, this is a worst case example, showing that -disconnected communities may be identiļ¬ed by the Lou- -vain algorithm. More subtle problems may occur as well, -causing Louvain to ļ¬nd communities that are connected, -but only in a very weak sense. Hence, in general, Louvain -may ļ¬nd arbitrarily badly connected communities. -This problem is diļ¬€erent from the well-known issue -of the resolution limit of modularity [14]. Due to the -resolution limit, modularity may cause smaller commu- -nities to be clustered into larger communities. In other -words, modularity may ā€œhideā€ smaller communities and -may yield communities containing signiļ¬cant substruc- -ture. CPM does not suļ¬€er from this issue [13]. Never- -theless, when CPM is used as the quality function, the -Louvain algorithm may still ļ¬nd arbitrarily badly con- -nected communities. Hence, the problem of Louvain out- -linedaboveisindependentfromtheissueoftheresolution -limit. In the case of modularity, communities may have -signiļ¬cant substructure both because of the resolution -limit and because of the shortcomings of Louvain. -In fact, although it may seem that the Louvain al- -gorithm does a good job at ļ¬nding high quality parti- -tions, in its standard form the algorithm provides only -one guarantee: the algorithm yields partitions for which -it is guaranteed that no communities can be merged. In -other words, communities are guaranteed to be well sep- -arated. Somewhat stronger guarantees can be obtained -by iterating the algorithm, using the partition obtainedin one iteration of the algorithm as starting point for the -next iteration. When iterating Louvain, the quality of -the partitions will keep increasing until the algorithm is -unable to make any further improvements. At this point, -it is guaranteed that each individual node is optimally -assigned. In this iterative scheme, Louvain provides two -guarantees: (1) no communities can be merged and (2) -no nodes can be moved. -Contrary to what might be expected, iterating the -Louvain algorithm aggravates the problem of badly con- -nected communities, as we will also see in Section IV. -This is not too diļ¬ƒcult to explain. After the ļ¬rst itera- -tion of the Louvain algorithm, some partition has been -obtained. In the ļ¬rst step of the next iteration, Louvain -will again move individual nodes in the network. Some -of these nodes may very well act as bridges, similarly to -node 0in the above example. By moving these nodes, -Louvain creates badly connected communities. More- -over, Louvain has no mechanism for ļ¬xing these commu- -nities. Iterating the Louvain algorithm can therefore be -seen as a double-edged sword: it improves the partition -in some way, but degrades it in another way. -The problem of disconnected communities has been -observed before in the context of the label propagation -algorithm [19]. However, so far this problem has never -been studied for the Louvain algorithm. Moreover, the -deeper signiļ¬cance of the problem was not recognised: -disconnected communities are merely the most extreme -manifestation of the problem of arbitrarily badly con- -nected communities. Trying to ļ¬x the problem by sim- -ply considering the connected components of communi- -ties[19ā€“21]isunsatisfactorybecauseitaddressesonlythe -most extreme case and does not resolve the more funda- -mental problem. We therefore require a more principled -solution, which we will introduce in the next section. -III. LEIDEN ALGORITHM -We here introduce the Leiden algorithm, which guaran- -tees that communities are well connected. The Leiden -algorithm is partly based on the previously introduced -smart local move algorithm [15], which itself can be seen -as an improvement of the Louvain algorithm. The Lei- -den algorithm also takes advantage of the idea of speed- -ing up the local moving of nodes [16, 17] and the idea of -moving nodes to random neighbours [18]. We consider -these ideas to represent the most promising directions -in which the Louvain algorithm can be improved, even -though we recognise that other improvements have been -suggested as well [22]. The Leiden algorithm consists of -three phases: (1) local moving of nodes, (2) reļ¬nement of -the partition and (3) aggregation of the network based on -the reļ¬ned partition, using the non-reļ¬ned partition to -create an initial partition for the aggregate network. The -Leiden algorithm is considerably more complex than the -Louvain algorithm. Fig. 3 provides an illustration of the -algorithm. The algorithm is described in pseudo-code in4 -a) b) c) -d) e) f)Move nodes Reļ¬ne -Aggregate -Move nodes Reļ¬neLevel 1 -Level 2 -FIG. 3.Leiden algorithm . The Leiden algorithm starts from a singleton partition (a). The algorithm moves individual nodes -from one community to another to ļ¬nd a partition (b), which is then reļ¬ned (c). An aggregate network (d) is created based -on the reļ¬ned partition, using the non-reļ¬ned partition to create an initial partition for the aggregate network. For example, -the red community in (b) is reļ¬ned into two subcommunities in (c), which after aggregation become two separate nodes in (d), -both belonging to the same community. The algorithm then moves individual nodes in the aggregate network (e). In this case, -reļ¬nement does not change the partition (f). These steps are repeated until no further improvements can be made. -Algorithm A.2 in Appendix A. -In the Louvain algorithm, an aggregate network is cre- -ated based on the partition Presulting from the local -moving phase. The idea of the reļ¬nement phase in the -Leiden algorithm is to identify a partition Preļ¬nedthat -is a reļ¬nement of P. Communities in Pmay be split -into multiple subcommunities in Preļ¬ned. The aggregate -network is created based on the partition Preļ¬ned. How- -ever, the initial partition for the aggregate network is -based on P, just like in the Louvain algorithm. By cre- -ating the aggregate network based on Preļ¬nedrather than -P, the Leiden algorithm has more room for identifying -high-quality partitions. In fact, by implementing the re- -ļ¬nement phase in the right way, several attractive guar- -antees can be given for partitions produced by the Leidenalgorithm. -The reļ¬ned partition Preļ¬nedis obtained as follows. -Initially, Preļ¬nedis set to a singleton partition, in which -each node is in its own community. The algorithm then -locally merges nodes in Preļ¬ned: nodes that are on their -own in a community in Preļ¬nedcan be merged with a dif- -ferent community. Importantly, mergers are performed -only within each community of the partition P. In ad- -dition, a node is merged with a community in Preļ¬ned -only if both are suļ¬ƒciently well connected to their com- -munity in P. After the reļ¬nement phase is concluded, -communities in Poften will have been split into multiple -communities in Preļ¬ned, but not always. -In the reļ¬nement phase, nodes are not necessarily -greedily merged with the community that yields the5 -largest increase in the quality function. Instead, a node -maybemergedwithanycommunityforwhichthequality -function increases. The community with which a node is -merged is selected randomly (similar to [18]). The larger -the increase in the quality function, the more likely a -community is to be selected. The degree of randomness -intheselectionofacommunityisdeterminedbyaparam- -eterĪø>0. Randomness in the selection of a community -allows the partition space to be explored more broadly. -Node mergers that cause the quality function to decrease -are not considered. This contrasts with optimisation al- -gorithms such as simulated annealing, which do allow the -quality function to decrease [4, 8]. Such algorithms are -rather slow, making them ineļ¬€ective for large networks. -Excluding node mergers that decrease the quality func- -tion makes the reļ¬nement phase more eļ¬ƒcient. As we -prove in Appendix C1, even when node mergers that -decrease the quality function are excluded, the optimal -partition of a set of nodes can still be uncovered. This -is not the case when nodes are greedily merged with the -community that yields the largest increase in the quality -function. In that case, some optimal partitions cannot -be found, as we show in Appendix C2. -Another important diļ¬€erence between the Leiden al- -gorithm and the Louvain algorithm is the implementa- -tion of the local moving phase. Unlike the Louvain algo- -rithm, the Leiden algorithm uses a fast local move pro- -cedure in this phase. Louvain keeps visiting all nodes -in a network until there are no more node movements -that increase the quality function. In doing so, Louvain -keeps visiting nodes that cannot be moved to a diļ¬€er- -ent community. In the fast local move procedure in the -Leiden algorithm, only nodes whose neighbourhood has -changed are visited. This is similar to ideas proposed re- -cently as ā€œpruningā€ [16] and in a slightly diļ¬€erent form -as ā€œprioritisationā€ [17]. The fast local move procedure -can be summarised as follows. We start by initialising -a queue with all nodes in the network. The nodes are -added to the queue in a random order. We then remove -the ļ¬rst node from the front of the queue and we deter- -mine whether the quality function can be increased by -moving this node from its current community to a diļ¬€er- -ent one. If we move the node to a diļ¬€erent community, -we add to the rear of the queue all neighbours of the -node that do not belong to the nodeā€™s new community -and that are not yet in the queue. We keep removing -nodes from the front of the queue, possibly moving these -nodes to a diļ¬€erent community. This continues until the -queue is empty. For a full speciļ¬cation of the fast local -move procedure, we refer to the pseudo-code of the Lei- -den algorithm in Algorithm A.2 in Appendix A. Using -the fast local move procedure, the ļ¬rst visit to all nodes -in a network in the Leiden algorithm is the same as in -the Louvain algorithm. However, after all nodes have -been visited once, Leiden visits only nodes whose neigh- -bourhood has changed, whereas Louvain keeps visiting -all nodes in the network. In this way, Leiden implements -the local moving phase more eļ¬ƒciently than Louvain.TABLE I. Overview of the guarantees provided by the Lou- -vain algorithm and the Leiden algorithm. -Louvain Leiden -Each -iterationĪ³-separation 3 3 -Ī³-connectivity 3 -Stable -iterationNode optimality 3 3 -Subpartition Ī³-density 3 -AsymptoticUniformĪ³-density 3 -Subset optimality 3 -A. Guarantees -We now consider the guarantees provided by the Lei- -den algorithm. The algorithm is run iteratively, using -the partition identiļ¬ed in one iteration as starting point -for the next iteration. We can guarantee a number of -properties of the partitions found by the Leiden algo- -rithm at various stages of the iterative process. Below -we oļ¬€er an intuitive explanation of these properties. We -provide the full deļ¬nitions of the properties as well as the -mathematical proofs in Appendix D. -After each iteration of the Leiden algorithm, it is guar- -anteed that: -1. All communities are Ī³-separated. -2. All communities are Ī³-connected. -In these properties, Ī³refers to the resolution parameter -in the quality function that is optimised, which can be -either modularity or CPM. The property of Ī³-separation -is also guaranteed by the Louvain algorithm. It states -that there are no communities that can be merged. The -property of Ī³-connectivity is a slightly stronger variant -of ordinary connectivity. As discussed in Section IIA, -the Louvain algorithm does not guarantee connectivity. -It therefore does not guarantee Ī³-connectivity either. -An iteration of the Leiden algorithm in which the par- -tition does not change is called a stable iteration. After a -stable iteration of the Leiden algorithm, it is guaranteed -that: -3. All nodes are locally optimally assigned. -4. All communities are subpartition Ī³-dense. -Node optimality is also guaranteed after a stable itera- -tion of the Louvain algorithm. It means that there are no -individual nodes that can be moved to a diļ¬€erent com- -munity. Subpartition Ī³-density is not guaranteed by the -Louvain algorithm. A community is subpartition Ī³-dense -if it can be partitioned into two parts such that: (1) the -two parts are well connected to each other; (2) neither -part can be separated from its community; and (3) each -part is also subpartition Ī³-dense itself. Subpartition Ī³- -density does not imply that individual nodes are locally -optimally assigned. It only implies that individual nodes -are well connected to their community.6 -TABLE II. Overview of the empirical networks and of the -maximalmodularityafter 10replicationsof 10iterationseach, -both for the Louvain and for the Leiden algorithm. -Max. modularity -Nodes Degree Louvain Leiden -DBLPa317 080 6 .6 0.8262 0.8387 -Amazona334 863 5 .6 0.9301 0.9341 -IMDBb374 511 80 .2 0.7062 0.7069 -Live Journala3 997 962 17 .4 0.7653 0.7739 -Web of Sciencec9 811 130 21 .2 0.7911 0.7951 -Web UKd39 252 879 39 .8 0.9796 0.9801 -ahttps://snap.stanford.edu/data/ -bhttps://sparse.tamu.edu/Barabasi/NotreDame_actors -cData cannot be shared due to license restrictions. -dhttp://law.di.unimi.it/webdata/uk-2005/ -In the case of the Louvain algorithm, after a stable it- -eration, all subsequent iterations will be stable as well. -Hence, no further improvements can be made after a sta- -ble iteration of the Louvain algorithm. This contrasts -with the Leiden algorithm. After a stable iteration of -the Leiden algorithm, the algorithm may still be able to -make further improvements in later iterations. In fact, -when we keep iterating the Leiden algorithm, it will con- -verge to a partition for which it is guaranteed that: -5. All communities are uniformly Ī³-dense. -6. All communities are subset optimal. -A community is uniformly Ī³-dense if there are no subsets -of the community that can be separated from the com- -munity. Uniform Ī³-density means that no matter how a -community is partitioned into two parts, the two parts -willalwaysbewellconnectedtoeachother. Furthermore, -if all communities in a partition are uniformly Ī³-dense, -the quality of the partition is not too far from optimal, -as shown in Appendix E. A community is subset optimal -if all subsets of the community are locally optimally as- -signed. That is, no subset can be moved to a diļ¬€erent -community. Subset optimality is the strongest guarantee -that is provided by the Leiden algorithm. It implies uni- -formĪ³-density and all the other above-mentioned prop- -erties. -An overview of the various guarantees is presented in -Table I. -IV. EXPERIMENTAL ANALYSIS -In the previous section, we showed that the Leiden -algorithm guarantees a number of properties of the par- -titions uncovered at diļ¬€erent stages of the algorithm. We -also suggested that the Leiden algorithm is faster than -the Louvain algorithm, because of the fast local move -approach. In this section, we analyse and compare theperformance of the two algorithms in practice1. All ex- -periments were run on a computer with 64 Intel Xeon -E5-4667v3 2GHz CPUs and 1TB internal memory. In all -experiments reported here, we used a value of 0.01for the -parameterĪøthat determines the degree of randomness in -the reļ¬nement phase of the Leiden algorithm. However, -values ofĪøwithin a range of roughly [0.0005,0.1]all pro- -vide reasonable results, thus allowing for some, but not -too much randomness. We use six empirical networks -in our analysis. These are the same networks that were -also studied in an earlier paper introducing the smart lo- -cal move algorithm [15]. Table II provides an overview of -the six networks. First, we show that the Louvain algo- -rithm ļ¬nds disconnected communities, and more gener- -ally, badly connected communities in the empirical net- -works. Second, to study the scaling of the Louvain and -the Leiden algorithm, we use benchmark networks, al- -lowing us to compare the algorithms in terms of both -computational time and quality of the partitions. Fi- -nally, we compare the performance of the algorithms on -the empirical networks. We ļ¬nd that the Leiden algo- -rithm commonly ļ¬nds partitions of higher quality in less -time. The diļ¬€erence in computational time is especially -pronounced for larger networks, with Leiden being up to -20times faster than Louvain in empirical networks. -A. Badly connected communities -We study the problem of badly connected communi- -ties when using the Louvain algorithm for several empir- -ical networks. For each community in a partition that -was uncovered by the Louvain algorithm, we determined -whether it is internally connected or not. In addition, -to analyse whether a community is badly connected, we -ran the Leiden algorithm on the subnetwork consisting of -all nodes belonging to the community.2The Leiden al- -gorithm was run until a stable iteration was obtained. -When the Leiden algorithm found that a community -could be split into multiple subcommunities, we counted -the community as badly connected. Note that if Lei- -den ļ¬nds subcommunities, splitting up the community is -guaranteed to increase modularity. Conversely, if Leiden -does not ļ¬nd subcommunities, there is no guarantee that -modularity cannot be increased by splitting up the com- -munity. Hence, by counting the number of communities -thathavebeensplitup, weobtainedalowerboundonthe -number of communities that are badly connected. The -1We implemented both algorithms in Java, available from -github.com/CWTSLeiden/networkanalysis and deposited at -Zenodo [23]. Additionally, we implemented a Python pack- -age, available from github.com/vtraag/leidenalg and deposited -at Zenodo [24]. -2We ensured that modularity optimisation for the subnetwork -was fully consistent with modularity optimisation for the whole -network [13].7 -051015202530DBLP Amazon -05IMDB Live Journal -1 2 3 40510152025Web of Science -1 2 3 4Web UK (2005)Disconnected (Louvain) Badly connected (Louvain) -Badly connected (Leiden)% communities -Iterations -FIG. 4. Badly connected communities . Percentage of -communities found by the Louvain algorithm that are either -disconnected or badly connected compared to percentage of -badly connected communities found by the Leiden algorithm. -Note that communities found by the Leiden algorithm are -guaranteed to be connected. -count of badly connected communities also included dis- -connected communities. For each network, we repeated -the experiment 10times. We used modularity with a -resolution parameter of Ī³= 1for the experiments. -As can be seen in Fig. 4, in the ļ¬rst iteration of the -Louvain algorithm, the percentage of badly connected -communities can be quite high. For the Amazon, DBLP -and Web UK networks, Louvain yields on average respec- -tively 23%,16%and14%badly connected communities. -The percentage of disconnected communities is more lim- -ited, usually around 1%. However, in the case of the Web -of Science network, more than 5%of the communities are -disconnected in the ļ¬rst iteration. -Later iterations of the Louvain algorithm only ag- -gravate the problem of disconnected communities, even -though the quality function (i.e. modularity) increases. -The second iteration of Louvain shows a large increase -in the percentage of disconnected communities. In sub- -sequent iterations, the percentage of disconnected com- -munities remains fairly stable. The increase in the per- -centage of disconnected communities is relatively limited -for the Live Journal and Web of Science networks. Other -networks show an almost tenfold increase in the percent- -age of disconnected communities. The percentage of dis- -connected communities even jumps to 16%for the DBLP -network. The percentage of badly connected communi- -ties is less aļ¬€ected by the number of iterations of the -Louvain algorithm. Presumably, many of the badly con- -nected communities in the ļ¬rst iteration of Louvain be-come disconnected in the second iteration. Indeed, the -percentage of disconnected communities becomes more -comparable to the percentage of badly connected com- -munities in later iterations. Nonetheless, some networks -still show large diļ¬€erences. For example, after four itera- -tions, the Web UK network has 8%disconnected commu- -nities, but twice as many badly connected communities. -Even worse, the Amazon network has 5%disconnected -communities, but 25%badly connected communities. -The above results shows that the problem of discon- -nected and badly connected communities is quite per- -vasive in practice. Because the percentage of discon- -nected communities in the ļ¬rst iteration of the Louvain -algorithm usually seems to be relatively low, the prob- -lem may have escaped attention from users of the algo- -rithm. However, focussing only on disconnected commu- -nities masks the more fundamental issue: Louvain ļ¬nds -arbitrarily badly connected communities. The high per- -centage of badly connected communities attests to this. -Besides being pervasive, the problem is also sizeable. In -the worst case, almost a quarter of the communities are -badly connected. This may have serious consequences -for analyses based on the resulting partitions. For exam- -ple, nodes in a community in biological or neurological -networks are often assumed to share similar functions or -behaviour [25]. However, if communities are badly con- -nected, this may lead to incorrect attributions of shared -functionality. Similarly, in citation networks, such as the -Web of Science network, nodes in a community are usu- -ally considered to share a common topic [26, 27]. Again, -if communities are badly connected, this may lead to in- -correct inferences of topics, which will aļ¬€ect bibliomet- -ric analyses relying on the inferred topics. In short, the -problem of badly connected communities has important -practical consequences. -The Leiden algorithm has been speciļ¬cally designed -to address the problem of badly connected communities. -Fig. 4 shows how well it does compared to the Louvain -algorithm. The Leiden algorithm guarantees all commu- -nities to be connected, but it may yield badly connected -communities. In terms of the percentage of badly con- -nectedcommunitiesintheļ¬rstiteration,Leidenperforms -even worse than Louvain, as can be seen in Fig. 4. Cru- -cially, however, the percentage of badly connected com- -munities decreases with each iteration of the Leiden al- -gorithm. Starting from the second iteration, Leiden out- -performed Louvain in terms of the percentage of badly -connected communities. In fact, if we keep iterating the -Leiden algorithm, it will converge to a partition with- -out any badly connected communities, as discussed in -Section III. Hence, the Leiden algorithm eļ¬€ectively ad- -dresses the problem of badly connected communities. -B. Benchmark networks -To study the scaling of the Louvain and the Leiden al- -gorithm, we rely on a variant of a well-known approach8 -0.740.760.780.8Āµ= 0.2 -0.540.560.580.6Āµ= 0.4 -0.340.360.380.4Āµ= 0.6 -0.220.240.26Āµ= 0.8 -10310510710āˆ’1101103 -10310510710āˆ’1101103 -10310510710āˆ’1101103 -10310510710āˆ’1102105Louvain Leiden -NodesQuality Time (s) -FIG. 5. Scaling of benchmark results for network size . Speed and quality of the Louvain and the Leiden algorithm -for benchmark networks of increasing size (two iterations). For larger networks and higher values of Āµ, Louvain is much slower -than Leiden. For higher values of Āµ, Leiden ļ¬nds better partitions than Louvain. -204060801000.797250.79730.797350.7974Āµ= 0.2 -4060801001200.5950.5960.597Āµ= 0.4 -100 200 3000.320.340.360.380.4Āµ= 0.6 -1021030.2020.2040.2060.2080.21Āµ= 0.8 -500 1,0000.800260.800280.80030.800320.80034 -500 1,000 1,5000.5990.59950.60.6005 -1,000 2,000 3,000 4,0000.320.340.360.380.4 -1031040.2040.2060.208Louvain Leiden -Time (s)Qualityn= 106n= 107 -FIG. 6.Runtime versus quality for benchmark networks . Speed and quality for the ļ¬rst 10iterations of the Louvain -and the Leiden algorithm for benchmark networks ( n= 106andn= 107). The horizontal axis indicates the cumulative time -taken to obtain the quality indicated on the vertical axis. Each point corresponds to a certain iteration of an algorithm, with -results averaged over 10experiments. In general, Leiden is both faster than Louvain and ļ¬nds better partitions. -for constructing benchmark networks [28]. We generated -benchmark networks in the following way. First, we cre- -ated a speciļ¬ed number of nodes and we assigned each -node to a community. Communities were all of equal -size. A community size of 50nodes was used for the re- -sultspresentedbelow, butlargercommunitysizesyielded -qualitatively similar results. We then created a certain -number of edges such that a speciļ¬ed average degree /angbracketleftk/angbracketright -wasobtained. Fortheresultsreportedbelow, theaverage -degree was set to /angbracketleftk/angbracketright= 10. Edges were created in such -a way that an edge fell between two communities with aprobability Āµand within a community with a probability -1āˆ’Āµ. We applied the Louvain and the Leiden algorithm -to exactly the same networks, using the same seed for -the random number generator. For both algorithms, 10 -iterations were performed. We used the CPM quality -function. The value of the resolution parameter was de- -termined based on the so-called mixing parameter Āµ[13]. -We generated networks with n= 103ton= 107nodes. -For each set of parameters, we repeated the experiment -10times. Below, the quality of a partition is reported as -H -2m, where His deļ¬ned in Eq. (2) and mis the number9 -0.2 0.4 0.6 0.8102103104105Louvain -Leiden -ĀµTime (s) -FIG. 7.Scaling of benchmark results for diļ¬ƒculty of -the partition . Speed of the ļ¬rst iteration of the Louvain -and the Leiden algorithm for benchmark networks with in- -creasingly diļ¬ƒcult partitions ( n= 107). In the most diļ¬ƒcult -case (Āµ= 0.9), Louvain requires almost 2.5days, while Leiden -needs fewer than 10minutes. -of edges. -As shown in Fig. 5, for lower values of Āµthe partition -is well deļ¬ned, and neither the Louvain nor the Leiden -algorithm has a problem in determining the correct par- -tition in only two iterations. Hence, for lower values of -Āµ, the diļ¬€erence in quality is negligible. However, as Āµ -increases, the Leiden algorithm starts to outperform the -Louvain algorithm. The diļ¬€erences are not very large, -which is probably because both algorithms ļ¬nd parti- -tions for which the quality is close to optimal, related to -the issue of the degeneracy of quality functions [29]. -TheLeidenalgorithmisclearlyfasterthantheLouvain -algorithm. For lower values of Āµ, the correct partition -is easy to ļ¬nd and Leiden is only about twice as fast as -Louvain. However, forhighervaluesof Āµ, Leidenbecomes -ordersofmagnitudefasterthanLouvain, reaching 10ā€“100 -times faster runtimes for the largest networks. As can be -seen in Fig. 7, whereas Louvain becomes much slower for -more diļ¬ƒcult partitions, Leiden is much less aļ¬€ected by -the diļ¬ƒculty of the partition. -Fig. 6 presents total runtime versus quality for all iter- -ations of the Louvain and the Leiden algorithm. As can -be seen in the ļ¬gure, Louvain quickly reaches a state in -which it is unable to ļ¬nd better partitions. On the other -hand, Leiden keeps ļ¬nding better partitions, especially -for higher values of Āµ, for which it is more diļ¬ƒcult to -identify good partitions. A number of iterations of the -Leiden algorithm can be performed before the Louvain -algorithm has ļ¬nished its ļ¬rst iteration. Later iterations -of the Louvain algorithm are very fast, but this is only -because the partition remains the same. With one ex- -ception (Āµ= 0.2andn= 107), all results in Fig. 6 show -that Leiden outperforms Louvain in terms of both com- -putational time and quality of the partitions. -DBLPAmazonIMDB -Live Journal -Web of ScienceWeb UK (2005)1101001,00010,000 Time (s)Louvain -LeidenFIG. 8. First iteration runtime for empirical net- -works. Speed of the ļ¬rst iteration of the Louvain and the -Leiden algorithm for six empirical networks. Leiden is faster -than Louvain especially for larger networks. -C. Empirical networks -Analyses based on benchmark networks have only a -limited value because these networks are not represen- -tative of empirical real-world networks. In particular, -benchmarknetworkshavea rather simplestructure. Em- -pirical networks show a much richer and more complex -structure. We now compare how the Leiden and the Lou- -vain algorithm perform for the six empirical networks -listed in Table II. Our analysis is based on modularity -with resolution parameter Ī³= 1. For each network, Ta- -ble II reports the maximal modularity obtained using the -Louvain and the Leiden algorithm. -AscanbeseeninFig.8, theLeidenalgorithmissigniļ¬- -cantlyfasterthantheLouvainalgorithmalsoinempirical -networks. In the ļ¬rst iteration, Leiden is roughly 2ā€“20 -times faster than Louvain. The speed diļ¬€erence is espe- -cially large for larger networks. This is similar to what -we have seen for benchmark networks. For the Ama- -zon and IMDB networks, the ļ¬rst iteration of the Leiden -algorithm is only about 1.6times faster than the ļ¬rst -iteration of the Louvain algorithm. However, Leiden is -more than 7times faster for the Live Journal network, -more than 11times faster for the Web of Science network -and more than 20times faster for the Web UK network. -In fact, for the Web of Science and Web UK networks, -Fig. 9 shows that more than 10iterations of the Leiden -algorithmcanbeperformedbeforetheLouvainalgorithm -has ļ¬nished its ļ¬rst iteration. -As shown in Fig. 9, the Leiden algorithm also performs -better than the Louvain algorithm in terms of the qual-10 -ity of the partitions that are obtained. For all networks, -Leidenidentiļ¬essubstantiallybetterpartitionsthanLou- -vain. Louvainquicklyconvergestoapartitionandisthen -unable to make further improvements. In contrast, Lei- -den keeps ļ¬nding better partitions in each iteration. -The quality improvement realised by the Leiden algo- -rithm relative to the Louvain algorithm is larger for em- -pirical networks than for benchmark networks. Hence, -the complex structure of empirical networks creates an -even stronger need for the use of the Leiden algorithm. -Leiden keeps ļ¬nding better partitions for empirical net- -works also after the ļ¬rst 10iterations of the algorithm. -This contrasts to benchmark networks, for which Leiden -often converges after a few iterations. For empirical net- -works, it may take quite some time before the Leiden -algorithm reaches its ļ¬rst stable iteration. As can be -seen in Fig. 10, for the IMDB and Amazon networks, -Leiden reaches a stable iteration relatively quickly, pre- -sumablybecausethesenetworkshaveafairlysimplecom- -munity structure. The DBLP network is somewhat more -challenging, requiring almost 80iterations on average to -reach a stable iteration. The Web of Science network is -the most diļ¬ƒcult one. For this network, Leiden requires -over 750iterations on average to reach a stable iteration. -Importantly, the ļ¬rst iteration of the Leiden algorithm is -the most computationally intensive one, and subsequent -iterations are faster. For example, for the Web of Science -network, the ļ¬rst iteration takes about 110ā€“120seconds,while subsequent iterations require about 40seconds. -V. DISCUSSION -Community detection is an important task in the anal- -ysis of complex networks. Finding communities in large -networks is far from trivial: algorithms need to be fast, -but they also need to provide high-quality results. One -of the most widely used algorithms is the Louvain algo- -rithm [10], which is reported to be among the fastest and -bestperformingcommunitydetectionalgorithms[11,12]. -However, as shown in this paper, the Louvain algorithm -has a major shortcoming: the algorithm yields commu- -nities that may be arbitrarily badly connected. Commu- -nities may even be disconnected. -To overcome the problem of arbitrarily badly con- -nected communities, we introduced a new algorithm, -which we refer to as the Leiden algorithm. This algo- -rithm provides a number of explicit guarantees. In par- -ticular, it yields communities that are guaranteed to be -connected. Moreover, when the algorithm is applied it- -eratively, it converges to a partition in which all subsets -of all communities are guaranteed to be locally optimally -assigned. In practical applications, the Leiden algorithm -convincingly outperforms the Louvain algorithm, both in -terms of speed and in terms of quality of the results, as -shown by the experimental analysis presented in this pa- -per. We conclude that the Leiden algorithm is strongly -preferable to the Louvain algorithm. -[1] S. Fortunato, Phys. Rep. 486, 75 (2010). -[2] M. A. Porter, J.-P. Onnela, and P. J. Mucha, Not. AMS -56, 1082 (2009). -[3] M. E. J. Newman and M. Girvan, Phys. Rev. E 69, -026113 (2004). -[4] J. Reichardt and S. Bornholdt, Phys. Rev. E 74, 016110 -(2006). -[5] U. Brandes, D. Delling, M. Gaertler, R. Gorke, M. Hoe- -fer, Z. Nikoloski, D. Wagner, R. G, M. Hoefer, -Z. Nikoloski, and D. Wagner, IEEE Trans. Knowl. Data -Eng.20, 172 (2008). -[6] A. Clauset, M. E. J. Newman, and C. Moore, Phys. Rev. -E70, 066111 (2004). -[7] J. Duch and A. Arenas, Phys. Rev. E 72, 027104 (2005). -[8] R. GuimerĆ  and L. A. Nunes Amaral, Nature 433, 895 -(2005). -[9] M. E. J. Newman, Phys. Rev. E 74, 036104 (2006). -[10] V. D. Blondel, J.-L. Guillaume, R. Lambiotte, and -E. Lefebvre, J. Stat. Mech.Theory Exp. 10008, 6 (2008). -[11] A. Lancichinetti and S. Fortunato, Phys. Rev. E 80, -056117 (2009). -[12] Z. Yang, R. Algesheimer, and C. J. Tessone, Sci. Rep. -6, 30750 (2016). -[13] V. A. Traag, P. Van Dooren, and Y. Nesterov, Phys. -Rev. E84, 016114 (2011). -[14] S. Fortunato and M. BarthĆ©lemy, Proc. Natl. Acad. Sci. -U. S. A.104, 36 (2007).[15] L. Waltman and N. J. van Eck, Eur. Phys. J. B 86, 471 -(2013). -[16] N. Ozaki, H. Tezuka, and M. Inaba, Int. J. Comput. -Electr. Eng. 8, 207 (2016). -[17] S. Bae, D. Halperin, J. D. West, M. Rosvall, and -B. Howe, ACM Trans. Knowl. Discov. Data 11, 1 (2017). -[18] V. A. Traag, Phys. Rev. E 92, 032801 (2015). -[19] U. Raghavan, R. Albert, and S. Kumara, Phys. Rev. E -76, 036106 (2007). -[20] M. D. Luecken, Application of multi-resolution partition- -ing of interaction networks to the study of complex dis- -ease, Ph.D. thesis, University of Oxford (2016). -[21] F. A. Wolf, F. Hamey, M. Plass, J. Solana, J. S. Dahlin, -B. Gottgens, N. Rajewsky, L. Simon, and F. J. Theis, -bioRxiv (2018), 10.1101/208819. -[22] R. Rotta and A. Noack, J. Exp. Algorithmics 16, 2.1 -(2011). -[23] V. A. Traag, L. Waltman, and N. J. van Eck, ā€œnet- -workanalysis,ā€ Zenodo, 10.5281/zenodo.1466831 (2018), -Source Code. -[24] V. A. Traag, ā€œleidenalg 0.7.0,ā€ Zenodo, 10.5281/zen- -odo.1469357 (2018), Source Code. -[25] E. Bullmore and O. Sporns, Nat. Rev. Neurosci. 10, 186 -(2009). -[26] L. Waltman and N. J. van Eck, J. Am. Soc. Inf. Sci. -Technol.63, 2378 (2012). -[27] R.KlavansandK.W.Boyack,J.Assoc.Inf.Sci.Technol.11 -1 2 3 40.8250.830.835QualityDBLP -1 2 3 40.9260.9280.930.9320.934Amazon -10 200.6980.70.7020.704QualityIMDB -100 2000.750.760.77Live Journal -1,000 2,0000.7750.780.7850.79 -Time (s)QualityWeb of Science -2,000 4,0000.97940.97960.97980.98 -Time (s)Web UKLouvain Leiden -FIG. 9. Runtime versus quality for empirical net- -works. Speed and quality for the ļ¬rst 10iterations of the -Louvain and the Leiden algorithm for six empirical networks. -The horizontal axis indicates the cumulative time taken to -obtain the quality indicated on the vertical axis. Each point -corresponds to a certain iteration of an algorithm, with re- -sults averaged over 10experiments. Leiden is both faster -than Louvain and ļ¬nds better partitions. -68, 984 (2017). -[28] A. Lancichinetti, S. Fortunato, and F. Radicchi, Phys. -Rev. E78, 046110 (2008). -[29] B. H. Good, Y. A. De Montjoye, and A. Clauset, Phys. -Rev. E81, 046106 (2010). -[30] V. A. Traag and J. Bruggeman, Phys. Rev. E 80, 036115 -(2009). -[31] T. N. Dinh, X. Li, and M. T. Thai, in 2015 IEEE Int. -Conf. Data Min. (IEEE, 2015) pp. 101ā€“110. -ACKNOWLEDGMENTS -We gratefully acknowledge computational facilities pro- -vided by the LIACS Data Science Lab Computing Facili- -ties through Frank Takes. We thank Lovro Ė˜Subelj for his -comments on an earlier version of this paper. -DBLPAmazonIMDB -Live Journal -Web of ScienceWeb UK (2005)301003001,000No. iterations until stabilityFIG. 10. Number of iterations until stability . Number -of iterations before the Leiden algorithm has reached a stable -iteration for six empirical networks. In a stable iteration, the -partition is guaranteed to be node optimal and subpartition -Ī³-dense. -AUTHOR CONTRIBUTIONS STATEMENT -All authors conceived the algorithm and contributed to -the source code. VAT performed the experimental analy- -sis. VAT and LW wrote the manuscript. NJvE reviewed -the manuscript. -ADDITIONAL INFORMATION -Competing interests -The authors act as bibliometric consultants to CWTS -B.V., which makes use of community detection algo- -rithms in commercial products and services.12 -Appendix A: Pseudo-code and mathematical notation -Pseudo-codefortheLouvainalgorithmandtheLeidenalgorithmisprovidedinAlgorithmsA.1andA.2, respectively. -Below we discuss the mathematical notation that is used in the pseudo-code and also in the mathematical results -presented in Appendices C, D, and E. There are some uncommon elements in the notation. In particular, the idea of -sets of sets plays an important role, and some concepts related to this idea need to be introduced. -LetG= (V,E)be a graph with n=|V|nodes andm=|E|edges. Graphs are assumed to be undirected. With the -exception of Theorem 14 in Appendix E, the mathematical results presented in this paper apply to both unweighted -and weighted graphs. For simplicity, our mathematical notation assumes graphs to be unweighted, although the -notation does allow for multigraphs. A partition P={C1,...,Cr}consists of r=|P|communities, where each -community CiāŠ†Vconsists of a set of nodes such that V=/uniontext -iCiandCiāˆ©Cj=āˆ…for alli/negationslash=j. For two sets Rand -S, we sometimes use R+Sto denote the union RāˆŖSandRāˆ’Sto denote the diļ¬€erence R\S. -A quality function H(G,P)assigns a ā€œqualityā€ to a partition Pof a graphG. We aim to ļ¬nd a partition with the -highest possible quality. The graph Gis often clear from the context, and we therefore usually write H(P)instead -ofH(G,P). Based on partition P, graphGcan beaggregated into a new graph G/prime. GraphGis then called the base -graph, while graph G/primeis called the aggregate graph . The nodes of the aggregate graph G/primeare the communities in the -partition Pof the base graph G, i.e.V(G/prime) =P. The edges of the aggregate graph G/primeare multi-edges. The number of -edgesbetweentwonodesintheaggregategraph G/primeequalsthenumberofedgesbetweennodesinthetwocorresponding -communities in the base graph G. Hence,E(G/prime) ={(C,D)|(u,v)āˆˆE(G),uāˆˆCāˆˆP,vāˆˆDāˆˆP}, whereE(G/prime) -is a multiset. A quality function must have the property that H(G,P) =H(G/prime,P/prime), where P/prime={{v}|vāˆˆV(G/prime)} -denotes the singleton partition of the aggregate graph G/prime. This ensures that a quality function gives consistent results -for base graphs and aggregate graphs. -We denote by P(v/mapstoā†’C)the partition that is obtained when we start from partition Pand we then move node -vto community C. We write āˆ†HP(v/mapstoā†’C)for the change in the quality function by moving node vto community -Cfor some partition P. In other words, āˆ†HP(v/mapstoā†’C) =H(P(v/mapstoā†’C))āˆ’H(P). We usually leave the partition P -implicit and simply write āˆ†H(v/mapstoā†’C). Similarly, we denote by āˆ†HP(S/mapstoā†’C)the change in the quality function by -moving a set of nodes Sto community C. An empty community is denoted by āˆ…. Hence, āˆ†HP(S/mapstoā†’āˆ…)is the change -in the quality function by moving a set of nodes Sto an empty (i.e. new) community. -Now consider a community Cthat consists of two parts S1andS2such thatC=S1āˆŖS2andS1āˆ©S2=āˆ…. Suppose -thatS1andS2are disconnected. In other words, there are no edges between nodes in S1andS2. We then require a -quality function to have the property that āˆ†H(S1/mapstoā†’āˆ…)>0andāˆ†H(S2/mapstoā†’āˆ…)>0. This guarantees that a partition -can always be improved by splitting a community into its connected components. This comes naturally for most -deļ¬nitions of a community, but this is not the case when considering for example negative links [30]. -Because nodes in an aggregate graph are sets themselves, it is convenient to deļ¬ne some recursive properties. -Deļ¬nition 1. Therecursive size of a setSis deļ¬ned as -/bardblS/bardbl=/summationdisplay -sāˆˆS/bardbls/bardbl, (A1) -where/bardbls/bardbl= 1ifsis not a set itself. The ļ¬‚attening operation for a set Sis deļ¬ned as -ļ¬‚at(S) =/uniondisplay -sāˆˆSļ¬‚at(s), (A2) -where ļ¬‚at(s) =sifsis not a set itself. A set that has been ļ¬‚attened is called a ļ¬‚at set. -The recursive size of a set corresponds to the usual deļ¬nition of set size in case the elements of a set are not -sets themselves, but it generalizes this deļ¬nition whenever the elements are sets themselves. For example, if S= -{{a,b},{c},{d,e,f}}, then -/bardblS/bardbl=/bardbl{a,b}/bardbl+/bardbl{c}/bardbl+/bardbl{d,e,f}/bardbl -= (/bardbla/bardbl+/bardblb/bardbl) +/bardblc/bardbl+ (/bardbld/bardbl+/bardble/bardbl+/bardblf/bardbl) -= 2 + 1 + 3 = 6 . -This contrasts with the traditional size of a set, which is |S|= 3, becauseScontains 3elements. The fact that the -elements are sets themselves plays no role in the traditional size of a set. The ļ¬‚attening of Sis -ļ¬‚at(S) = ļ¬‚at({a,b})āˆŖļ¬‚at({c})āˆŖļ¬‚at({d,e,f}) -=aāˆŖbāˆŖcāˆŖdāˆŖeāˆŖf -={a,b,c,d,e,f}.13 -Note that/bardblS/bardbl=|ļ¬‚at(S)|. -Deļ¬nition 2. Theļ¬‚attening operation for a partition Pis deļ¬ned as -ļ¬‚atāˆ—(P) ={ļ¬‚at(C)|CāˆˆP}. (A3) -Hence, ļ¬‚atāˆ—(P)denotes the operation in which each community CāˆˆPis ļ¬‚attened. A partition that has been -ļ¬‚attened is called a ļ¬‚at partition . -For any partition of an aggregate graph, the equivalent partition of the base graph can be obtained by applying -the ļ¬‚attening operation. -Additionally, we need some terminology to describe the connectivity of communities. -Deļ¬nition 3. LetG= (V,E)be a graph, and let Pbe a partition of G. Furthermore, let H(C)be the subgraph -induced by a community CāˆˆP, i.e.V(H) =CandE(H) ={(u,v)|(u,v)āˆˆE(G),u,vāˆˆC}. A community CāˆˆP -is calledconnected ifH(C)is a connected graph. Conversely, a community CāˆˆPis calleddisconnected ifH(C)is a -disconnected graph. -The mathematical proofs presented in this paper rely on the Constant Potts Model (CPM) [13]. This quality -function has important advantages over modularity. In particular, unlike modularity, CPM does not suļ¬€er from the -problem of the resolution limit [13, 14]. Moreover, our mathematical deļ¬nitions and proofs are quite elegant when -expressed in terms of CPM. The CPM quality function is deļ¬ned as -H(G,P) =/summationdisplay -CāˆˆP/bracketleftbigg -E(C,C)āˆ’Ī³/parenleftbigg/bardblC/bardbl -2/parenrightbigg/bracketrightbigg -, (A4) -whereE(C,D) =|{(u,v)āˆˆE(G)|uāˆˆC,vāˆˆD}|denotes the number of edges between nodes in communities Cand -D. Note that this deļ¬nition can also be used for aggregate graphs because E(G)is a multiset. -The mathematical results presented in this paper also extend to modularity, although the formulations are less -elegant. Results for modularity are straightforward to prove by redeļ¬ning the recursive size /bardblS/bardblof a setS. We need -to deļ¬ne the size of a node vin the base graph as /bardblv/bardbl=kvinstead of/bardblv/bardbl= 1, wherekvis the degree of node v. -Furthermore, we need to rescale the resolution parameter Ī³by2m. Modularity can then be written as -H(G,P) =/summationdisplay -CāˆˆP/bracketleftbigg -E(C,C)āˆ’Ī³ -2m/parenleftbigg/bardblC/bardbl -2/parenrightbigg/bracketrightbigg -. (A5) -Note that, in addition to the overall multiplicative factor of1 -2m, this adds a constantĪ³ -2m/summationtext -C/bardblC/bardbl -2=Ī³ -2to the ordinary -deļ¬nition of modularity [3]. However, this does not matter for optimisation or for the proofs. -As discussed in the main text, the Louvain and the Leiden algorithm can be iterated by performing multiple -consecutive iterations of the algorithm, using the partition identiļ¬ed in one iteration as starting point for the next -iteration. In this way, a sequence of partitions P0,P1,...is obtained such that Pt+1=Louvain (G,Pt)orPt+1= -Leiden (G,Pt). The initial partition P0usually is the singleton partition of the graph G, i.e.P0={{v}|vāˆˆV}. -Appendix B: Disconnected communities in the Louvain algorithm -In this appendix, we analyse the problem that communities obtained using the Louvain algorithm may be discon- -nected. This problem is also discussed in the main text (Section IIA), using the example presented in Fig. 2. However, -the main text oļ¬€ers no numerical details. These details are provided below. -We consider the CPM quality function with a resolution of Ī³=1 -7. In the example presented in Fig. 2, the edges -between nodes 0 and 1 and between nodes 0 and 4 have a weight of 2, as indicated by the thick lines in the ļ¬gure. -All other edges have a weight of 1. The Louvain algorithm starts from a singleton partition, with each node being -assigned to its own community. The algorithm then keeps iterating over all nodes, moving each node to its optimal -community. Depending on the order in which the nodes are visited, the following could happen. Node 1 is visited -ļ¬rst, followed by node 4. Nodes 1 and 4 join the community of node 0, because the weight of the edges between nodes -0 and 1 and between nodes 0 and 4 is suļ¬ƒciently high. For node 1, the best move clearly is to join the community of -node 0. For node 4, the beneļ¬t of joining the community of nodes 0 and 1 then is 2āˆ’Ī³Ā·2 =12 -7. This is larger than the -beneļ¬t of joining the community of node 5 or 6, which is 1āˆ’Ī³Ā·1 =6 -7. Next, nodes 2, 3, 5 and 6 are visited. For these -nodes, it is beneļ¬cial to join the community of nodes 0, 1 and 4, because joining this community has a beneļ¬t of at14 -1:function Louvain (GraphG, Partition P) -2:do -3: Pā†MoveNodes (G,P) āŠæMove nodes between communities -4: doneā†|P|=|V(G)| āŠæTerminate when each community consists of only one node -5:ifnot donethen -6: Gā†AggregateGraph (G,P) āŠæCreate aggregate graph based on partition P -7: Pā†SingletonPartition (G) āŠæAssign each node in aggregate graph to its own community -8:end if -9:whilenot done -10:return ļ¬‚atāˆ—(P) -11:end function -12:function MoveNodes (GraphG, Partition P) -13:do -14: Hold=H(P) -15: forvāˆˆV(G)do āŠæVisit nodes (in random order) -16: C/primeā†arg maxCāˆˆPāˆŖāˆ…āˆ†HP(v/mapstoā†’C) āŠæDetermine best community for node v -17: ifāˆ†HP(v/mapstoā†’C/prime)>0then āŠæPerform only strictly positive node movements -18: v/mapstoā†’C/primeāŠæMove node vto community C/prime -19: end if -20: end for -21:while H(P)>Hold āŠæContinue until no more nodes can be moved -22:return P -23:end function -24:function AggregateGraph (GraphG, Partition P) -25:Vā†P āŠæCommunities become nodes in aggregate graph -26:Eā†{(C,D)|(u,v)āˆˆE(G),uāˆˆCāˆˆP,vāˆˆDāˆˆP} āŠæ Eis a multiset -27:return Graph (V,E) -28:end function -29:function SingletonPartition (GraphG) -30:return{{v}|vāˆˆV(G)} āŠæAssign each node to its own community -31:end function -ALGORITHM A.1. Louvain algorithm. -least 1āˆ’Ī³Ā·6 =1 -7>0. This then yields the situation portrayed in Fig. 2(a). After some node movements in the rest -of the graph, some neighbours of node 0 in the rest of the graph end up together in a new community. Consequently, -when node 0 is visited, it can best be moved to this new community, which gives the situation depicted in Fig. 2(b). -In particular, suppose there are 5nodes in the new community, all of which are connected to node 0. In that case, the -beneļ¬t for node 0 of moving to this community is 5āˆ’Ī³Ā·5 =30 -7, while the beneļ¬t of staying in the current community -is only 2Ā·2āˆ’Ī³Ā·6 =22 -7. After node 0 has moved, nodes 1 and 4 are still locally optimally assigned. For these nodes, -the beneļ¬t of moving to the new community of node 0 is 2āˆ’Ī³Ā·6 =8 -7. This is smaller than the beneļ¬t of staying -in the current community, which is 2āˆ’Ī³Ā·5 =9 -7. Finally, nodes 2, 3, 5 and 6 are all locally optimally assigned, as -1āˆ’Ī³Ā·5 =2 -7>0. Hence, we end up with a community that is disconnected. In later stages of the Louvain algorithm, -there will be no possibility to repair this. -The example presented above considers a weighted graph, but this graph can be assumed to be an aggregate graph -of an unweighted base graph, thus extending the example also to unweighted graphs. Although the example uses -the CPM quality function, similar examples can be given for modularity. However, because of the dependency of -modularity on the number of edges m, the calculations for modularity are a bit more complex. Importantly, both for -CPM and for modularity, the Louvain algorithm suļ¬€ers from the problem of disconnected communities. -Appendix C: Reachability of optimal partitions -In this appendix, we consider two types of move sequences: non-decreasing move sequences and greedy move -sequences. For each type of move sequence, we study whether all optimal partitions are reachable. We ļ¬rst show that -this is not the case for greedy move sequences. In particular, we show that for some optimal partitions there does not -exist a greedy move sequence that is able to reach the partition. We then show that optimal partitions can always15 -1:function Leiden(GraphG, Partition P) -2:do -3: Pā†MoveNodesFast (G,P) āŠæMove nodes between communities -4: doneā†|P|=|V(G)| āŠæTerminate when each community consists of only one node -5:ifnot donethen -6: Preļ¬nedā†RefinePartition (G,P) āŠæReļ¬ne partition P -7: Gā†AggregateGraph (G,Preļ¬ned ) āŠæCreate aggregate graph based on reļ¬ned partition Preļ¬ned -8: Pā†{{v|vāŠ†C,vāˆˆV(G)}|CāˆˆP} āŠæBut maintain partition P -9:end if -10:whilenot done -11:return ļ¬‚atāˆ—(P) -12:end function -13:function MoveNodesFast (GraphG, Partition P) -14:Qā†Queue (V(G)) āŠæMake sure that all nodes will be visited (in random order) -15:do -16:vā†Q.remove() āŠæDetermine next node to visit -17:C/primeā†arg maxCāˆˆPāˆŖāˆ…āˆ†HP(v/mapstoā†’C) āŠæDetermine best community for node v -18: ifāˆ†HP(v/mapstoā†’C/prime)>0then āŠæPerform only strictly positive node movements -19: v/mapstoā†’C/primeāŠæMove node vto community C/prime -20: Nā†{u|(u,v)āˆˆE(G),u /āˆˆC/prime} āŠæIdentify neighbours of node vthat are not in community C/prime -21: Q.add(Nāˆ’Q) āŠæMake sure that these neighbours will be visited -22: end if -23:whileQ/negationslash=āˆ… āŠæContinue until there are no more nodes to visit -24:return P -25:end function -26:function RefinePartition (GraphG, Partition P) -27: Preļ¬nedā†SingletonPartition (G) āŠæAssign each node to its own community -28:forCāˆˆPdo āŠæVisit communities -29: Preļ¬nedā†MergeNodesSubset (G,Preļ¬ned,C) āŠæReļ¬ne community C -30:end for -31:return Preļ¬ned -32:end function -33:function MergeNodesSubset (GraphG, Partition P, SubsetS) -34:R={v|vāˆˆS,E(v,Sāˆ’v)ā‰„Ī³/bardblv/bardblĀ·(/bardblS/bardblāˆ’/bardblv/bardbl)}āŠæConsider only nodes that are well connected within subset S -35:forvāˆˆRdo āŠæVisit nodes (in random order) -36: ifvin singleton community then āŠæConsider only nodes that have not yet been merged -37: Tā†{C|CāˆˆP,CāŠ†S,E(C,Sāˆ’C)ā‰„Ī³/bardblC/bardblĀ·(/bardblS/bardblāˆ’/bardblC/bardbl)}āŠæConsider only well-connected communities -38: Pr(C/prime=C)āˆ¼/braceleftbigg -exp/parenleftbig1 -Īøāˆ†HP(v/mapstoā†’C)/parenrightbig -ifāˆ†HP(v/mapstoā†’C)ā‰„0 -0 otherwiseforCāˆˆTāŠæChoose random community C/prime -39: v/mapstoā†’C/primeāŠæMove node vto community C/prime -40: end if -41:end for -42:return P -43:end function -44:function AggregateGraph (GraphG, Partition P) -45:Vā†P āŠæCommunities become nodes in aggregate graph -46:Eā†{(C,D)|(u,v)āˆˆE(G),uāˆˆCāˆˆP,vāˆˆDāˆˆP} āŠæ Eis a multiset -47:return Graph (V,E) -48:end function -49:function SingletonPartition (GraphG) -50:return{{v}|vāˆˆV(G)} āŠæAssign each node to its own community -51:end function -ALGORITHM A.2. Leiden algorithm.16 -be reached using a non-decreasing move sequence. This result forms the basis for the asymptotic guarantees of the -Leiden algorithm, which are discussed in Appendix D3. -We ļ¬rst deļ¬ne the diļ¬€erent types of move sequences. -Deļ¬nition 4. LetG= (V,E)be a graph, and let P0,...,PĻ„be partitions of G. A sequence of partitions P0,...,PĻ„ -is called a move sequence if for eacht= 0,...,Ļ„āˆ’1there exists a node vtāˆˆVand a community CtāˆˆPtāˆŖāˆ…such -thatPt+1=Pt(vt/mapstoā†’Ct). A move sequence is called non-decreasing ifH(Pt+1)ā‰„H(Pt)for allt= 0,...,Ļ„āˆ’1. A -move sequence is called greedyifH(Pt+1) = maxCH(Pt(vt/mapstoā†’C))for allt= 0,...,Ļ„āˆ’1. -In other words, the next partition in a move sequence is obtained by moving a single node to a diļ¬€erent community. -Clearly, a greedy move sequence must be non-decreasing, but a non-decreasing move sequence does not need to be -greedy. A natural question is whether for any optimal partition Pāˆ—there exists a move sequence that starts from the -singleton partition and that reaches the optimal partition, i.e., a move sequence P0,...,PĻ„withP0={{v}|vāˆˆV} -andPĻ„=Pāˆ—. Trivially, it is always possible to reach the optimal partition if we allow all movesā€”even moves that -decrease the quality functionā€”as is done for example in simulated annealing [4, 8]. However, it can be shown that -there is no need to consider all moves in order to reach the optimal partition. It is suļ¬ƒcient to consider only non- -decreasing moves. On the other hand, considering only greedy moves turns out to be too restrictive to guarantee that -the optimal partition can be reached. -1. Non-decreasing move sequences -We here prove that for any graph there exists a non-decreasing move sequence that reaches the optimal partition -Pāˆ—. The optimal partition can be reached in nāˆ’|Pāˆ—|steps. -Theorem 1. LetG= (V,E)be a graph, and let Pāˆ—be an optimal partition of G. There then exists a non-decreasing -move sequence P0,...,PĻ„withP0={{v}|vāˆˆV},PĻ„=Pāˆ—, andĻ„=nāˆ’|Pāˆ—|. -Proof.LetCāˆ—āˆˆPāˆ—be a community in the optimal partition Pāˆ—, letv0āˆˆCāˆ—be a node in this community, and -letC0={v0}. LetP0={{v}|vāˆˆV}be the singleton partition. For t= 1,...,|Cāˆ—|āˆ’1, letvtāˆˆCāˆ—āˆ’Ctāˆ’1, -letCt={v0,...,vt}āˆˆPt, and let Pt=Ptāˆ’1(vt/mapstoā†’Ctāˆ’1). We prove by contradiction that there always exists a -non-decreasing move sequence P0,...,P|Cāˆ—|āˆ’1. Assume that for some tthere does not exist a node vtfor which -āˆ†H(vt/mapstoā†’Ctāˆ’1)ā‰„0. LetS=Cāˆ—āˆ’Ctāˆ’1andR=Ctāˆ’1. For allvāˆˆS, -E(v,R)āˆ’Ī³/bardblv/bardblĀ·/bardblR/bardbl<0. -This implies that -E(S,R) =/summationdisplay -vāˆˆSE(v,R)<Ī³/bardblS/bardblĀ·/bardblR/bardbl. -However, by optimality, for all SāŠ†Cāˆ—andR=Cāˆ—āˆ’S, -E(S,R)ā‰„Ī³/bardblS/bardblĀ·/bardblR/bardbl. -We therefore have a contradiction. Hence, there always exists a non-decreasing move sequence P0,...,P|Cāˆ—|āˆ’1. This -move sequence reaches the community Ct=Cāˆ—. The above reasoning can be applied to each community Cāˆ—āˆˆPāˆ—. -Consequently, each of these communities can be reached using a non-decreasing move sequence. In addition, for each -community Cāˆ—āˆˆPāˆ—, this can be done in |Cāˆ—|āˆ’1steps, so that in total Ļ„=/summationtext -Cāˆ—āˆˆPāˆ—(|Cāˆ—|āˆ’1) =nāˆ’|Pāˆ—|steps are -needed.  -2. Greedy move sequences -We here show that there does not always exist a greedy move sequence that reaches the optimal partition of a -graph. To show this, we provide a counterexample in which we have a graph for which there is no greedy move -sequence that reaches the optimal partition. Our counterexample includes two nodes that should be assigned to -diļ¬€erent communities. However, because there is a strong connection between the nodes, in a greedy move sequence -the nodes are always assigned to the same community. We use the CPM quality function in our counterexample, but -a similar counterexample can be given for modularity. The counterexample is illustrated in Fig. C.1. The thick edges17 -a) -b)0 1 23 -45 -67 -0 1 23 -45 -67 -FIG. C.1. Unreachable optimal partition. A greedy move sequence always reaches the partition in (a), whereas the -partition in (b) is optimal. This demonstrates that for some graphs there does not exist a greedy move sequence that reaches -the optimal partition. -have a weight of 3, while the thin ones have a weight of3 -2. The resolution is set to Ī³= 1. In this situation, nodes 0 -and 1 are always joined together in a community. This has a beneļ¬t of 3āˆ’Ī³= 2, which is larger than the beneļ¬t of -3Ā·3 -2āˆ’Ī³Ā·3 =3 -2obtained by node 0 joining the community of nodes 2, 3 and 4 or node 1 joining the community of -nodes 5, 6 and 7. Hence, regardless of the exact node order, the partition reached by a greedy move sequence always -consists of three communities. This gives a total quality of -2Ā·/parenleftbigg -3Ā·3āˆ’Ī³3Ā·2 -2/parenrightbigg -+/parenleftbigg -3āˆ’Ī³2Ā·1 -2/parenrightbigg -= 14, -while the optimal partition has only two communities, consisting of nodes {0,2,3,4}and{1,5,6,7}and resulting in -a total quality of -2Ā·/parenleftbigg -3Ā·3 + 3Ā·3 -2āˆ’Ī³4Ā·3 -2/parenrightbigg -= 15. -Hence, a greedy move sequence always reaches the partition in Fig. C.1(a), whereas the partition in Fig. C.1(b) is -optimal. -Appendix D: Guarantees of the Leiden algorithm -In this appendix, we discuss the guarantees provided by the Leiden algorithm. The guarantees of the Leiden -algorithm partly rely on the randomness in the algorithm. We therefore require that Īø > 0. Before stating the -guarantees of the Leiden algorithm, we ļ¬rst deļ¬ne a number of properties. We start by introducing some relatively -weak properties, and we then move on to stronger properties. In the following deļ¬nitions, Pis a ļ¬‚at partition of a -graphG= (V,E). -Deļ¬nition5 (Ī³-separation) .Wecallapairofcommunities C,DāˆˆPĪ³-separated ifāˆ†H(C/mapstoā†’D) = āˆ†H(D/mapstoā†’C)ā‰¤0. -A community CāˆˆPisĪ³-separated if CisĪ³-separated with respect to all DāˆˆP. A partition PisĪ³-separated if all -CāˆˆPareĪ³-separated. -Deļ¬nition 6 (Ī³-connectivity) .We call a set of nodes SāŠ†CāˆˆPĪ³-connected if|S|= 1or ifScan be partitioned into -two setsRandTsuch thatE(R,T)ā‰„Ī³/bardblR/bardblĀ·/bardblT/bardblandRandTareĪ³-connected. A community CāˆˆPisĪ³-connected -ifS=CisĪ³-connected. A partition PisĪ³-connected if all CāˆˆPareĪ³-connected. -Deļ¬nition 7 (Subpartition Ī³-density).We call a set of nodes SāŠ†CāˆˆPsubpartition Ī³-denseif the following two -conditions are satisļ¬ed: (i) āˆ†H(S/mapstoā†’āˆ…)ā‰¤0and (ii)|S|= 1orScan be partitioned into two sets RandTsuch that18 -E(R,T)ā‰„Ī³/bardblR/bardblĀ·/bardblT/bardblandRandTare subpartition Ī³-dense. A community CāˆˆPis subpartition Ī³-dense ifS=C -is subpartition Ī³-dense. A partition Pis subpartition Ī³-dense if all CāˆˆPare subpartition Ī³-dense. -Deļ¬nition 8 (Node optimality) .We call a community CāˆˆPnode optimal ifāˆ†H(v/mapstoā†’D)ā‰¤0for allvāˆˆCand all -DāˆˆP(orD=āˆ…). A partition Pis node optimal if all CāˆˆPare node optimal. -Deļ¬nition 9 (UniformĪ³-density).We call a community CāˆˆPuniformlyĪ³-denseifāˆ†H(S/mapstoā†’āˆ…)ā‰¤0for allSāŠ†C. -A partition Pis uniformly Ī³-dense if all CāˆˆPare uniformly Ī³-dense. -Deļ¬nition 10 (Subset optimality) .We call a community CāˆˆPsubset optimal ifāˆ†H(S/mapstoā†’D)ā‰¤0for allSāŠ†C -and allDāˆˆP(orD=āˆ…). A partition Pis subset optimal if all CāˆˆPare subset optimal. -Subsetoptimalityclearlyisthestrongestpropertyandsubsumesallotherproperties. Uniform Ī³-densityissubsumed -by subset optimality but may be somewhat more intuitive to grasp. It states that any subset of nodes in a community -is always connected to the rest of the community with a density of at least Ī³. In other words, for all SāŠ†CāˆˆPwe -have -E(S,Cāˆ’S)ā‰„Ī³/bardblS/bardblĀ·/bardblCāˆ’S/bardbl. (D1) -Imposingtherestriction D=āˆ…inthedeļ¬nitionofsubsetoptimalitygivesthepropertyofuniform Ī³-density, restricting -Sto consist of only one node gives the property of node optimality, and imposing the restriction S=Cyields the -property of Ī³-separation. Uniform Ī³-density implies subpartition Ī³-density, which in turn implies Ī³-connectivity. -Subpartition Ī³-density also implies that individual nodes cannot be split from their community (but notice that this -is a weaker property than node optimality). Ordinary connectivity is implied by Ī³-connectivity, but not vice versa. -Obviously, any optimal partition is subset optimal, but not the other way around: a subset optimal partition is not -necessarily an optimal partition (see Fig. C.1(a) for an example). -In the rest of this appendix, we show that the Leiden algorithm guarantees that the above properties hold for -partitions produced by the algorithm. The properties hold either in each iteration, in every stable iteration, or -asymptotically. The ļ¬rst two properties of Ī³-separation and Ī³-connectivity are guaranteed in each iteration of the -Leiden algorithm. We prove this in Appendix D1. The next two properties of subpartition Ī³-density and node -optimality are guaranteed in every stable iteration of the Leiden algorithm, as we prove in Appendix D2. Finally, -in Appendix D3 we prove that asymptotically the Leiden algorithm guarantees the last two properties of uniform -Ī³-density and subset optimality. -1. Guarantees in each iteration -In order to show that the property of Ī³-separation is guaranteed in each iteration of the Leiden algorithm, we ļ¬rst -need to prove some results for the MoveNodesFast function in the Leiden algorithm. -We start by introducing some notation. The MoveNodesFast function iteratively evaluates nodes. When a node -is evaluated, either it is moved to a diļ¬€erent (possibly empty) community or it is kept in its current community, -depending on what is most beneļ¬cial for the quality function. Let G= (V,E)be a graph, let Pbe a partition -ofG, and let P/prime=MoveNodesFast (G,P). We denote by P0,...,Pra sequence of partitions generated by the -MoveNodesFast function, with P0=Pdenoting the initial partition, P1denoting the partition after the ļ¬rst -evaluation of a node has taken place, and so on. Pr=P/primedenotes the partition after the ļ¬nal evaluation of a node -has taken place. The MoveNodesFast function maintains a queue of nodes that still need to be evaluated. Let Qs -be the set of nodes that still need to be evaluated after snode evaluations have taken place, with Q0=V. Also, for -allvāˆˆV, letCv -sāˆˆPsbe the community in which node vļ¬nds itself after snode evaluations have taken place. -The following lemma states that at any point in the MoveNodesFast function, if a node is disconnected from the -rest of its community, the node will ļ¬nd itself in the queue of nodes that still need to be evaluated. -Lemma2. Usingthenotationintroducedabove,forall vāˆˆVandalls,wehavevāˆˆQsor|Cv -s|= 1orE(v,Cv -sāˆ’v)>0. -Proof.We are going to prove the lemma for an arbitrary node vāˆˆV. We provide a proof by induction. We observe -thatvāˆˆQ0, which provides our inductive base. Suppose that vāˆˆQsāˆ’1or|Cv -sāˆ’1|= 1orE(v,Cv -sāˆ’1āˆ’v)>0. This is -our inductive hypothesis. We are going to show that vāˆˆQsor|Cv -s|= 1orE(v,Cv -sāˆ’v)>0. IfvāˆˆQs, this result is -obtained in a trivial way. Suppose therefore that v /āˆˆQs. We then need to show that |Cv -s|= 1orE(v,Cv -sāˆ’v)>0. -To do so, we distinguish between two cases. -We ļ¬rst consider the case in which vāˆˆQsāˆ’1. IfvāˆˆQsāˆ’1andv /āˆˆQs, nodevhas just been evaluated. We then -obviously have|Cv -s|= 1orE(v,Cv -sāˆ’v)>0. Otherwise we would have |Cv -s|>1andE(v,Cv -sāˆ’v) = 0, which would19 -mean that node vis disconnected from the rest of its community. Since node vhas just been evaluated, this is not -possible. -We now consider the case in which v /āˆˆQsāˆ’1. LetuāˆˆVbe the node that has just been evaluated, i.e., uāˆˆQsāˆ’1 -andu /āˆˆQs. If nodeuhas not been moved to a diļ¬€erent community, then Ps=Psāˆ’1. Obviously, if|Cv -sāˆ’1|= 1or -E(v,Cv -sāˆ’1āˆ’v)>0, we then have|Cv -s|= 1orE(v,Cv -sāˆ’v)>0. On the other hand, if node uhas been moved to a -diļ¬€erent community, we have (u,v)/āˆˆE(G)orvāˆˆCu -s. To see this, note that if (u,v)āˆˆE(G)andv /āˆˆCu -s, we would -havevāˆˆQs(following line 21 in Algorithm A.2). This contradicts our assumption that v /āˆˆQs, so that we must have -(u,v)/āˆˆE(G)orvāˆˆCu -s. In other words, either there is no edge between nodes uandvor nodeuhas been moved -to the community of node v. In either case, it is not possible that the movement of node ucauses node vto become -disconnected from the rest of its community. Hence, in either case, if |Cv -sāˆ’1|= 1orE(v,Cv -sāˆ’1āˆ’v)>0, then|Cv -s|= 1 -orE(v,Cv -sāˆ’v)>0.  -Using Lemma 2, we now prove the following lemma, which states that for partitions provided by the MoveNodes- -Fastfunction it is guaranteed that singleton communities cannot be merged with each other. -Lemma 3. LetG= (V,E)be a graph, let Pbe a partition of G, and let P/prime=MoveNodesFast (G,P). Then for -all pairsC,DāˆˆP/primesuch that|C|=|D|= 1, we have āˆ†H(C/mapstoā†’D) = āˆ†H(D/mapstoā†’C)ā‰¤0. -Proof.We are going to prove the lemma for an arbitrary pair of communities C,DāˆˆP/primesuch that|C|=|D|= 1. -We use the notation introduced above. If C,DāˆˆPsfor alls, it is clear that āˆ†H(C/mapstoā†’D) = āˆ†H(D/mapstoā†’C)ā‰¤0. -Otherwise, consider tsuch thatC,DāˆˆPsfor allsā‰„tand eitherC /āˆˆPtāˆ’1orD /āˆˆPtāˆ’1. Without loss of generality, -we assume that C /āˆˆPtāˆ’1andDāˆˆPtāˆ’1. ConsidervāˆˆVsuch thatC={v}. Aftertāˆ’1node evaluations have taken -place, there are two possibilities. -One possibility is that node vis evaluated and is moved to an empty community. This means that moving node v -to an empty community is more beneļ¬cial for the quality function than moving node vto community D. It is then -clear that āˆ†H(C/mapstoā†’D) = āˆ†H(D/mapstoā†’C)ā‰¤0. -The second possibility is that node vis in a community together with one other node uāˆˆV(i.e.{u,v}āˆˆPtāˆ’1) -and that this node uis evaluated and is moved to a diļ¬€erent community. In this case, vāˆˆQt, as we will now show. -If(u,v)āˆˆE(G), this follows from line 21 in Algorithm A.2. If (u,v)/āˆˆE(G), we have|Cv -tāˆ’1|=|{u,v}|= 2and -E(v,Cv -tāˆ’1āˆ’v) = 0. It then follows from Lemma 2 that vāˆˆQtāˆ’1. Since node vis not evaluated in node evaluation t -(nodeuis evaluated in this node evaluation), vāˆˆQtāˆ’1implies that vāˆˆQt. IfvāˆˆQt, at some point sā‰„t, nodevis -evaluated. Since C,DāˆˆPsfor allsā‰„t, keeping node vin its own singleton community Cis more beneļ¬cial for the -quality function than moving node vto community D. This means that āˆ†H(C/mapstoā†’D) = āˆ†H(D/mapstoā†’C)ā‰¤0.  -Lemma 3 enables us to prove that the property of Ī³-separation is guaranteed in each iteration of the Leiden -algorithm, as stated in the following theorem. -Theorem 4. LetG= (V,E)be a graph, let Ptbe a ļ¬‚at partition of G, and let Pt+1=Leiden (G,Pt). Then Pt+1 -isĪ³-separated. -Proof.LetG/lscript= (V/lscript,E/lscript)be the aggregate graph at the highest level in the Leiden algorithm, let P/lscriptbe the initial -partition of G/lscript, and let P/prime -/lscript=MoveNodesFast (G/lscript,P/lscript). Since we are at the highest level of aggregation, it follows -from line 4 in Algorithm A.2 that |P/prime -/lscript|=|V/lscript|, which means that |C|= 1for allCāˆˆP/prime -/lscript. In other words, P/prime -/lscriptis a -singleton partition of G/lscript. Lemma 3 then implies that for all C,DāˆˆP/prime -/lscriptwe have āˆ†H(C/mapstoā†’D) = āˆ†H(D/mapstoā†’C)ā‰¤0. -SincePt+1= ļ¬‚atāˆ—(P/prime -/lscript), it follows that for all C,DāˆˆPt+1we have āˆ†H(C/mapstoā†’D) = āˆ†H(D/mapstoā†’C)ā‰¤0. Hence, Pt+1is -Ī³-separated.  -The property of Ī³-separation also holds after each iteration of the Louvain algorithm. In fact, for the Louvain -algorithm this is much easier to see than for the Leiden algorithm. The Louvain algorithm uses the MoveNodes -functioninsteadofthe MoveNodesFast function. Unlikethe MoveNodesFast function, the MoveNodes function -yieldspartitionsthatareguaranteedtobenodeoptimal. Thisguaranteeleadsinastraightforwardwaytotheproperty -ofĪ³-separation for partitions obtained in each iteration of the Louvain algorithm. -We now consider the property of Ī³-connectivity. By constructing a tree corresponding to the decomposition of -Ī³-connectivity, we are going to prove that this property is guaranteed in each iteration of the Leiden algorithm. -Theorem 5. LetG= (V,E)be a graph, let Ptbe a ļ¬‚at partition of G, and let Pt+1=Leiden (G,Pt). Then Pt+1 -isĪ³-connected. -Proof.LetG/lscript= (V/lscript,E/lscript)be the aggregate graph at level /lscriptin the Leiden algorithm, with G0=Gbeing the base graph. -We say that a node vāˆˆV/lscriptisĪ³-connected if ļ¬‚at(v)isĪ³-connected. We are going to proceed inductively. Each node -in the base graph G0is triviallyĪ³-connected. This provides our inductive base. Suppose that each node vāˆˆV/lscriptāˆ’1is20 -Ī³-connected, which is our inductive hypothesis. Each node vāˆˆV/lscriptis obtained by merging one or more nodes at the -preceding level, i.e. v={u|uāˆˆS}for some set SāŠ†V/lscriptāˆ’1. Ifvconsists of only one node at the preceding level, vis -immediately Ī³-connected by our inductive hypothesis. The set of nodes Sis constructed in the MergeNodesSubset -function. There exists some order u1,...,ukin which nodes are added to S. LetSi={u1,...,ui}be the set obtained -after adding node ui. It follows from line 38 in Algorithm A.2 that E(ui+1,Si)ā‰„Ī³/bardblui+1/bardblĀ·/bardblSi/bardblfori= 1,...,kāˆ’1. -Taking into account that each uiisĪ³-connected by our inductive hypothesis, this implies that each set SiisĪ³- -connected. Since S=SkisĪ³-connected, node visĪ³-connected. Hence, each node vāˆˆV/lscriptisĪ³-connected. This -also holds for the nodes in the aggregate graph at the highest level in the Leiden algorithm, which implies that all -communities in Pt+1areĪ³-connected. In other words, Pt+1isĪ³-connected.  -Note that the theorem does not require Ptto be connected. Even if a disconnected partition is provided as input -to the Leiden algorithm, performing a single iteration of the algorithm will give a partition that is Ī³-connected. -2. Guarantees in stable iterations -As discussed earlier, the Leiden algorithm can be iterated until Pt+1=Leiden (G,Pt). Likewise, the Louvain -algorithm can be iterated until Pt+1=Louvain (G,Pt). We say that an iteration is stableifPt+1=Pt, in which -case we call Pt(orPt+1) astable partition . -There is a subtle point when considering stable iterations. In order for the below guarantees to hold, we need to -ensure that H(Pt+1) =H(Pt)implies Pt+1=Pt. In both the Leiden algorithm and the Louvain algorithm, we -therefore consider only strictly positive improvements (see line 17 in Algorithm A.1 and line 18 in Algorithm A.2). In -other words, if a node movement leads to a partition that has the same quality as the current partition, the current -partition is preferred and the node movement will not take place. This then also implies that H(Pt+1)>H(Pt)if -Pt+1/negationslash=Pt. -The Leiden algorithm guarantees that a stable partition is subpartition Ī³-dense, as stated in the following theorem. -Note that the proof of the theorem has a structure that is similar to the structure of the proof of Theorem 5 presented -above. -Theorem 6. LetG= (V,E)be a graph, let Ptbe a ļ¬‚at partition of G, and let Pt+1=Leiden (G,Pt). IfPt+1=Pt, -thenPt+1=Ptis subpartition Ī³-dense. -Proof.Suppose we have a stable iteration. Hence, Pt+1=Pt. LetG/lscript= (V/lscript,E/lscript)be the aggregate graph at level -/lscriptin the Leiden algorithm, with G0=Gbeing the base graph. We say that a node vāˆˆV/lscriptis subpartition Ī³-dense -if the set of nodes ļ¬‚at(v)is subpartition Ī³-dense. We ļ¬rst observe that for all levels /lscriptand all nodes vāˆˆV/lscriptwe -have āˆ†H(v/mapstoā†’āˆ…)ā‰¤0. To see this, note that if āˆ†H(v/mapstoā†’āˆ…)>0for some level /lscriptand some node vāˆˆV/lscript, the -MoveNodesFast function would have removed node vfrom its community, which means that the iteration would -not have been stable. We are now going to proceed inductively. Since āˆ†H(v/mapstoā†’ āˆ…)ā‰¤0for all nodes vāˆˆV0, -each node in the base graph G0is subpartition Ī³-dense. This provides our inductive base. Suppose that each node -vāˆˆV/lscriptāˆ’1is subpartition Ī³-dense, which is our inductive hypothesis. Each node vāˆˆV/lscriptis obtained by merging one -or more nodes at the preceding level, i.e. v={u|uāˆˆS}for some set SāŠ†V/lscriptāˆ’1. Ifvconsists of only one node -at the preceding level, vis immediately subpartition Ī³-dense by our inductive hypothesis. The set of nodes Sis -constructed in the MergeNodesSubset function. There exists some order u1,...,ukin which nodes are added to -S. LetSi={u1,...,ui}be the set obtained after adding node ui. It follows from line 38 in Algorithm A.2 that -E(ui+1,Si)ā‰„Ī³/bardblui+1/bardblĀ·/bardblSi/bardblfori= 1,...,kāˆ’1. Furthermore, line 37 in Algorithm A.2 ensures that āˆ†H(Si/mapstoā†’āˆ…)ā‰¤0 -fori= 1,...,kāˆ’1. We also have āˆ†H(Sk/mapstoā†’āˆ…)ā‰¤0, sinceSk=S=vand since āˆ†H(v/mapstoā†’āˆ…)ā‰¤0, as observed -above. Taking into account that each uiis subpartition Ī³-dense by our inductive hypothesis, this implies that each -setSiis subpartition Ī³-dense. Since S=Skis subpartition Ī³-dense, node vis subpartition Ī³-dense. Hence, each node -vāˆˆV/lscriptis subpartition Ī³-dense. This also holds for the nodes in the aggregate graph at the highest level in the Leiden -algorithm, which implies that all communities in Pt+1=Ptare subpartition Ī³-dense. In other words, Pt+1=Ptis -subpartition Ī³-dense.  -Subpartition Ī³-density does not imply node optimality. It guarantees only that āˆ†H(v/mapstoā†’āˆ…)ā‰¤0for allvāˆˆV, not -that āˆ†H(v/mapstoā†’D)ā‰¤0for allvāˆˆVand allDāˆˆP. However, it is easy to see that all nodes are locally optimally -assigned in a stable iteration of the Leiden algorithm. This is stated in the following theorem. -Theorem 7. LetG= (V,E)be a graph, let Ptbe a ļ¬‚at partition of G, and let Pt+1=Leiden (G,Pt). IfPt+1=Pt, -thenPt+1=Ptis node optimal.21 -Proof.Suppose we have a stable iteration. Hence, Pt+1=Pt. We are going to give a proof by contradiction. Assume -thatPt+1=Ptis not node optimal. There then exists a node vāˆˆCāˆˆPtand a community DāˆˆPt(orD=āˆ…) -such that āˆ†H(v/mapstoā†’D)>0. The MoveNodesFast function then moves node vto community D. This means that -Pt+1/negationslash=Ptand that the iteration is not stable. We now have a contradiction, which implies that the assumption of -Pt+1=Ptnot being node optimal must be false. Hence, Pt+1=Ptis node optimal.  -In the same way, it is straightforward to see that the Louvain algorithm also guarantees node optimality in a stable -iteration. -When the Louvain algorithm reaches a stable iteration, the partition is Ī³-separated and node optimal. Since the -Louvain algorithm considers only moving nodes and merging communities, additional iterations of the algorithm will -not lead to further improvements of the partition. Hence, in the case of the Louvain algorithm, if Pt+1=Pt, then -PĻ„=Ptfor allĻ„ā‰„t. In other words, when the Louvain algorithm reaches a stable iteration, all future iterations will -be stable as well. This contrasts with the Leiden algorithm, which may continue to improve a partition after a stable -iteration. We consider this in more detail below. -3. Asymptotic guarantees -When an iteration of the Leiden algorithm is stable, this does not imply that the next iteration will also be stable. -Because of randomness in the reļ¬nement phase of the Leiden algorithm, a partition that is stable in one iteration -may be improved in the next iteration. However, at some point, a partition will be obtained for which the Leiden -algorithm is unable to make any further improvements. We call this an asymptotically stable partition. Below, we -prove that an asymptotically stable partition is uniformly Ī³-dense and subset optimal. -We ļ¬rst need to show what it means to deļ¬ne asymptotic properties for the Leiden algorithm. The Leiden algorithm -considers moving a node to a diļ¬€erent community only if this results in a strict increase in the quality function. As -stated in the following lemma, this ensures that at some point the Leiden algorithm will ļ¬nd a partition for which it -can make no further improvements. -Lemma 8. LetG= (V,E)be a graph, and let Pt+1=Leiden (G,Pt). There exists a Ļ„such that Pt=PĻ„for all -tā‰„Ļ„. -Proof.Only strict improvements can be made in the Leiden algorithm. Consequently, if Pt+1/negationslash=Pt, thenPt+1/negationslash=Pt/prime -for allt/primeā‰¤t. Assume that there does not exist a Ļ„such that Pt=PĻ„for alltā‰„Ļ„. Then for any Ļ„there exists a -t>Ļ„such that Pt/negationslash=Pt/primefor allt/prime0for someDāˆˆPor forD=āˆ…. A setSāŠ†CāˆˆPis called a minimal non-optimal subset if -Sis a non-optimal subset and if there does not exist a non-optimal subset S/primeāŠ‚S. -The following lemma states an important property of minimal non-optimal subsets. -Lemma 9. LetG= (V,E)be a graph, let Pbe a partition of G, and letSāŠ†CāˆˆPbe a minimal non-optimal -subset. Then{S}is an optimal partition of the subgraph induced by S. -Proof.Assume that{S}is not an optimal partition of the subgraph induced by S. There then exists a set S1āˆˆS -such that -E(S1,S2)āˆ’Ī³/bardblS1/bardblĀ·/bardblS2/bardbl<0, (D2) -whereS2=Sāˆ’S1. LetDāˆˆPorD=āˆ…such that āˆ†H(Sā†’D)>0. Hence, -E(S,D)āˆ’Ī³/bardblS/bardblĀ·/bardblD/bardbl>E(S,Cāˆ’S)āˆ’Ī³/bardblS/bardblĀ·/bardblCāˆ’S/bardbl. (D3)22 -BecauseSis a minimal non-optimal subset, S1andS2cannot be non-optimal subsets. Therefore, āˆ†H(S1ā†’D)ā‰¤0 -andāˆ†H(S2ā†’D)ā‰¤0, or equivalently, -E(S1,D)āˆ’Ī³/bardblS1/bardblĀ·/bardblD/bardblā‰¤E(S1,Cāˆ’S1)āˆ’Ī³/bardblS1/bardblĀ·/bardblCāˆ’S1/bardbl (D4) -and -E(S2,D)āˆ’Ī³/bardblS2/bardblĀ·/bardblD/bardblā‰¤E(S2,Cāˆ’S2)āˆ’Ī³/bardblS2/bardblĀ·/bardblCāˆ’S2/bardbl. (D5) -It then follows from Eqs. (D4) and (D5) that -E(S,D)āˆ’Ī³/bardblS/bardblĀ·/bardblD/bardbl=/parenleftbig -E(S1,D)āˆ’Ī³/bardblS1/bardblĀ·/bardblD/bardbl/parenrightbig -+/parenleftbig -E(S2,D)āˆ’Ī³/bardblS2/bardblĀ·/bardblD/bardbl/parenrightbig -ā‰¤/parenleftbig -E(S1,Cāˆ’S1)āˆ’Ī³/bardblS1/bardblĀ·/bardblCāˆ’S1/bardbl/parenrightbig -+/parenleftbig -E(S2,Cāˆ’S2)āˆ’Ī³/bardblS2/bardblĀ·/bardblCāˆ’S2/bardbl/parenrightbig -. -This can be written as -E(S,D)āˆ’Ī³/bardblS/bardblĀ·/bardblD/bardblā‰¤/parenleftbig -E(S1,Cāˆ’S) +E(S1,S2)āˆ’Ī³/bardblS1/bardblĀ·/bardblCāˆ’S/bardblāˆ’Ī³/bardblS1/bardblĀ·/bardblS2/bardbl/parenrightbig -+/parenleftbig -E(S2,Cāˆ’S) +E(S2,S1)āˆ’Ī³/bardblS2/bardblĀ·/bardblCāˆ’S/bardblāˆ’Ī³/bardblS2/bardblĀ·/bardblS1/bardbl/parenrightbig -=E(S,Cāˆ’S) + 2E(S1,S2)āˆ’Ī³/bardblS/bardblĀ·/bardblCāˆ’S/bardblāˆ’2Ī³/bardblS1/bardblĀ·/bardblS2/bardbl. -Using Eq. (D2), we then obtain -E(S,D)āˆ’Ī³/bardblS/bardblĀ·/bardblD/bardbl ./ragtest/input/book.txt\n\nNext we'll inject some required config variables:\n\nSet Up Your Workspace Variables\n-------------------------------\n\nFirst let's make sure to setup the required environment variables. For details on these environment variables, and what environment variables are available, see the [variables documentation](/graphrag/posts/config/overview/)\n.\n\nTo initialize your workspace, let's first run the `graphrag.index --init` command. Since we have already configured a directory named .ragtest\\` in the previous step, we can run the following command:\n\n python -m graphrag.index --init --root ./ragtest\n\nThis will create two files: `.env` and `settings.yaml` in the `./ragtest` directory.\n\n* `.env` contains the environment variables required to run the GraphRAG pipeline. If you inspect the file, you'll see a single environment variable defined, `GRAPHRAG_API_KEY=`. This is the API key for the OpenAI API or Azure OpenAI endpoint. You can replace this with your own API key.\n* `settings.yaml` contains the settings for the pipeline. You can modify this file to change the settings for the pipeline. \n \n\n#### OpenAI and Azure OpenAI\n\nTo run in OpenAI mode, just make sure to update the value of `GRAPHRAG_API_KEY` in the `.env` file with your OpenAI API key.\n\n#### Azure OpenAI\n\nIn addition, Azure OpenAI users should set the following variables in the settings.yaml file. To find the appropriate sections, just search for the `llm:` configuration, you should see two sections, one for the chat endpoint and one for the embeddings endpoint. Here is an example of how to configure the chat endpoint:\n\n type: azure_openai_chat # Or azure_openai_embedding for embeddings\n api_base: https://.openai.azure.com\n api_version: 2024-02-15-preview # You can customize this for other versions\n deployment_name: \n\n* For more details about configuring GraphRAG, see the [configuration documentation](/graphrag/posts/config/overview/)\n .\n* To learn more about Initialization, refer to the [Initialization documentation](/graphrag/posts/config/init/)\n .\n* For more details about using the CLI, refer to the [CLI documentation](/graphrag/posts/query/3-cli/)\n .\n\nRunning the Indexing pipeline\n-----------------------------\n\nFinally we'll run the pipeline!\n\n python -m graphrag.index --root ./ragtest\n\n![pipeline executing from the CLI](https://microsoft.github.io/graphrag/img/pipeline-running.png)\n\nThis process will take some time to run. This depends on the size of your input data, what model you're using, and the text chunk size being used (these can be configured in your `.env` file). Once the pipeline is complete, you should see a new folder called `./ragtest/output//artifacts` with a series of parquet files.\n\nUsing the Query Engine\n======================\n\nRunning the Query Engine\n------------------------\n\nNow let's ask some questions using this dataset.\n\nHere is an example using Global search to ask a high-level question:\n\n python -m graphrag.query \\\n --root ./ragtest \\\n --method global \\\n \"What are the top themes in this story?\"\n\nHere is an example using Local search to ask a more specific question about a particular character:\n\n python -m graphrag.query \\\n --root ./ragtest \\\n --method local \\\n \"Who is Scrooge, and what are his main relationships?\"\n\nPlease refer to [Query Engine](/graphrag/posts/query/overview)\n docs for detailed information about how to leverage our Local and Global search mechanisms for extracting meaningful insights from data after the Indexer has wrapped up execution.", - "markdown": "Get Started\n===========\n\nRequirements\n------------\n\n[Python 3.10-3.12](https://www.python.org/downloads/)\n\nTo get started with the GraphRAG system, you have a few options:\n\nšŸ‘‰ [Use the GraphRAG Accelerator solution](https://github.com/Azure-Samples/graphrag-accelerator)\n \nšŸ‘‰ [Install from pypi](https://pypi.org/project/graphrag/)\n. \nšŸ‘‰ [Use it from source](/graphrag/posts/developing)\n \n\nQuickstart\n----------\n\nTo get started with the GraphRAG system we recommend trying the [Solution Accelerator](https://github.com/Azure-Samples/graphrag-accelerator)\n package. This provides a user-friendly end-to-end experience with Azure resources.\n\nTop-Level Modules\n=================\n\n[Indexing Pipeline Overview](/graphrag/posts/index/overview)\n \n[Query Engine Overview](/graphrag/posts/query/overview)\n\nOverview\n========\n\nThe following is a simple end-to-end example for using the GraphRAG system. It shows how to use the system to index some text, and then use the indexed data to answer questions about the documents.\n\nInstall GraphRAG\n================\n\n pip install graphrag\n\nRunning the Indexer\n===================\n\nNow we need to set up a data project and some initial configuration. Let's set that up. We're using the [default configuration mode](/graphrag/posts/config/overview/)\n, which you can customize as needed using a [config file](/graphrag/posts/config/json_yaml/)\n, which we recommend, or [environment variables](/graphrag/posts/config/env_vars/)\n.\n\nFirst let's get a sample dataset ready:\n\n mkdir -p ./ragtest/input\n\nNow let's get a copy of A Christmas Carol by Charles Dickens from a trusted source\n\n curl https://www.gutenberg.org/cache/epub/24022/pg24022.txt > ./ragtest/input/book.txt\n\nNext we'll inject some required config variables:\n\nSet Up Your Workspace Variables\n-------------------------------\n\nFirst let's make sure to setup the required environment variables. For details on these environment variables, and what environment variables are available, see the [variables documentation](/graphrag/posts/config/overview/)\n.\n\nTo initialize your workspace, let's first run the `graphrag.index --init` command. Since we have already configured a directory named .ragtest\\` in the previous step, we can run the following command:\n\n python -m graphrag.index --init --root ./ragtest\n\nThis will create two files: `.env` and `settings.yaml` in the `./ragtest` directory.\n\n* `.env` contains the environment variables required to run the GraphRAG pipeline. If you inspect the file, you'll see a single environment variable defined, `GRAPHRAG_API_KEY=`. This is the API key for the OpenAI API or Azure OpenAI endpoint. You can replace this with your own API key.\n* `settings.yaml` contains the settings for the pipeline. You can modify this file to change the settings for the pipeline. \n \n\n#### OpenAI and Azure OpenAI\n\nTo run in OpenAI mode, just make sure to update the value of `GRAPHRAG_API_KEY` in the `.env` file with your OpenAI API key.\n\n#### Azure OpenAI\n\nIn addition, Azure OpenAI users should set the following variables in the settings.yaml file. To find the appropriate sections, just search for the `llm:` configuration, you should see two sections, one for the chat endpoint and one for the embeddings endpoint. Here is an example of how to configure the chat endpoint:\n\n type: azure_openai_chat # Or azure_openai_embedding for embeddings\n api_base: https://.openai.azure.com\n api_version: 2024-02-15-preview # You can customize this for other versions\n deployment_name: \n\n* For more details about configuring GraphRAG, see the [configuration documentation](/graphrag/posts/config/overview/)\n .\n* To learn more about Initialization, refer to the [Initialization documentation](/graphrag/posts/config/init/)\n .\n* For more details about using the CLI, refer to the [CLI documentation](/graphrag/posts/query/3-cli/)\n .\n\nRunning the Indexing pipeline\n-----------------------------\n\nFinally we'll run the pipeline!\n\n python -m graphrag.index --root ./ragtest\n\n![pipeline executing from the CLI](https://microsoft.github.io/graphrag/img/pipeline-running.png)\n\nThis process will take some time to run. This depends on the size of your input data, what model you're using, and the text chunk size being used (these can be configured in your `.env` file). Once the pipeline is complete, you should see a new folder called `./ragtest/output//artifacts` with a series of parquet files.\n\nUsing the Query Engine\n======================\n\nRunning the Query Engine\n------------------------\n\nNow let's ask some questions using this dataset.\n\nHere is an example using Global search to ask a high-level question:\n\n python -m graphrag.query \\\n --root ./ragtest \\\n --method global \\\n \"What are the top themes in this story?\"\n\nHere is an example using Local search to ask a more specific question about a particular character:\n\n python -m graphrag.query \\\n --root ./ragtest \\\n --method local \\\n \"Who is Scrooge, and what are his main relationships?\"\n\nPlease refer to [Query Engine](/graphrag/posts/query/overview)\n docs for detailed information about how to leverage our Local and Global search mechanisms for extracting meaningful insights from data after the Indexer has wrapped up execution.", - "metadata": { - "title": "Get Started", - "sourceURL": "https://microsoft.github.io/graphrag/posts/get_started/", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://www.python.org/downloads/", - "https://github.com/Azure-Samples/graphrag-accelerator", - "https://pypi.org/project/graphrag/", - "https://microsoft.github.io/graphrag/posts/developing", - "https://microsoft.github.io/graphrag/posts/index/overview", - "https://microsoft.github.io/graphrag/posts/query/overview", - "https://microsoft.github.io/graphrag/posts/config/json_yaml/", - "https://microsoft.github.io/graphrag/posts/config/env_vars/", - "https://microsoft.github.io/graphrag/posts/config/init/", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag" - ] - }, - { - "content": "Developing GraphRAG\n===================\n\nRequirements\n============\n\n| Name | Installation | Purpose |\n| --- | --- | --- |\n| Python 3.10-3.12 | [Download](https://www.python.org/downloads/) | The library is Python-based. |\n| Poetry | [Instructions](https://python-poetry.org/docs/#installation) | Poetry is used for package management and virtualenv management in Python codebases |\n\nGetting Started\n===============\n\nInstall Dependencies\n--------------------\n\n # Install Python dependencies.\n poetry install\n\nExecute the Indexing Engine\n---------------------------\n\n poetry run poe index <...args>\n\nExecuting Queries\n-----------------\n\n poetry run poe query <...args>\n\nAzurite\n=======\n\nSome unit and smoke tests use Azurite to emulate Azure resources. This can be started by running:\n\n ./scripts/start-azurite.sh\n\nor by simply running `azurite` in the terminal if already installed globally. See the [Azurite documentation](https://learn.microsoft.com/en-us/azure/storage/common/storage-use-azurite)\n for more information about how to install and use Azurite.\n\nLifecycle Scripts\n=================\n\nOur Python package utilizes Poetry to manage dependencies and [poethepoet](https://pypi.org/project/poethepoet/)\n to manage build scripts.\n\nAvailable scripts are:\n\n* `poetry run poe index` - Run the Indexing CLI\n* `poetry run poe query` - Run the Query CLI\n* `poetry build` - This invokes `poetry build`, which will build a wheel file and other distributable artifacts.\n* `poetry run poe test` - This will execute all tests.\n* `poetry run poe test_unit` - This will execute unit tests.\n* `poetry run poe test_integration` - This will execute integration tests.\n* `poetry run poe test_smoke` - This will execute smoke tests.\n* `poetry run poe check` - This will perform a suite of static checks across the package, including:\n * formatting\n * documentation formatting\n * linting\n * security patterns\n * type-checking\n* `poetry run poe fix` - This will apply any available auto-fixes to the package. Usually this is just formatting fixes.\n* `poetry run poe fix_unsafe` - This will apply any available auto-fixes to the package, including those that may be unsafe.\n* `poetry run poe format` - Explicitly run the formatter across the package.\n\nTroubleshooting\n---------------\n\n### \"RuntimeError: llvm-config failed executing, please point LLVM\\_CONFIG to the path for llvm-config\" when running poetry install\n\nMake sure llvm-9 and llvm-9-dev are installed:\n\n`sudo apt-get install llvm-9 llvm-9-dev`\n\nand then in your bashrc, add\n\n`export LLVM_CONFIG=/usr/bin/llvm-config-9`\n\n### \"numba/\\_pymodule.h:6:10: fatal error: Python.h: No such file or directory\" when running poetry install\n\nMake sure you have python3.10-dev installed or more generally `python-dev`\n\n`sudo apt-get install python3.10-dev`\n\n### LLM call constantly exceeds TPM, RPM or time limits\n\n`GRAPHRAG_LLM_THREAD_COUNT` and `GRAPHRAG_EMBEDDING_THREAD_COUNT` are both set to 50 by default. You can modify this values to reduce concurrency. Please refer to the [Configuration Documents](../config/overview)", - "markdown": "Developing GraphRAG\n===================\n\nRequirements\n============\n\n| Name | Installation | Purpose |\n| --- | --- | --- |\n| Python 3.10-3.12 | [Download](https://www.python.org/downloads/) | The library is Python-based. |\n| Poetry | [Instructions](https://python-poetry.org/docs/#installation) | Poetry is used for package management and virtualenv management in Python codebases |\n\nGetting Started\n===============\n\nInstall Dependencies\n--------------------\n\n # Install Python dependencies.\n poetry install\n\nExecute the Indexing Engine\n---------------------------\n\n poetry run poe index <...args>\n\nExecuting Queries\n-----------------\n\n poetry run poe query <...args>\n\nAzurite\n=======\n\nSome unit and smoke tests use Azurite to emulate Azure resources. This can be started by running:\n\n ./scripts/start-azurite.sh\n\nor by simply running `azurite` in the terminal if already installed globally. See the [Azurite documentation](https://learn.microsoft.com/en-us/azure/storage/common/storage-use-azurite)\n for more information about how to install and use Azurite.\n\nLifecycle Scripts\n=================\n\nOur Python package utilizes Poetry to manage dependencies and [poethepoet](https://pypi.org/project/poethepoet/)\n to manage build scripts.\n\nAvailable scripts are:\n\n* `poetry run poe index` - Run the Indexing CLI\n* `poetry run poe query` - Run the Query CLI\n* `poetry build` - This invokes `poetry build`, which will build a wheel file and other distributable artifacts.\n* `poetry run poe test` - This will execute all tests.\n* `poetry run poe test_unit` - This will execute unit tests.\n* `poetry run poe test_integration` - This will execute integration tests.\n* `poetry run poe test_smoke` - This will execute smoke tests.\n* `poetry run poe check` - This will perform a suite of static checks across the package, including:\n * formatting\n * documentation formatting\n * linting\n * security patterns\n * type-checking\n* `poetry run poe fix` - This will apply any available auto-fixes to the package. Usually this is just formatting fixes.\n* `poetry run poe fix_unsafe` - This will apply any available auto-fixes to the package, including those that may be unsafe.\n* `poetry run poe format` - Explicitly run the formatter across the package.\n\nTroubleshooting\n---------------\n\n### \"RuntimeError: llvm-config failed executing, please point LLVM\\_CONFIG to the path for llvm-config\" when running poetry install\n\nMake sure llvm-9 and llvm-9-dev are installed:\n\n`sudo apt-get install llvm-9 llvm-9-dev`\n\nand then in your bashrc, add\n\n`export LLVM_CONFIG=/usr/bin/llvm-config-9`\n\n### \"numba/\\_pymodule.h:6:10: fatal error: Python.h: No such file or directory\" when running poetry install\n\nMake sure you have python3.10-dev installed or more generally `python-dev`\n\n`sudo apt-get install python3.10-dev`\n\n### LLM call constantly exceeds TPM, RPM or time limits\n\n`GRAPHRAG_LLM_THREAD_COUNT` and `GRAPHRAG_EMBEDDING_THREAD_COUNT` are both set to 50 by default. You can modify this values to reduce concurrency. Please refer to the [Configuration Documents](../config/overview)", - "metadata": { - "title": "Developing GraphRAG", - "sourceURL": "https://microsoft.github.io/graphrag/posts/developing/", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://www.python.org/downloads/", - "https://python-poetry.org/docs/#installation", - "https://learn.microsoft.com/en-us/azure/storage/common/storage-use-azurite", - "https://pypi.org/project/poethepoet/", - "https://microsoft.github.io/graphrag/posts/developing//../config/overview", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "GraphRAG Indexing šŸ¤–\n====================\n\nThe GraphRAG indexing package is a data pipeline and transformation suite that is designed to extract meaningful, structured data from unstructured text using LLMs.\n\nIndexing Pipelines are configurable. They are composed of workflows, standard and custom steps, prompt templates, and input/output adapters. Our standard pipeline is designed to:\n\n* extract entities, relationships and claims from raw text\n* perform community detection in entities\n* generate community summaries and reports at multiple levels of granularity\n* embed entities into a graph vector space\n* embed text chunks into a textual vector space\n\nThe outputs of the pipeline can be stored in a variety of formats, including JSON and Parquet - or they can be handled manually via the Python API.\n\nGetting Started\n---------------\n\n### Requirements\n\nSee the [requirements](/graphrag/posts/developing#requirements)\n section in [Get Started](/graphrag/posts/get_started)\n for details on setting up a development environment.\n\nThe Indexing Engine can be used in either a default configuration mode or with a custom pipeline. To configure GraphRAG, see the [configuration](/graphrag/posts/config/overview)\n documentation. After you have a config file you can run the pipeline using the CLI or the Python API.\n\nUsage\n-----\n\n### CLI\n\n # Via Poetry\n poetry run poe cli --root # default config mode\n poetry run poe cli --config your_pipeline.yml # custom config mode\n \n # Via Node\n yarn run:index --root # default config mode\n yarn run:index --config your_pipeline.yml # custom config mode\n \n\n### Python API\n\n from graphrag.index import run_pipeline\n from graphrag.index.config import PipelineWorkflowReference\n \n workflows: list[PipelineWorkflowReference] = [\\\n PipelineWorkflowReference(\\\n steps=[\\\n {\\\n # built-in verb\\\n \"verb\": \"derive\", # https://github.com/microsoft/datashaper/blob/main/python/datashaper/datashaper/engine/verbs/derive.py\\\n \"args\": {\\\n \"column1\": \"col1\", # from above\\\n \"column2\": \"col2\", # from above\\\n \"to\": \"col_multiplied\", # new column name\\\n \"operator\": \"*\", # multiply the two columns\\\n },\\\n # Since we're trying to act on the default input, we don't need explicitly to specify an input\\\n }\\\n ]\\\n ),\\\n ]\n \n dataset = pd.DataFrame([{\"col1\": 2, \"col2\": 4}, {\"col1\": 5, \"col2\": 10}])\n outputs = []\n async for output in await run_pipeline(dataset=dataset, workflows=workflows):\n outputs.append(output)\n pipeline_result = outputs[-1]\n print(pipeline_result)\n\nFurther Reading\n---------------\n\n* To start developing within the _GraphRAG_ project, see [getting started](/graphrag/posts/developing/)\n \n* To understand the underlying concepts and execution model of the indexing library, see [the architecture documentation](/graphrag/posts/index/0-architecture/)\n \n* To get running with a series of examples, see [the examples documentation](https://github.com/microsoft/graphrag/blob/main/examples/README.md)\n \n* To read more about configuring the indexing engine, see [the configuration documentation](/graphrag/posts/config/overview)", - "markdown": "GraphRAG Indexing šŸ¤–\n====================\n\nThe GraphRAG indexing package is a data pipeline and transformation suite that is designed to extract meaningful, structured data from unstructured text using LLMs.\n\nIndexing Pipelines are configurable. They are composed of workflows, standard and custom steps, prompt templates, and input/output adapters. Our standard pipeline is designed to:\n\n* extract entities, relationships and claims from raw text\n* perform community detection in entities\n* generate community summaries and reports at multiple levels of granularity\n* embed entities into a graph vector space\n* embed text chunks into a textual vector space\n\nThe outputs of the pipeline can be stored in a variety of formats, including JSON and Parquet - or they can be handled manually via the Python API.\n\nGetting Started\n---------------\n\n### Requirements\n\nSee the [requirements](/graphrag/posts/developing#requirements)\n section in [Get Started](/graphrag/posts/get_started)\n for details on setting up a development environment.\n\nThe Indexing Engine can be used in either a default configuration mode or with a custom pipeline. To configure GraphRAG, see the [configuration](/graphrag/posts/config/overview)\n documentation. After you have a config file you can run the pipeline using the CLI or the Python API.\n\nUsage\n-----\n\n### CLI\n\n # Via Poetry\n poetry run poe cli --root # default config mode\n poetry run poe cli --config your_pipeline.yml # custom config mode\n \n # Via Node\n yarn run:index --root # default config mode\n yarn run:index --config your_pipeline.yml # custom config mode\n \n\n### Python API\n\n from graphrag.index import run_pipeline\n from graphrag.index.config import PipelineWorkflowReference\n \n workflows: list[PipelineWorkflowReference] = [\\\n PipelineWorkflowReference(\\\n steps=[\\\n {\\\n # built-in verb\\\n \"verb\": \"derive\", # https://github.com/microsoft/datashaper/blob/main/python/datashaper/datashaper/engine/verbs/derive.py\\\n \"args\": {\\\n \"column1\": \"col1\", # from above\\\n \"column2\": \"col2\", # from above\\\n \"to\": \"col_multiplied\", # new column name\\\n \"operator\": \"*\", # multiply the two columns\\\n },\\\n # Since we're trying to act on the default input, we don't need explicitly to specify an input\\\n }\\\n ]\\\n ),\\\n ]\n \n dataset = pd.DataFrame([{\"col1\": 2, \"col2\": 4}, {\"col1\": 5, \"col2\": 10}])\n outputs = []\n async for output in await run_pipeline(dataset=dataset, workflows=workflows):\n outputs.append(output)\n pipeline_result = outputs[-1]\n print(pipeline_result)\n\nFurther Reading\n---------------\n\n* To start developing within the _GraphRAG_ project, see [getting started](/graphrag/posts/developing/)\n \n* To understand the underlying concepts and execution model of the indexing library, see [the architecture documentation](/graphrag/posts/index/0-architecture/)\n \n* To get running with a series of examples, see [the examples documentation](https://github.com/microsoft/graphrag/blob/main/examples/README.md)\n \n* To read more about configuring the indexing engine, see [the configuration documentation](/graphrag/posts/config/overview)", - "metadata": { - "title": "GraphRAG Indexing šŸ¤–", - "sourceURL": "https://microsoft.github.io/graphrag/posts/index/overview/", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://microsoft.github.io/graphrag/posts/developing#requirements", - "https://microsoft.github.io/graphrag/posts/get_started", - "https://microsoft.github.io/graphrag/posts/config/overview", - "https://github.com/microsoft/graphrag/blob/main/examples/README.md", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Indexing Architecture\n=====================\n\nKey Concepts\n------------\n\n### Knowledge Model\n\nIn order to support the GraphRAG system, the outputs of the indexing engine (in the Default Configuration Mode) are aligned to a knowledge model we call the _GraphRAG Knowledge Model_. This model is designed to be an abstraction over the underlying data storage technology, and to provide a common interface for the GraphRAG system to interact with. In normal use-cases the outputs of the GraphRAG Indexer would be loaded into a database system, and the GraphRAG's Query Engine would interact with the database using the knowledge model data-store types.\n\n### DataShaper Workflows\n\nGraphRAG's Indexing Pipeline is built on top of our open-source library, [DataShaper](https://github.com/microsoft/datashaper)\n. DataShaper is a data processing library that allows users to declaratively express data pipelines, schemas, and related assets using well-defined schemas. DataShaper has implementations in JavaScript and Python, and is designed to be extensible to other languages.\n\nOne of the core resource types within DataShaper is a [Workflow](https://github.com/microsoft/datashaper/blob/main/javascript/schema/src/workflow/WorkflowSchema.ts)\n. Workflows are expressed as sequences of steps, which we call [verbs](https://github.com/microsoft/datashaper/blob/main/javascript/schema/src/workflow/verbs.ts)\n. Each step has a verb name and a configuration object. In DataShaper, these verbs model relational concepts such as SELECT, DROP, JOIN, etc.. Each verb transforms an input data table, and that table is passed down the pipeline.\n\n### LLM-based Workflow Steps\n\nGraphRAG's Indexing Pipeline implements a handful of custom verbs on top of the standard, relational verbs that our DataShaper library provides. These verbs give us the ability to augment text documents with rich, structured data using the power of LLMs such as GPT-4. We utilize these verbs in our standard workflow to extract entities, relationships, claims, community structures, and community reports and summaries. This behavior is customizable and can be extended to support many kinds of AI-based data enrichment and extraction tasks.\n\n### Workflow Graphs\n\nBecause of the complexity of our data indexing tasks, we needed to be able to express our data pipeline as series of multiple, interdependent workflows. In the GraphRAG Indexing Pipeline, each workflow may define dependencies on other workflows, effectively forming a directed acyclic graph (DAG) of workflows, which is then used to schedule processing.\n\n\\---\ntitle: Sample Workflow DAG\n---\nstateDiagram-v2\n \\[\\*\\] --> Prepare\n Prepare --> Chunk\n Chunk --> ExtractGraph\n Chunk --> EmbedDocuments\n ExtractGraph --> GenerateReports\n ExtractGraph --> EmbedGraph\n EntityResolution --> EmbedGraph\n EntityResolution --> GenerateReports\n ExtractGraph --> EntityResolution\n\n### Dataframe Message Format\n\nThe primary unit of communication between workflows, and between workflow steps is an instance of `pandas.DataFrame`. Although side-effects are possible, our goal is to be _data-centric_ and _table-centric_ in our approach to data processing. This allows us to easily reason about our data, and to leverage the power of dataframe-based ecosystems. Our underlying dataframe technology may change over time, but our primary goal is to support the DataShaper workflow schema while retaining single-machine ease of use and developer ergonomics.\n\n### LLM Caching\n\nThe GraphRAG library was designed with LLM interactions in mind, and a common setback when working with LLM APIs is various errors errors due to network latency, throttling, etc.. Because of these potential error cases, we've added a cache layer around LLM interactions. When completion requests are made using the same input set (prompt and tuning parameters), we return a cached result if one exists. This allows our indexer to be more resilient to network issues, to act idempotently, and to provide a more efficient end-user experience.", - "markdown": "Indexing Architecture\n=====================\n\nKey Concepts\n------------\n\n### Knowledge Model\n\nIn order to support the GraphRAG system, the outputs of the indexing engine (in the Default Configuration Mode) are aligned to a knowledge model we call the _GraphRAG Knowledge Model_. This model is designed to be an abstraction over the underlying data storage technology, and to provide a common interface for the GraphRAG system to interact with. In normal use-cases the outputs of the GraphRAG Indexer would be loaded into a database system, and the GraphRAG's Query Engine would interact with the database using the knowledge model data-store types.\n\n### DataShaper Workflows\n\nGraphRAG's Indexing Pipeline is built on top of our open-source library, [DataShaper](https://github.com/microsoft/datashaper)\n. DataShaper is a data processing library that allows users to declaratively express data pipelines, schemas, and related assets using well-defined schemas. DataShaper has implementations in JavaScript and Python, and is designed to be extensible to other languages.\n\nOne of the core resource types within DataShaper is a [Workflow](https://github.com/microsoft/datashaper/blob/main/javascript/schema/src/workflow/WorkflowSchema.ts)\n. Workflows are expressed as sequences of steps, which we call [verbs](https://github.com/microsoft/datashaper/blob/main/javascript/schema/src/workflow/verbs.ts)\n. Each step has a verb name and a configuration object. In DataShaper, these verbs model relational concepts such as SELECT, DROP, JOIN, etc.. Each verb transforms an input data table, and that table is passed down the pipeline.\n\n### LLM-based Workflow Steps\n\nGraphRAG's Indexing Pipeline implements a handful of custom verbs on top of the standard, relational verbs that our DataShaper library provides. These verbs give us the ability to augment text documents with rich, structured data using the power of LLMs such as GPT-4. We utilize these verbs in our standard workflow to extract entities, relationships, claims, community structures, and community reports and summaries. This behavior is customizable and can be extended to support many kinds of AI-based data enrichment and extraction tasks.\n\n### Workflow Graphs\n\nBecause of the complexity of our data indexing tasks, we needed to be able to express our data pipeline as series of multiple, interdependent workflows. In the GraphRAG Indexing Pipeline, each workflow may define dependencies on other workflows, effectively forming a directed acyclic graph (DAG) of workflows, which is then used to schedule processing.\n\n\\---\ntitle: Sample Workflow DAG\n---\nstateDiagram-v2\n \\[\\*\\] --> Prepare\n Prepare --> Chunk\n Chunk --> ExtractGraph\n Chunk --> EmbedDocuments\n ExtractGraph --> GenerateReports\n ExtractGraph --> EmbedGraph\n EntityResolution --> EmbedGraph\n EntityResolution --> GenerateReports\n ExtractGraph --> EntityResolution\n\n### Dataframe Message Format\n\nThe primary unit of communication between workflows, and between workflow steps is an instance of `pandas.DataFrame`. Although side-effects are possible, our goal is to be _data-centric_ and _table-centric_ in our approach to data processing. This allows us to easily reason about our data, and to leverage the power of dataframe-based ecosystems. Our underlying dataframe technology may change over time, but our primary goal is to support the DataShaper workflow schema while retaining single-machine ease of use and developer ergonomics.\n\n### LLM Caching\n\nThe GraphRAG library was designed with LLM interactions in mind, and a common setback when working with LLM APIs is various errors errors due to network latency, throttling, etc.. Because of these potential error cases, we've added a cache layer around LLM interactions. When completion requests are made using the same input set (prompt and tuning parameters), we return a cached result if one exists. This allows our indexer to be more resilient to network issues, to act idempotently, and to provide a more efficient end-user experience.", - "metadata": { - "title": "Indexing Architecture", - "sourceURL": "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://github.com/microsoft/datashaper", - "https://github.com/microsoft/datashaper/blob/main/javascript/schema/src/workflow/WorkflowSchema.ts", - "https://github.com/microsoft/datashaper/blob/main/javascript/schema/src/workflow/verbs.ts", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Indexing Dataflow\n=================\n\nThe GraphRAG Knowledge Model\n----------------------------\n\nThe knowledge model is a specification for data outputs that conform to our data-model definition. You can find these definitions in the python/graphrag/graphrag/model folder within the GraphRAG repository. The following entity types are provided. The fields here represent the fields that are text-embedded by default.\n\n* `Document` - An input document into the system. These either represent individual rows in a CSV or individual .txt file.\n* `TextUnit` - A chunk of text to analyze. The size of these chunks, their overlap, and whether they adhere to any data boundaries may be configured below. A common use case is to set `CHUNK_BY_COLUMNS` to `id` so that there is a 1-to-many relationship between documents and TextUnits instead of a many-to-many.\n* `Entity` - An entity extracted from a TextUnit. These represent people, places, events, or some other entity-model that you provide.\n* `Relationship` - A relationship between two entities. These are generated from the covariates.\n* `Covariate` - Extracted claim information, which contains statements about entities which may be time-bound.\n* `Community Report` - Once entities are generated, we perform hierarchical community detection on them and generate reports for each community in this hierarchy.\n* `Node` - This table contains layout information for rendered graph-views of the Entities and Documents which have been embedded and clustered.\n\nThe Default Configuration Workflow\n----------------------------------\n\nLet's take a look at how the default-configuration workflow transforms text documents into the _GraphRAG Knowledge Model_. This page gives a general overview of the major steps in this process. To fully configure this workflow, check out the [configuration](/graphrag/posts/config/overview/)\n documentation.\n\nPhase 1: Compose TextUnits\n--------------------------\n\nThe first phase of the default-configuration workflow is to transform input documents into _TextUnits_. A _TextUnit_ is a chunk of text that is used for our graph extraction techniques. They are also used as source-references by extracted knowledge items in order to empower breadcrumbs and provenance by concepts back to their original source tex.\n\nThe chunk size (counted in tokens), is user-configurable. By default this is set to 300 tokens, although we've had positive experience with 1200-token chunks using a single \"glean\" step. (A \"glean\" step is a follow-on extraction). Larger chunks result in lower-fidelity output and less meaningful reference texts; however, using larger chunks can result in much faster processing time.\n\nThe group-by configuration is also user-configurable. By default, we align our chunks to document boundaries, meaning that there is a strict 1-to-many relationship between Documents and TextUnits. In rare cases, this can be turned into a many-to-many relationship. This is useful when the documents are very short and we need several of them to compose a meaningful analysis unit (e.g. Tweets or a chat log)\n\nEach of these text-units are text-embedded and passed into the next phase of the pipeline.\n\n\\---\ntitle: Documents into Text Chunks\n---\nflowchart LR\n doc1\\[Document 1\\] --> tu1\\[TextUnit 1\\]\n doc1 --> tu2\\[TextUnit 2\\]\n doc2\\[Document 2\\] --> tu3\\[TextUnit 3\\]\n doc2 --> tu4\\[TextUnit 4\\]\n\nPhase 2: Graph Extraction\n-------------------------\n\nIn this phase, we analyze each text unit and extract our graph primitives: _Entities_, _Relationships_, and _Claims_. Entities and Relationships are extracted at once in our _entity\\_extract_ verb, and claims are extracted in our _claim\\_extract_ verb. Results are then combined and passed into following phases of the pipeline.\n\n\\---\ntitle: Graph Extraction\n---\nflowchart LR\n tu\\[TextUnit\\] --> ge\\[Graph Extraction\\] --> gs\\[Graph Summarization\\] --> er\\[Entity Resolution\\]\n tu --> ce\\[Claim Extraction\\]\n\n### Entity & Relationship Extraction\n\nIn this first step of graph extraction, we process each text-unit in order to extract entities and relationships out of the raw text using the LLM. The output of this step is a subgraph-per-TextUnit containing a list of **entities** with a _name_, _type_, and _description_, and a list of **relationships** with a _source_, _target_, and _description_.\n\nThese subgraphs are merged together - any entities with the same _name_ and _type_ are merged by creating an array of their descriptions. Similarly, any relationships with the same _source_ and _target_ are merged by creating an array of their descriptions.\n\n### Entity & Relationship Summarization\n\nNow that we have a graph of entities and relationships, each with a list of descriptions, we can summarize these lists into a single description per entity and relationship. This is done by asking the LLM for a short summary that captures all of the distinct information from each description. This allows all of our entities and relationships to have a single concise description.\n\n### Entity Resolution (Not Enabled by Default)\n\nThe final step of graph extraction is to resolve any entities that represent the same real-world entity but but have different names. Since this is done via LLM, and we don't want to lose information, we want to take a conservative, non-destructive approach to this.\n\nOur current implementation of Entity Resolution, however, is destructive. It will provide the LLM with a series of entities and ask it to determine which ones should be merged. Those entities are then merged together into a single entity and their relationships are updated.\n\nWe are currently exploring other entity resolution techniques. In the near future, entity resolution will be executed by creating an edge between entity variants indicating that the entities have been resolved by the indexing engine. This will allow for end-users to undo indexing-side resolutions, and add their own non-destructive resolutions using a similar process.\n\n### Claim Extraction & Emission\n\nFinally, as an independent workflow, we extract claims from the source TextUnits. These claims represent positive factual statements with an evaluated status and time-bounds. These are emitted as a primary artifact called **Covariates**.\n\nPhase 3: Graph Augmentation\n---------------------------\n\nNow that we have a usable graph of entities and relationships, we want to understand their community structure and augment the graph with additional information. This is done in two steps: _Community Detection_ and _Graph Embedding_. These give us explicit (communities) and implicit (embeddings) ways of understanding the topological structure of our graph.\n\n\\---\ntitle: Graph Augmentation\n---\nflowchart LR\n cd\\[Leiden Hierarchical Community Detection\\] --> ge\\[Node2Vec Graph Embedding\\] --> ag\\[Graph Table Emission\\]\n\n### Community Detection\n\nIn this step, we generate a hierarchy of entity communities using the Hierarchical Leiden Algorithm. This method will apply a recursive community-clustering to our graph until we reach a community-size threshold. This will allow us to understand the community structure of our graph and provide a way to navigate and summarize the graph at different levels of granularity.\n\n### Graph Embedding\n\nIn this step, we generate a vector representation of our graph using the Node2Vec algorithm. This will allow us to understand the implicit structure of our graph and provide an additional vector-space in which to search for related concepts during our query phase.\n\n### Graph Tables Emission\n\nOnce our graph augmentation steps are complete, the final **Entities** and **Relationships** tables are emitted after their text fields are text-embedded.\n\nPhase 4: Community Summarization\n--------------------------------\n\n\\---\ntitle: Community Summarization\n---\nflowchart LR\n sc\\[Generate Community Reports\\] --> ss\\[Summarize Community Reports\\] --> ce\\[Community Embedding\\] --> co\\[Community Tables Emission\\]\n\nAt this point, we have a functional graph of entities and relationships, a hierarchy of communities for the entities, as well as node2vec embeddings.\n\nNow we want to build on the communities data and generate reports for each community. This gives us a high-level understanding of the graph at several points of graph granularity. For example, if community A is the top-level community, we'll get a report about the entire graph. If the community is lower-level, we'll get a report about a local cluster.\n\n### Generate Community Reports\n\nIn this step, we generate a summary of each community using the LLM. This will allow us to understand the distinct information contained within each community and provide a scoped understanding of the graph, from either a high-level or a low-level perspective. These reports contain an executive overview and reference the key entities, relationships, and claims within the community sub-structure.\n\n### Summarize Community Reports\n\nIn this step, each _community report_ is then summarized via the LLM for shorthand use.\n\n### Community Embedding\n\nIn this step, we generate a vector representation of our communities by generating text embeddings of the community report, the community report summary, and the title of the community report.\n\n### Community Tables Emission\n\nAt this point, some bookkeeping work is performed and we emit the **Communities** and **CommunityReports** tables.\n\nPhase 5: Document Processing\n----------------------------\n\nIn this phase of the workflow, we create the _Documents_ table for the knowledge model.\n\n\\---\ntitle: Document Processing\n---\nflowchart LR\n aug\\[Augment\\] --> dp\\[Link to TextUnits\\] --> de\\[Avg. Embedding\\] --> dg\\[Document Table Emission\\]\n\n### Augment with Columns (CSV Only)\n\nIf the workflow is operating on CSV data, you may configure your workflow to add additional fields to Documents output. These fields should exist on the incoming CSV tables. Details about configuring this can be found in the [configuration documentation](/graphrag/posts/config/overview/)\n.\n\n### Link to TextUnits\n\nIn this step, we link each document to the text-units that were created in the first phase. This allows us to understand which documents are related to which text-units and vice-versa.\n\n### Document Embedding\n\nIn this step, we generate a vector representation of our documents using an average embedding of document slices. We re-chunk documents without overlapping chunks, and then generate an embedding for each chunk. We create an average of these chunks weighted by token-count and use this as the document embedding. This will allow us to understand the implicit relationship between documents, and will help us generate a network representation of our documents.\n\n### Documents Table Emission\n\nAt this point, we can emit the **Documents** table into the knowledge Model.\n\nPhase 6: Network Visualization\n------------------------------\n\nIn this phase of the workflow, we perform some steps to support network visualization of our high-dimensional vector spaces within our existing graphs. At this point there are two logical graphs at play: the _Entity-Relationship_ graph and the _Document_ graph.\n\n\\---\ntitle: Network Visualization Workflows\n---\nflowchart LR\n nv\\[Umap Documents\\] --> ne\\[Umap Entities\\] --> ng\\[Nodes Table Emission\\]\n\nFor each of the logical graphs, we perform a UMAP dimensionality reduction to generate a 2D representation of the graph. This will allow us to visualize the graph in a 2D space and understand the relationships between the nodes in the graph. The UMAP embeddings are then emitted as a table of _Nodes_. The rows of this table include a discriminator indicating whether the node is a document or an entity, and the UMAP coordinates.", - "markdown": "Indexing Dataflow\n=================\n\nThe GraphRAG Knowledge Model\n----------------------------\n\nThe knowledge model is a specification for data outputs that conform to our data-model definition. You can find these definitions in the python/graphrag/graphrag/model folder within the GraphRAG repository. The following entity types are provided. The fields here represent the fields that are text-embedded by default.\n\n* `Document` - An input document into the system. These either represent individual rows in a CSV or individual .txt file.\n* `TextUnit` - A chunk of text to analyze. The size of these chunks, their overlap, and whether they adhere to any data boundaries may be configured below. A common use case is to set `CHUNK_BY_COLUMNS` to `id` so that there is a 1-to-many relationship between documents and TextUnits instead of a many-to-many.\n* `Entity` - An entity extracted from a TextUnit. These represent people, places, events, or some other entity-model that you provide.\n* `Relationship` - A relationship between two entities. These are generated from the covariates.\n* `Covariate` - Extracted claim information, which contains statements about entities which may be time-bound.\n* `Community Report` - Once entities are generated, we perform hierarchical community detection on them and generate reports for each community in this hierarchy.\n* `Node` - This table contains layout information for rendered graph-views of the Entities and Documents which have been embedded and clustered.\n\nThe Default Configuration Workflow\n----------------------------------\n\nLet's take a look at how the default-configuration workflow transforms text documents into the _GraphRAG Knowledge Model_. This page gives a general overview of the major steps in this process. To fully configure this workflow, check out the [configuration](/graphrag/posts/config/overview/)\n documentation.\n\nPhase 1: Compose TextUnits\n--------------------------\n\nThe first phase of the default-configuration workflow is to transform input documents into _TextUnits_. A _TextUnit_ is a chunk of text that is used for our graph extraction techniques. They are also used as source-references by extracted knowledge items in order to empower breadcrumbs and provenance by concepts back to their original source tex.\n\nThe chunk size (counted in tokens), is user-configurable. By default this is set to 300 tokens, although we've had positive experience with 1200-token chunks using a single \"glean\" step. (A \"glean\" step is a follow-on extraction). Larger chunks result in lower-fidelity output and less meaningful reference texts; however, using larger chunks can result in much faster processing time.\n\nThe group-by configuration is also user-configurable. By default, we align our chunks to document boundaries, meaning that there is a strict 1-to-many relationship between Documents and TextUnits. In rare cases, this can be turned into a many-to-many relationship. This is useful when the documents are very short and we need several of them to compose a meaningful analysis unit (e.g. Tweets or a chat log)\n\nEach of these text-units are text-embedded and passed into the next phase of the pipeline.\n\n\\---\ntitle: Documents into Text Chunks\n---\nflowchart LR\n doc1\\[Document 1\\] --> tu1\\[TextUnit 1\\]\n doc1 --> tu2\\[TextUnit 2\\]\n doc2\\[Document 2\\] --> tu3\\[TextUnit 3\\]\n doc2 --> tu4\\[TextUnit 4\\]\n\nPhase 2: Graph Extraction\n-------------------------\n\nIn this phase, we analyze each text unit and extract our graph primitives: _Entities_, _Relationships_, and _Claims_. Entities and Relationships are extracted at once in our _entity\\_extract_ verb, and claims are extracted in our _claim\\_extract_ verb. Results are then combined and passed into following phases of the pipeline.\n\n\\---\ntitle: Graph Extraction\n---\nflowchart LR\n tu\\[TextUnit\\] --> ge\\[Graph Extraction\\] --> gs\\[Graph Summarization\\] --> er\\[Entity Resolution\\]\n tu --> ce\\[Claim Extraction\\]\n\n### Entity & Relationship Extraction\n\nIn this first step of graph extraction, we process each text-unit in order to extract entities and relationships out of the raw text using the LLM. The output of this step is a subgraph-per-TextUnit containing a list of **entities** with a _name_, _type_, and _description_, and a list of **relationships** with a _source_, _target_, and _description_.\n\nThese subgraphs are merged together - any entities with the same _name_ and _type_ are merged by creating an array of their descriptions. Similarly, any relationships with the same _source_ and _target_ are merged by creating an array of their descriptions.\n\n### Entity & Relationship Summarization\n\nNow that we have a graph of entities and relationships, each with a list of descriptions, we can summarize these lists into a single description per entity and relationship. This is done by asking the LLM for a short summary that captures all of the distinct information from each description. This allows all of our entities and relationships to have a single concise description.\n\n### Entity Resolution (Not Enabled by Default)\n\nThe final step of graph extraction is to resolve any entities that represent the same real-world entity but but have different names. Since this is done via LLM, and we don't want to lose information, we want to take a conservative, non-destructive approach to this.\n\nOur current implementation of Entity Resolution, however, is destructive. It will provide the LLM with a series of entities and ask it to determine which ones should be merged. Those entities are then merged together into a single entity and their relationships are updated.\n\nWe are currently exploring other entity resolution techniques. In the near future, entity resolution will be executed by creating an edge between entity variants indicating that the entities have been resolved by the indexing engine. This will allow for end-users to undo indexing-side resolutions, and add their own non-destructive resolutions using a similar process.\n\n### Claim Extraction & Emission\n\nFinally, as an independent workflow, we extract claims from the source TextUnits. These claims represent positive factual statements with an evaluated status and time-bounds. These are emitted as a primary artifact called **Covariates**.\n\nPhase 3: Graph Augmentation\n---------------------------\n\nNow that we have a usable graph of entities and relationships, we want to understand their community structure and augment the graph with additional information. This is done in two steps: _Community Detection_ and _Graph Embedding_. These give us explicit (communities) and implicit (embeddings) ways of understanding the topological structure of our graph.\n\n\\---\ntitle: Graph Augmentation\n---\nflowchart LR\n cd\\[Leiden Hierarchical Community Detection\\] --> ge\\[Node2Vec Graph Embedding\\] --> ag\\[Graph Table Emission\\]\n\n### Community Detection\n\nIn this step, we generate a hierarchy of entity communities using the Hierarchical Leiden Algorithm. This method will apply a recursive community-clustering to our graph until we reach a community-size threshold. This will allow us to understand the community structure of our graph and provide a way to navigate and summarize the graph at different levels of granularity.\n\n### Graph Embedding\n\nIn this step, we generate a vector representation of our graph using the Node2Vec algorithm. This will allow us to understand the implicit structure of our graph and provide an additional vector-space in which to search for related concepts during our query phase.\n\n### Graph Tables Emission\n\nOnce our graph augmentation steps are complete, the final **Entities** and **Relationships** tables are emitted after their text fields are text-embedded.\n\nPhase 4: Community Summarization\n--------------------------------\n\n\\---\ntitle: Community Summarization\n---\nflowchart LR\n sc\\[Generate Community Reports\\] --> ss\\[Summarize Community Reports\\] --> ce\\[Community Embedding\\] --> co\\[Community Tables Emission\\]\n\nAt this point, we have a functional graph of entities and relationships, a hierarchy of communities for the entities, as well as node2vec embeddings.\n\nNow we want to build on the communities data and generate reports for each community. This gives us a high-level understanding of the graph at several points of graph granularity. For example, if community A is the top-level community, we'll get a report about the entire graph. If the community is lower-level, we'll get a report about a local cluster.\n\n### Generate Community Reports\n\nIn this step, we generate a summary of each community using the LLM. This will allow us to understand the distinct information contained within each community and provide a scoped understanding of the graph, from either a high-level or a low-level perspective. These reports contain an executive overview and reference the key entities, relationships, and claims within the community sub-structure.\n\n### Summarize Community Reports\n\nIn this step, each _community report_ is then summarized via the LLM for shorthand use.\n\n### Community Embedding\n\nIn this step, we generate a vector representation of our communities by generating text embeddings of the community report, the community report summary, and the title of the community report.\n\n### Community Tables Emission\n\nAt this point, some bookkeeping work is performed and we emit the **Communities** and **CommunityReports** tables.\n\nPhase 5: Document Processing\n----------------------------\n\nIn this phase of the workflow, we create the _Documents_ table for the knowledge model.\n\n\\---\ntitle: Document Processing\n---\nflowchart LR\n aug\\[Augment\\] --> dp\\[Link to TextUnits\\] --> de\\[Avg. Embedding\\] --> dg\\[Document Table Emission\\]\n\n### Augment with Columns (CSV Only)\n\nIf the workflow is operating on CSV data, you may configure your workflow to add additional fields to Documents output. These fields should exist on the incoming CSV tables. Details about configuring this can be found in the [configuration documentation](/graphrag/posts/config/overview/)\n.\n\n### Link to TextUnits\n\nIn this step, we link each document to the text-units that were created in the first phase. This allows us to understand which documents are related to which text-units and vice-versa.\n\n### Document Embedding\n\nIn this step, we generate a vector representation of our documents using an average embedding of document slices. We re-chunk documents without overlapping chunks, and then generate an embedding for each chunk. We create an average of these chunks weighted by token-count and use this as the document embedding. This will allow us to understand the implicit relationship between documents, and will help us generate a network representation of our documents.\n\n### Documents Table Emission\n\nAt this point, we can emit the **Documents** table into the knowledge Model.\n\nPhase 6: Network Visualization\n------------------------------\n\nIn this phase of the workflow, we perform some steps to support network visualization of our high-dimensional vector spaces within our existing graphs. At this point there are two logical graphs at play: the _Entity-Relationship_ graph and the _Document_ graph.\n\n\\---\ntitle: Network Visualization Workflows\n---\nflowchart LR\n nv\\[Umap Documents\\] --> ne\\[Umap Entities\\] --> ng\\[Nodes Table Emission\\]\n\nFor each of the logical graphs, we perform a UMAP dimensionality reduction to generate a 2D representation of the graph. This will allow us to visualize the graph in a 2D space and understand the relationships between the nodes in the graph. The UMAP embeddings are then emitted as a table of _Nodes_. The rows of this table include a discriminator indicating whether the node is a document or an entity, and the UMAP coordinates.", - "metadata": { - "title": "Indexing Dataflow", - "sourceURL": "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Indexer CLI\n===========\n\nThe GraphRAG indexer CLI allows for no-code usage of the GraphRAG Indexer.\n\n python -m graphrag.index --verbose --root --config \n --resume --reporter --emit json,csv,parquet\n --nocache\n\nCLI Arguments\n-------------\n\n* `--verbose` - Adds extra logging information during the run.\n* `--root ` - the data root directory. This should contain an `input` directory with the input data, and an `.env` file with environment variables. These are described below.\n* `--init` - This will initialize the data project directory at the specified `root` with bootstrap configuration and prompt-overrides.\n* `--resume ` - if specified, the pipeline will attempt to resume a prior run. The parquet files from the prior run will be loaded into the system as inputs, and the workflows that generated those files will be skipped. The input value should be the timestamped output folder, e.g. \"20240105-143721\".\n* `--config ` - This will opt-out of the Default Configuration mode and execute a custom configuration. If this is used, then none of the environment-variables below will apply.\n* `--reporter ` - This will specify the progress reporter to use. The default is `rich`. Valid values are `rich`, `print`, and `none`.\n* `--emit ` - This specifies the table output formats the pipeline should emit. The default is `parquet`. Valid values are `parquet`, `csv`, and `json`, comma-separated.\n* `--nocache` - This will disable the caching mechanism. This is useful for debugging and development, but should not be used in production.", - "markdown": "Indexer CLI\n===========\n\nThe GraphRAG indexer CLI allows for no-code usage of the GraphRAG Indexer.\n\n python -m graphrag.index --verbose --root --config \n --resume --reporter --emit json,csv,parquet\n --nocache\n\nCLI Arguments\n-------------\n\n* `--verbose` - Adds extra logging information during the run.\n* `--root ` - the data root directory. This should contain an `input` directory with the input data, and an `.env` file with environment variables. These are described below.\n* `--init` - This will initialize the data project directory at the specified `root` with bootstrap configuration and prompt-overrides.\n* `--resume ` - if specified, the pipeline will attempt to resume a prior run. The parquet files from the prior run will be loaded into the system as inputs, and the workflows that generated those files will be skipped. The input value should be the timestamped output folder, e.g. \"20240105-143721\".\n* `--config ` - This will opt-out of the Default Configuration mode and execute a custom configuration. If this is used, then none of the environment-variables below will apply.\n* `--reporter ` - This will specify the progress reporter to use. The default is `rich`. Valid values are `rich`, `print`, and `none`.\n* `--emit ` - This specifies the table output formats the pipeline should emit. The default is `parquet`. Valid values are `parquet`, `csv`, and `json`, comma-separated.\n* `--nocache` - This will disable the caching mechanism. This is useful for debugging and development, but should not be used in production.", - "metadata": { - "title": "Indexer CLI", - "sourceURL": "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Configuring GraphRAG Indexing\n=============================\n\nThe GraphRAG system is highly configurable. This page provides an overview of the configuration options available for the GraphRAG indexing engine.\n\nDefault Configuration Mode\n--------------------------\n\nThe default configuration mode is the simplest way to get started with the GraphRAG system. It is designed to work out-of-the-box with minimal configuration. The primary configuration sections for the Indexing Engine pipelines are described below. The main ways to set up GraphRAG in Default Configuration mode are via:\n\n* [Init command](/graphrag/posts/config/init)\n (recommended)\n* [Purely using environment variables](/graphrag/posts/config/env_vars)\n \n* [Using JSON or YAML for deeper control](/graphrag/posts/config/json_yaml)\n \n\nCustom Configuration Mode\n-------------------------\n\nCustom configuration mode is an advanced use-case. Most users will want to use the Default Configuration instead. The primary configuration sections for Indexing Engine pipelines are described below. Details about how to use custom configuration are available in the [Custom Configuration Mode](/graphrag/posts/config/custom)\n documentation.", - "markdown": "Configuring GraphRAG Indexing\n=============================\n\nThe GraphRAG system is highly configurable. This page provides an overview of the configuration options available for the GraphRAG indexing engine.\n\nDefault Configuration Mode\n--------------------------\n\nThe default configuration mode is the simplest way to get started with the GraphRAG system. It is designed to work out-of-the-box with minimal configuration. The primary configuration sections for the Indexing Engine pipelines are described below. The main ways to set up GraphRAG in Default Configuration mode are via:\n\n* [Init command](/graphrag/posts/config/init)\n (recommended)\n* [Purely using environment variables](/graphrag/posts/config/env_vars)\n \n* [Using JSON or YAML for deeper control](/graphrag/posts/config/json_yaml)\n \n\nCustom Configuration Mode\n-------------------------\n\nCustom configuration mode is an advanced use-case. Most users will want to use the Default Configuration instead. The primary configuration sections for Indexing Engine pipelines are described below. Details about how to use custom configuration are available in the [Custom Configuration Mode](/graphrag/posts/config/custom)\n documentation.", - "metadata": { - "title": "Configuring GraphRAG Indexing", - "sourceURL": "https://microsoft.github.io/graphrag/posts/config/overview/", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Configuring GraphRAG Indexing\n=============================\n\nTo start using GraphRAG, you need to configure the system. The `init` command is the easiest way to get started. It will create a `.env` and `settings.yaml` files in the specified directory with the necessary configuration settings. It will also output the default LLM prompts used by GraphRAG.\n\nUsage\n-----\n\n python -m graphrag.index [--init] [--root PATH]\n\nOptions\n-------\n\n* `--init` - Initialize the directory with the necessary configuration files.\n* `--root PATH` - The root directory to initialize. Default is the current directory.\n\nExample\n-------\n\n python -m graphrag.index --init --root ./ragtest\n\nOutput\n------\n\nThe `init` command will create the following files in the specified directory:\n\n* `settings.yaml` - The configuration settings file. This file contains the configuration settings for GraphRAG.\n* `.env` - The environment variables file. These are referenced in the `settings.yaml` file.\n* `prompts/` - The LLM prompts folder. This contains the default prompts used by GraphRAG, you can modify them or run the [Auto Prompt Tuning](/graphrag/posts/prompt_tuning/auto_prompt_tuning)\n command to generate new prompts adapted to your data.\n\nNext Steps\n----------\n\nAfter initializing your workspace, you can either run the [Prompt Tuning](/graphrag/posts/prompt_tuning/auto_prompt_tuning)\n command to adapt the prompts to your data or even start running the [Indexing Pipeline](/graphrag/posts/index/overview)\n to index your data. For more information on configuring GraphRAG, see the [Configuration](/graphrag/posts/config/overview)\n documentation.", - "markdown": "Configuring GraphRAG Indexing\n=============================\n\nTo start using GraphRAG, you need to configure the system. The `init` command is the easiest way to get started. It will create a `.env` and `settings.yaml` files in the specified directory with the necessary configuration settings. It will also output the default LLM prompts used by GraphRAG.\n\nUsage\n-----\n\n python -m graphrag.index [--init] [--root PATH]\n\nOptions\n-------\n\n* `--init` - Initialize the directory with the necessary configuration files.\n* `--root PATH` - The root directory to initialize. Default is the current directory.\n\nExample\n-------\n\n python -m graphrag.index --init --root ./ragtest\n\nOutput\n------\n\nThe `init` command will create the following files in the specified directory:\n\n* `settings.yaml` - The configuration settings file. This file contains the configuration settings for GraphRAG.\n* `.env` - The environment variables file. These are referenced in the `settings.yaml` file.\n* `prompts/` - The LLM prompts folder. This contains the default prompts used by GraphRAG, you can modify them or run the [Auto Prompt Tuning](/graphrag/posts/prompt_tuning/auto_prompt_tuning)\n command to generate new prompts adapted to your data.\n\nNext Steps\n----------\n\nAfter initializing your workspace, you can either run the [Prompt Tuning](/graphrag/posts/prompt_tuning/auto_prompt_tuning)\n command to adapt the prompts to your data or even start running the [Indexing Pipeline](/graphrag/posts/index/overview)\n to index your data. For more information on configuring GraphRAG, see the [Configuration](/graphrag/posts/config/overview)\n documentation.", - "metadata": { - "title": "Configuring GraphRAG Indexing", - "sourceURL": "https://microsoft.github.io/graphrag/posts/config/init", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning", - "https://microsoft.github.io/graphrag/posts/index/overview", - "https://microsoft.github.io/graphrag/posts/config/overview", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Default Configuration Mode (using Env Vars)\n===========================================\n\nText-Embeddings Customization\n-----------------------------\n\nBy default, the GraphRAG indexer will only emit embeddings required for our query methods. However, the model has embeddings defined for all plaintext fields, and these can be generated by setting the `GRAPHRAG_EMBEDDING_TARGET` environment variable to `all`.\n\nIf the embedding target is `all`, and you want to only embed a subset of these fields, you may specify which embeddings to skip using the `GRAPHRAG_EMBEDDING_SKIP` argument described below.\n\n### Embedded Fields\n\n* `text_unit.text`\n* `document.raw_content`\n* `entity.name`\n* `entity.description`\n* `relationship.description`\n* `community.title`\n* `community.summary`\n* `community.full_content`\n\nInput Data\n----------\n\nOur pipeline can ingest .csv or .txt data from an input folder. These files can be nested within subfolders. To configure how input data is handled, what fields are mapped over, and how timestamps are parsed, look for configuration values starting with `GRAPHRAG_INPUT_` below. In general, CSV-based data provides the most customizeability. Each CSV should at least contain a `text` field (which can be mapped with environment variables), but it's helpful if they also have `title`, `timestamp`, and `source` fields. Additional fields can be included as well, which will land as extra fields on the `Document` table.\n\nBase LLM Settings\n-----------------\n\nThese are the primary settings for configuring LLM connectivity.\n\n| Parameter | Required? | Description | Type | Default Value |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_API_KEY` | **Yes for OpenAI. Optional for AOAI** | The API key. (Note: \\`OPENAI\\_API\\_KEY is also used as a fallback). If not defined when using AOAI, managed identity will be used. | `str` | `None` |\n| `GRAPHRAG_API_BASE` | **For AOAI** | The API Base URL | `str` | `None` |\n| `GRAPHRAG_API_VERSION` | **For AOAI** | The AOAI API version. | `str` | `None` |\n| `GRAPHRAG_API_ORGANIZATION` | | The AOAI organization. | `str` | `None` |\n| `GRAPHRAG_API_PROXY` | | The AOAI proxy. | `str` | `None` |\n\nText Generation Settings\n------------------------\n\nThese settings control the text generation model used by the pipeline. Any settings with a fallback will use the base LLM settings, if available.\n\n| Parameter | Required? | Description | Type | Default Value |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_LLM_TYPE` | **For AOAI** | The LLM operation type. Either `openai_chat` or `azure_openai_chat` | `str` | `openai_chat` |\n| `GRAPHRAG_LLM_DEPLOYMENT_NAME` | **For AOAI** | The AOAI model deployment name. | `str` | `None` |\n| `GRAPHRAG_LLM_API_KEY` | Yes (uses fallback) | The API key. If not defined when using AOAI, managed identity will be used. | `str` | `None` |\n| `GRAPHRAG_LLM_API_BASE` | For AOAI (uses fallback) | The API Base URL | `str` | `None` |\n| `GRAPHRAG_LLM_API_VERSION` | For AOAI (uses fallback) | The AOAI API version. | `str` | `None` |\n| `GRAPHRAG_LLM_API_ORGANIZATION` | For AOAI (uses fallback) | The AOAI organization. | `str` | `None` |\n| `GRAPHRAG_LLM_API_PROXY` | | The AOAI proxy. | `str` | `None` |\n| `GRAPHRAG_LLM_MODEL` | | The LLM model. | `str` | `gpt-4-turbo-preview` |\n| `GRAPHRAG_LLM_MAX_TOKENS` | | The maximum number of tokens. | `int` | `4000` |\n| `GRAPHRAG_LLM_REQUEST_TIMEOUT` | | The maximum number of seconds to wait for a response from the chat client. | `int` | `180` |\n| `GRAPHRAG_LLM_MODEL_SUPPORTS_JSON` | | Indicates whether the given model supports JSON output mode. `True` to enable. | `str` | `None` |\n| `GRAPHRAG_LLM_THREAD_COUNT` | | The number of threads to use for LLM parallelization. | `int` | 50 |\n| `GRAPHRAG_LLM_THREAD_STAGGER` | | The time to wait (in seconds) between starting each thread. | `float` | 0.3 |\n| `GRAPHRAG_LLM_CONCURRENT_REQUESTS` | | The number of concurrent requests to allow for the embedding client. | `int` | 25 |\n| `GRAPHRAG_LLM_TOKENS_PER_MINUTE` | | The number of tokens per minute to allow for the LLM client. 0 = Bypass | `int` | 0 |\n| `GRAPHRAG_LLM_REQUESTS_PER_MINUTE` | | The number of requests per minute to allow for the LLM client. 0 = Bypass | `int` | 0 |\n| `GRAPHRAG_LLM_MAX_RETRIES` | | The maximum number of retries to attempt when a request fails. | `int` | 10 |\n| `GRAPHRAG_LLM_MAX_RETRY_WAIT` | | The maximum number of seconds to wait between retries. | `int` | 10 |\n| `GRAPHRAG_LLM_SLEEP_ON_RATE_LIMIT_RECOMMENDATION` | | Whether to sleep on rate limit recommendation. (Azure Only) | `bool` | `True` |\n| `GRAPHRAG_LLM_TEMPERATURE` | | The temperature to use generation. | `float` | 0 |\n| `GRAPHRAG_LLM_TOP_P` | | The top\\_p to use for sampling. | `float` | 1 |\n| `GRAPHRAG_LLM_N` | | The number of responses to generate. | `int` | 1 |\n\nText Embedding Settings\n-----------------------\n\nThese settings control the text embedding model used by the pipeline. Any settings with a fallback will use the base LLM settings, if available.\n\n| Parameter | Required ? | Description | Type | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_EMBEDDING_TYPE` | **For AOAI** | The embedding client to use. Either `openai_embedding` or `azure_openai_embedding` | `str` | `openai_embedding` |\n| `GRAPHRAG_EMBEDDING_DEPLOYMENT_NAME` | **For AOAI** | The AOAI deployment name. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_API_KEY` | Yes (uses fallback) | The API key to use for the embedding client. If not defined when using AOAI, managed identity will be used. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_API_BASE` | For AOAI (uses fallback) | The API base URL. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_API_VERSION` | For AOAI (uses fallback) | The AOAI API version to use for the embedding client. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_API_ORGANIZATION` | For AOAI (uses fallback) | The AOAI organization to use for the embedding client. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_API_PROXY` | | The AOAI proxy to use for the embedding client. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_MODEL` | | The model to use for the embedding client. | `str` | `text-embedding-3-small` |\n| `GRAPHRAG_EMBEDDING_BATCH_SIZE` | | The number of texts to embed at once. [(Azure limit is 16)](https://learn.microsoft.com/en-us/azure/ai-ce) | `int` | 16 |\n| `GRAPHRAG_EMBEDDING_BATCH_MAX_TOKENS` | | The maximum tokens per batch [(Azure limit is 8191)](https://learn.microsoft.com/en-us/azure/ai-services/openai/reference) | `int` | 8191 |\n| `GRAPHRAG_EMBEDDING_TARGET` | | The target fields to embed. Either `required` or `all`. | `str` | `required` |\n| `GRAPHRAG_EMBEDDING_SKIP` | | A comma-separated list of fields to skip embeddings for . (e.g. 'relationship.description') | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_THREAD_COUNT` | | The number of threads to use for parallelization for embeddings. | `int` | |\n| `GRAPHRAG_EMBEDDING_THREAD_STAGGER` | | The time to wait (in seconds) between starting each thread for embeddings. | `float` | 50 |\n| `GRAPHRAG_EMBEDDING_CONCURRENT_REQUESTS` | | The number of concurrent requests to allow for the embedding client. | `int` | 25 |\n| `GRAPHRAG_EMBEDDING_TOKENS_PER_MINUTE` | | The number of tokens per minute to allow for the embedding client. 0 = Bypass | `int` | 0 |\n| `GRAPHRAG_EMBEDDING_REQUESTS_PER_MINUTE` | | The number of requests per minute to allow for the embedding client. 0 = Bypass | `int` | 0 |\n| `GRAPHRAG_EMBEDDING_MAX_RETRIES` | | The maximum number of retries to attempt when a request fails. | `int` | 10 |\n| `GRAPHRAG_EMBEDDING_MAX_RETRY_WAIT` | | The maximum number of seconds to wait between retries. | `int` | 10 |\n| `GRAPHRAG_EMBEDDING_TARGET` | | The target fields to embed. Either `required` or `all`. | `str` | `required` |\n| `GRAPHRAG_EMBEDDING_SLEEP_ON_RATE_LIMIT_RECOMMENDATION` | | Whether to sleep on rate limit recommendation. (Azure Only) | `bool` | `True` |\n\nInput Settings\n--------------\n\nThese settings control the data input used by the pipeline. Any settings with a fallback will use the base LLM settings, if available.\n\n### Plaintext Input Data (`GRAPHRAG_INPUT_FILE_TYPE`\\=text)\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_INPUT_FILE_PATTERN` | The file pattern regexp to use when reading input files from the input directory. | `str` | optional | `.*\\.txt$` |\n\n### CSV Input Data (`GRAPHRAG_INPUT_FILE_TYPE`\\=csv)\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_INPUT_TYPE` | The input storage type to use when reading files. (`file` or `blob`) | `str` | optional | `file` |\n| `GRAPHRAG_INPUT_FILE_PATTERN` | The file pattern regexp to use when reading input files from the input directory. | `str` | optional | `.*\\.txt$` |\n| `GRAPHRAG_INPUT_SOURCE_COLUMN` | The 'source' column to use when reading CSV input files. | `str` | optional | `source` |\n| `GRAPHRAG_INPUT_TIMESTAMP_COLUMN` | The 'timestamp' column to use when reading CSV input files. | `str` | optional | `None` |\n| `GRAPHRAG_INPUT_TIMESTAMP_FORMAT` | The timestamp format to use when parsing timestamps in the timestamp column. | `str` | optional | `None` |\n| `GRAPHRAG_INPUT_TEXT_COLUMN` | The 'text' column to use when reading CSV input files. | `str` | optional | `text` |\n| `GRAPHRAG_INPUT_DOCUMENT_ATTRIBUTE_COLUMNS` | A list of CSV columns, comma-separated, to incorporate as document fields. | `str` | optional | `id` |\n| `GRAPHRAG_INPUT_TITLE_COLUMN` | The 'title' column to use when reading CSV input files. | `str` | optional | `title` |\n| `GRAPHRAG_INPUT_STORAGE_ACCOUNT_BLOB_URL` | The Azure Storage blob endpoint to use when in `blob` mode and using managed identity. Will have the format `https://.blob.core.windows.net` | `str` | optional | `None` |\n| `GRAPHRAG_INPUT_CONNECTION_STRING` | The connection string to use when reading CSV input files from Azure Blob Storage. | `str` | optional | `None` |\n| `GRAPHRAG_INPUT_CONTAINER_NAME` | The container name to use when reading CSV input files from Azure Blob Storage. | `str` | optional | `None` |\n| `GRAPHRAG_INPUT_BASE_DIR` | The base directory to read input files from. | `str` | optional | `None` |\n\nData Mapping Settings\n---------------------\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_INPUT_FILE_TYPE` | The type of input data, `csv` or `text` | `str` | optional | `text` |\n| `GRAPHRAG_INPUT_ENCODING` | The encoding to apply when reading CSV/text input files. | `str` | optional | `utf-8` |\n\nData Chunking\n-------------\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_CHUNK_SIZE` | The chunk size in tokens for text-chunk analysis windows. | `str` | optional | 1200 |\n| `GRAPHRAG_CHUNK_OVERLAP` | The chunk overlap in tokens for text-chunk analysis windows. | `str` | optional | 100 |\n| `GRAPHRAG_CHUNK_BY_COLUMNS` | A comma-separated list of document attributes to groupby when performing TextUnit chunking. | `str` | optional | `id` |\n| `GRAPHRAG_CHUNK_ENCODING_MODEL` | The encoding model to use for chunking. | `str` | optional | The top-level encoding model. |\n\nPrompting Overrides\n-------------------\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_ENTITY_EXTRACTION_PROMPT_FILE` | The path (relative to the root) of an entity extraction prompt template text file. | `str` | optional | `None` |\n| `GRAPHRAG_ENTITY_EXTRACTION_MAX_GLEANINGS` | The maximum number of redrives (gleanings) to invoke when extracting entities in a loop. | `int` | optional | 1 |\n| `GRAPHRAG_ENTITY_EXTRACTION_ENTITY_TYPES` | A comma-separated list of entity types to extract. | `str` | optional | `organization,person,event,geo` |\n| `GRAPHRAG_ENTITY_EXTRACTION_ENCODING_MODEL` | The encoding model to use for entity extraction. | `str` | optional | The top-level encoding model. |\n| `GRAPHRAG_SUMMARIZE_DESCRIPTIONS_PROMPT_FILE` | The path (relative to the root) of an description summarization prompt template text file. | `str` | optional | `None` |\n| `GRAPHRAG_SUMMARIZE_DESCRIPTIONS_MAX_LENGTH` | The maximum number of tokens to generate per description summarization. | `int` | optional | 500 |\n| `GRAPHRAG_CLAIM_EXTRACTION_ENABLED` | Whether claim extraction is enabled for this pipeline. | `bool` | optional | `False` |\n| `GRAPHRAG_CLAIM_EXTRACTION_DESCRIPTION` | The claim\\_description prompting argument to utilize. | `string` | optional | \"Any claims or facts that could be relevant to threat analysis.\" |\n| `GRAPHRAG_CLAIM_EXTRACTION_PROMPT_FILE` | The claim extraction prompt to utilize. | `string` | optional | `None` |\n| `GRAPHRAG_CLAIM_EXTRACTION_MAX_GLEANINGS` | The maximum number of redrives (gleanings) to invoke when extracting claims in a loop. | `int` | optional | 1 |\n| `GRAPHRAG_CLAIM_EXTRACTION_ENCODING_MODEL` | The encoding model to use for claim extraction. | `str` | optional | The top-level encoding model |\n| `GRAPHRAG_COMMUNITY_REPORTS_PROMPT_FILE` | The community reports extraction prompt to utilize. | `string` | optional | `None` |\n| `GRAPHRAG_COMMUNITY_REPORTS_MAX_LENGTH` | The maximum number of tokens to generate per community reports. | `int` | optional | 1500 |\n\nStorage\n-------\n\nThis section controls the storage mechanism used by the pipeline used for emitting output tables.\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_STORAGE_TYPE` | The type of reporter to use. Options are `file`, `memory`, or `blob` | `str` | optional | `file` |\n| `GRAPHRAG_STORAGE_STORAGE_ACCOUNT_BLOB_URL` | The Azure Storage blob endpoint to use when in `blob` mode and using managed identity. Will have the format `https://.blob.core.windows.net` | `str` | optional | None |\n| `GRAPHRAG_STORAGE_CONNECTION_STRING` | The Azure Storage connection string to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_STORAGE_CONTAINER_NAME` | The Azure Storage container name to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_STORAGE_BASE_DIR` | The base path to data outputs outputs. | `str` | optional | None |\n\nCache\n-----\n\nThis section controls the cache mechanism used by the pipeline. This is used to cache LLM invocation results.\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_CACHE_TYPE` | The type of cache to use. Options are `file`, `memory`, `none` or `blob` | `str` | optional | `file` |\n| `GRAPHRAG_CACHE_STORAGE_ACCOUNT_BLOB_URL` | The Azure Storage blob endpoint to use when in `blob` mode and using managed identity. Will have the format `https://.blob.core.windows.net` | `str` | optional | None |\n| `GRAPHRAG_CACHE_CONNECTION_STRING` | The Azure Storage connection string to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_CACHE_CONTAINER_NAME` | The Azure Storage container name to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_CACHE_BASE_DIR` | The base path to the reporting outputs. | `str` | optional | None |\n\nReporting\n---------\n\nThis section controls the reporting mechanism used by the pipeline, for common events and error messages. The default is to write reports to a file in the output directory. However, you can also choose to write reports to the console or to an Azure Blob Storage container.\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_REPORTING_TYPE` | The type of reporter to use. Options are `file`, `console`, or `blob` | `str` | optional | `file` |\n| `GRAPHRAG_REPORTING_STORAGE_ACCOUNT_BLOB_URL` | The Azure Storage blob endpoint to use when in `blob` mode and using managed identity. Will have the format `https://.blob.core.windows.net` | `str` | optional | None |\n| `GRAPHRAG_REPORTING_CONNECTION_STRING` | The Azure Storage connection string to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_REPORTING_CONTAINER_NAME` | The Azure Storage container name to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_REPORTING_BASE_DIR` | The base path to the reporting outputs. | `str` | optional | None |\n\nNode2Vec Parameters\n-------------------\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_NODE2VEC_ENABLED` | Whether to enable Node2Vec | `bool` | optional | False |\n| `GRAPHRAG_NODE2VEC_NUM_WALKS` | The Node2Vec number of walks to perform | `int` | optional | 10 |\n| `GRAPHRAG_NODE2VEC_WALK_LENGTH` | The Node2Vec walk length | `int` | optional | 40 |\n| `GRAPHRAG_NODE2VEC_WINDOW_SIZE` | The Node2Vec window size | `int` | optional | 2 |\n| `GRAPHRAG_NODE2VEC_ITERATIONS` | The number of iterations to run node2vec | `int` | optional | 3 |\n| `GRAPHRAG_NODE2VEC_RANDOM_SEED` | The random seed to use for node2vec | `int` | optional | 597832 |\n\nData Snapshotting\n-----------------\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_SNAPSHOT_GRAPHML` | Whether to enable GraphML snapshots. | `bool` | optional | False |\n| `GRAPHRAG_SNAPSHOT_RAW_ENTITIES` | Whether to enable raw entity snapshots. | `bool` | optional | False |\n| `GRAPHRAG_SNAPSHOT_TOP_LEVEL_NODES` | Whether to enable top-level node snapshots. | `bool` | optional | False |\n\nMiscellaneous Settings\n======================\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_ASYNC_MODE` | Which async mode to use. Either `asyncio` or `threaded`. | `str` | optional | `asyncio` |\n| `GRAPHRAG_ENCODING_MODEL` | The text encoding model, used in tiktoken, to encode text. | `str` | optional | `cl100k_base` |\n| `GRAPHRAG_MAX_CLUSTER_SIZE` | The maximum number of entities to include in a single Leiden cluster. | `int` | optional | 10 |\n| `GRAPHRAG_SKIP_WORKFLOWS` | A comma-separated list of workflow names to skip. | `str` | optional | `None` |\n| `GRAPHRAG_UMAP_ENABLED` | Whether to enable UMAP layouts | `bool` | optional | False |", - "markdown": "Default Configuration Mode (using Env Vars)\n===========================================\n\nText-Embeddings Customization\n-----------------------------\n\nBy default, the GraphRAG indexer will only emit embeddings required for our query methods. However, the model has embeddings defined for all plaintext fields, and these can be generated by setting the `GRAPHRAG_EMBEDDING_TARGET` environment variable to `all`.\n\nIf the embedding target is `all`, and you want to only embed a subset of these fields, you may specify which embeddings to skip using the `GRAPHRAG_EMBEDDING_SKIP` argument described below.\n\n### Embedded Fields\n\n* `text_unit.text`\n* `document.raw_content`\n* `entity.name`\n* `entity.description`\n* `relationship.description`\n* `community.title`\n* `community.summary`\n* `community.full_content`\n\nInput Data\n----------\n\nOur pipeline can ingest .csv or .txt data from an input folder. These files can be nested within subfolders. To configure how input data is handled, what fields are mapped over, and how timestamps are parsed, look for configuration values starting with `GRAPHRAG_INPUT_` below. In general, CSV-based data provides the most customizeability. Each CSV should at least contain a `text` field (which can be mapped with environment variables), but it's helpful if they also have `title`, `timestamp`, and `source` fields. Additional fields can be included as well, which will land as extra fields on the `Document` table.\n\nBase LLM Settings\n-----------------\n\nThese are the primary settings for configuring LLM connectivity.\n\n| Parameter | Required? | Description | Type | Default Value |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_API_KEY` | **Yes for OpenAI. Optional for AOAI** | The API key. (Note: \\`OPENAI\\_API\\_KEY is also used as a fallback). If not defined when using AOAI, managed identity will be used. | `str` | `None` |\n| `GRAPHRAG_API_BASE` | **For AOAI** | The API Base URL | `str` | `None` |\n| `GRAPHRAG_API_VERSION` | **For AOAI** | The AOAI API version. | `str` | `None` |\n| `GRAPHRAG_API_ORGANIZATION` | | The AOAI organization. | `str` | `None` |\n| `GRAPHRAG_API_PROXY` | | The AOAI proxy. | `str` | `None` |\n\nText Generation Settings\n------------------------\n\nThese settings control the text generation model used by the pipeline. Any settings with a fallback will use the base LLM settings, if available.\n\n| Parameter | Required? | Description | Type | Default Value |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_LLM_TYPE` | **For AOAI** | The LLM operation type. Either `openai_chat` or `azure_openai_chat` | `str` | `openai_chat` |\n| `GRAPHRAG_LLM_DEPLOYMENT_NAME` | **For AOAI** | The AOAI model deployment name. | `str` | `None` |\n| `GRAPHRAG_LLM_API_KEY` | Yes (uses fallback) | The API key. If not defined when using AOAI, managed identity will be used. | `str` | `None` |\n| `GRAPHRAG_LLM_API_BASE` | For AOAI (uses fallback) | The API Base URL | `str` | `None` |\n| `GRAPHRAG_LLM_API_VERSION` | For AOAI (uses fallback) | The AOAI API version. | `str` | `None` |\n| `GRAPHRAG_LLM_API_ORGANIZATION` | For AOAI (uses fallback) | The AOAI organization. | `str` | `None` |\n| `GRAPHRAG_LLM_API_PROXY` | | The AOAI proxy. | `str` | `None` |\n| `GRAPHRAG_LLM_MODEL` | | The LLM model. | `str` | `gpt-4-turbo-preview` |\n| `GRAPHRAG_LLM_MAX_TOKENS` | | The maximum number of tokens. | `int` | `4000` |\n| `GRAPHRAG_LLM_REQUEST_TIMEOUT` | | The maximum number of seconds to wait for a response from the chat client. | `int` | `180` |\n| `GRAPHRAG_LLM_MODEL_SUPPORTS_JSON` | | Indicates whether the given model supports JSON output mode. `True` to enable. | `str` | `None` |\n| `GRAPHRAG_LLM_THREAD_COUNT` | | The number of threads to use for LLM parallelization. | `int` | 50 |\n| `GRAPHRAG_LLM_THREAD_STAGGER` | | The time to wait (in seconds) between starting each thread. | `float` | 0.3 |\n| `GRAPHRAG_LLM_CONCURRENT_REQUESTS` | | The number of concurrent requests to allow for the embedding client. | `int` | 25 |\n| `GRAPHRAG_LLM_TOKENS_PER_MINUTE` | | The number of tokens per minute to allow for the LLM client. 0 = Bypass | `int` | 0 |\n| `GRAPHRAG_LLM_REQUESTS_PER_MINUTE` | | The number of requests per minute to allow for the LLM client. 0 = Bypass | `int` | 0 |\n| `GRAPHRAG_LLM_MAX_RETRIES` | | The maximum number of retries to attempt when a request fails. | `int` | 10 |\n| `GRAPHRAG_LLM_MAX_RETRY_WAIT` | | The maximum number of seconds to wait between retries. | `int` | 10 |\n| `GRAPHRAG_LLM_SLEEP_ON_RATE_LIMIT_RECOMMENDATION` | | Whether to sleep on rate limit recommendation. (Azure Only) | `bool` | `True` |\n| `GRAPHRAG_LLM_TEMPERATURE` | | The temperature to use generation. | `float` | 0 |\n| `GRAPHRAG_LLM_TOP_P` | | The top\\_p to use for sampling. | `float` | 1 |\n| `GRAPHRAG_LLM_N` | | The number of responses to generate. | `int` | 1 |\n\nText Embedding Settings\n-----------------------\n\nThese settings control the text embedding model used by the pipeline. Any settings with a fallback will use the base LLM settings, if available.\n\n| Parameter | Required ? | Description | Type | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_EMBEDDING_TYPE` | **For AOAI** | The embedding client to use. Either `openai_embedding` or `azure_openai_embedding` | `str` | `openai_embedding` |\n| `GRAPHRAG_EMBEDDING_DEPLOYMENT_NAME` | **For AOAI** | The AOAI deployment name. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_API_KEY` | Yes (uses fallback) | The API key to use for the embedding client. If not defined when using AOAI, managed identity will be used. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_API_BASE` | For AOAI (uses fallback) | The API base URL. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_API_VERSION` | For AOAI (uses fallback) | The AOAI API version to use for the embedding client. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_API_ORGANIZATION` | For AOAI (uses fallback) | The AOAI organization to use for the embedding client. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_API_PROXY` | | The AOAI proxy to use for the embedding client. | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_MODEL` | | The model to use for the embedding client. | `str` | `text-embedding-3-small` |\n| `GRAPHRAG_EMBEDDING_BATCH_SIZE` | | The number of texts to embed at once. [(Azure limit is 16)](https://learn.microsoft.com/en-us/azure/ai-ce) | `int` | 16 |\n| `GRAPHRAG_EMBEDDING_BATCH_MAX_TOKENS` | | The maximum tokens per batch [(Azure limit is 8191)](https://learn.microsoft.com/en-us/azure/ai-services/openai/reference) | `int` | 8191 |\n| `GRAPHRAG_EMBEDDING_TARGET` | | The target fields to embed. Either `required` or `all`. | `str` | `required` |\n| `GRAPHRAG_EMBEDDING_SKIP` | | A comma-separated list of fields to skip embeddings for . (e.g. 'relationship.description') | `str` | `None` |\n| `GRAPHRAG_EMBEDDING_THREAD_COUNT` | | The number of threads to use for parallelization for embeddings. | `int` | |\n| `GRAPHRAG_EMBEDDING_THREAD_STAGGER` | | The time to wait (in seconds) between starting each thread for embeddings. | `float` | 50 |\n| `GRAPHRAG_EMBEDDING_CONCURRENT_REQUESTS` | | The number of concurrent requests to allow for the embedding client. | `int` | 25 |\n| `GRAPHRAG_EMBEDDING_TOKENS_PER_MINUTE` | | The number of tokens per minute to allow for the embedding client. 0 = Bypass | `int` | 0 |\n| `GRAPHRAG_EMBEDDING_REQUESTS_PER_MINUTE` | | The number of requests per minute to allow for the embedding client. 0 = Bypass | `int` | 0 |\n| `GRAPHRAG_EMBEDDING_MAX_RETRIES` | | The maximum number of retries to attempt when a request fails. | `int` | 10 |\n| `GRAPHRAG_EMBEDDING_MAX_RETRY_WAIT` | | The maximum number of seconds to wait between retries. | `int` | 10 |\n| `GRAPHRAG_EMBEDDING_TARGET` | | The target fields to embed. Either `required` or `all`. | `str` | `required` |\n| `GRAPHRAG_EMBEDDING_SLEEP_ON_RATE_LIMIT_RECOMMENDATION` | | Whether to sleep on rate limit recommendation. (Azure Only) | `bool` | `True` |\n\nInput Settings\n--------------\n\nThese settings control the data input used by the pipeline. Any settings with a fallback will use the base LLM settings, if available.\n\n### Plaintext Input Data (`GRAPHRAG_INPUT_FILE_TYPE`\\=text)\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_INPUT_FILE_PATTERN` | The file pattern regexp to use when reading input files from the input directory. | `str` | optional | `.*\\.txt$` |\n\n### CSV Input Data (`GRAPHRAG_INPUT_FILE_TYPE`\\=csv)\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_INPUT_TYPE` | The input storage type to use when reading files. (`file` or `blob`) | `str` | optional | `file` |\n| `GRAPHRAG_INPUT_FILE_PATTERN` | The file pattern regexp to use when reading input files from the input directory. | `str` | optional | `.*\\.txt$` |\n| `GRAPHRAG_INPUT_SOURCE_COLUMN` | The 'source' column to use when reading CSV input files. | `str` | optional | `source` |\n| `GRAPHRAG_INPUT_TIMESTAMP_COLUMN` | The 'timestamp' column to use when reading CSV input files. | `str` | optional | `None` |\n| `GRAPHRAG_INPUT_TIMESTAMP_FORMAT` | The timestamp format to use when parsing timestamps in the timestamp column. | `str` | optional | `None` |\n| `GRAPHRAG_INPUT_TEXT_COLUMN` | The 'text' column to use when reading CSV input files. | `str` | optional | `text` |\n| `GRAPHRAG_INPUT_DOCUMENT_ATTRIBUTE_COLUMNS` | A list of CSV columns, comma-separated, to incorporate as document fields. | `str` | optional | `id` |\n| `GRAPHRAG_INPUT_TITLE_COLUMN` | The 'title' column to use when reading CSV input files. | `str` | optional | `title` |\n| `GRAPHRAG_INPUT_STORAGE_ACCOUNT_BLOB_URL` | The Azure Storage blob endpoint to use when in `blob` mode and using managed identity. Will have the format `https://.blob.core.windows.net` | `str` | optional | `None` |\n| `GRAPHRAG_INPUT_CONNECTION_STRING` | The connection string to use when reading CSV input files from Azure Blob Storage. | `str` | optional | `None` |\n| `GRAPHRAG_INPUT_CONTAINER_NAME` | The container name to use when reading CSV input files from Azure Blob Storage. | `str` | optional | `None` |\n| `GRAPHRAG_INPUT_BASE_DIR` | The base directory to read input files from. | `str` | optional | `None` |\n\nData Mapping Settings\n---------------------\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_INPUT_FILE_TYPE` | The type of input data, `csv` or `text` | `str` | optional | `text` |\n| `GRAPHRAG_INPUT_ENCODING` | The encoding to apply when reading CSV/text input files. | `str` | optional | `utf-8` |\n\nData Chunking\n-------------\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_CHUNK_SIZE` | The chunk size in tokens for text-chunk analysis windows. | `str` | optional | 1200 |\n| `GRAPHRAG_CHUNK_OVERLAP` | The chunk overlap in tokens for text-chunk analysis windows. | `str` | optional | 100 |\n| `GRAPHRAG_CHUNK_BY_COLUMNS` | A comma-separated list of document attributes to groupby when performing TextUnit chunking. | `str` | optional | `id` |\n| `GRAPHRAG_CHUNK_ENCODING_MODEL` | The encoding model to use for chunking. | `str` | optional | The top-level encoding model. |\n\nPrompting Overrides\n-------------------\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_ENTITY_EXTRACTION_PROMPT_FILE` | The path (relative to the root) of an entity extraction prompt template text file. | `str` | optional | `None` |\n| `GRAPHRAG_ENTITY_EXTRACTION_MAX_GLEANINGS` | The maximum number of redrives (gleanings) to invoke when extracting entities in a loop. | `int` | optional | 1 |\n| `GRAPHRAG_ENTITY_EXTRACTION_ENTITY_TYPES` | A comma-separated list of entity types to extract. | `str` | optional | `organization,person,event,geo` |\n| `GRAPHRAG_ENTITY_EXTRACTION_ENCODING_MODEL` | The encoding model to use for entity extraction. | `str` | optional | The top-level encoding model. |\n| `GRAPHRAG_SUMMARIZE_DESCRIPTIONS_PROMPT_FILE` | The path (relative to the root) of an description summarization prompt template text file. | `str` | optional | `None` |\n| `GRAPHRAG_SUMMARIZE_DESCRIPTIONS_MAX_LENGTH` | The maximum number of tokens to generate per description summarization. | `int` | optional | 500 |\n| `GRAPHRAG_CLAIM_EXTRACTION_ENABLED` | Whether claim extraction is enabled for this pipeline. | `bool` | optional | `False` |\n| `GRAPHRAG_CLAIM_EXTRACTION_DESCRIPTION` | The claim\\_description prompting argument to utilize. | `string` | optional | \"Any claims or facts that could be relevant to threat analysis.\" |\n| `GRAPHRAG_CLAIM_EXTRACTION_PROMPT_FILE` | The claim extraction prompt to utilize. | `string` | optional | `None` |\n| `GRAPHRAG_CLAIM_EXTRACTION_MAX_GLEANINGS` | The maximum number of redrives (gleanings) to invoke when extracting claims in a loop. | `int` | optional | 1 |\n| `GRAPHRAG_CLAIM_EXTRACTION_ENCODING_MODEL` | The encoding model to use for claim extraction. | `str` | optional | The top-level encoding model |\n| `GRAPHRAG_COMMUNITY_REPORTS_PROMPT_FILE` | The community reports extraction prompt to utilize. | `string` | optional | `None` |\n| `GRAPHRAG_COMMUNITY_REPORTS_MAX_LENGTH` | The maximum number of tokens to generate per community reports. | `int` | optional | 1500 |\n\nStorage\n-------\n\nThis section controls the storage mechanism used by the pipeline used for emitting output tables.\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_STORAGE_TYPE` | The type of reporter to use. Options are `file`, `memory`, or `blob` | `str` | optional | `file` |\n| `GRAPHRAG_STORAGE_STORAGE_ACCOUNT_BLOB_URL` | The Azure Storage blob endpoint to use when in `blob` mode and using managed identity. Will have the format `https://.blob.core.windows.net` | `str` | optional | None |\n| `GRAPHRAG_STORAGE_CONNECTION_STRING` | The Azure Storage connection string to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_STORAGE_CONTAINER_NAME` | The Azure Storage container name to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_STORAGE_BASE_DIR` | The base path to data outputs outputs. | `str` | optional | None |\n\nCache\n-----\n\nThis section controls the cache mechanism used by the pipeline. This is used to cache LLM invocation results.\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_CACHE_TYPE` | The type of cache to use. Options are `file`, `memory`, `none` or `blob` | `str` | optional | `file` |\n| `GRAPHRAG_CACHE_STORAGE_ACCOUNT_BLOB_URL` | The Azure Storage blob endpoint to use when in `blob` mode and using managed identity. Will have the format `https://.blob.core.windows.net` | `str` | optional | None |\n| `GRAPHRAG_CACHE_CONNECTION_STRING` | The Azure Storage connection string to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_CACHE_CONTAINER_NAME` | The Azure Storage container name to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_CACHE_BASE_DIR` | The base path to the reporting outputs. | `str` | optional | None |\n\nReporting\n---------\n\nThis section controls the reporting mechanism used by the pipeline, for common events and error messages. The default is to write reports to a file in the output directory. However, you can also choose to write reports to the console or to an Azure Blob Storage container.\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_REPORTING_TYPE` | The type of reporter to use. Options are `file`, `console`, or `blob` | `str` | optional | `file` |\n| `GRAPHRAG_REPORTING_STORAGE_ACCOUNT_BLOB_URL` | The Azure Storage blob endpoint to use when in `blob` mode and using managed identity. Will have the format `https://.blob.core.windows.net` | `str` | optional | None |\n| `GRAPHRAG_REPORTING_CONNECTION_STRING` | The Azure Storage connection string to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_REPORTING_CONTAINER_NAME` | The Azure Storage container name to use when in `blob` mode. | `str` | optional | None |\n| `GRAPHRAG_REPORTING_BASE_DIR` | The base path to the reporting outputs. | `str` | optional | None |\n\nNode2Vec Parameters\n-------------------\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_NODE2VEC_ENABLED` | Whether to enable Node2Vec | `bool` | optional | False |\n| `GRAPHRAG_NODE2VEC_NUM_WALKS` | The Node2Vec number of walks to perform | `int` | optional | 10 |\n| `GRAPHRAG_NODE2VEC_WALK_LENGTH` | The Node2Vec walk length | `int` | optional | 40 |\n| `GRAPHRAG_NODE2VEC_WINDOW_SIZE` | The Node2Vec window size | `int` | optional | 2 |\n| `GRAPHRAG_NODE2VEC_ITERATIONS` | The number of iterations to run node2vec | `int` | optional | 3 |\n| `GRAPHRAG_NODE2VEC_RANDOM_SEED` | The random seed to use for node2vec | `int` | optional | 597832 |\n\nData Snapshotting\n-----------------\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_SNAPSHOT_GRAPHML` | Whether to enable GraphML snapshots. | `bool` | optional | False |\n| `GRAPHRAG_SNAPSHOT_RAW_ENTITIES` | Whether to enable raw entity snapshots. | `bool` | optional | False |\n| `GRAPHRAG_SNAPSHOT_TOP_LEVEL_NODES` | Whether to enable top-level node snapshots. | `bool` | optional | False |\n\nMiscellaneous Settings\n======================\n\n| Parameter | Description | Type | Required or Optional | Default |\n| --- | --- | --- | --- | --- |\n| `GRAPHRAG_ASYNC_MODE` | Which async mode to use. Either `asyncio` or `threaded`. | `str` | optional | `asyncio` |\n| `GRAPHRAG_ENCODING_MODEL` | The text encoding model, used in tiktoken, to encode text. | `str` | optional | `cl100k_base` |\n| `GRAPHRAG_MAX_CLUSTER_SIZE` | The maximum number of entities to include in a single Leiden cluster. | `int` | optional | 10 |\n| `GRAPHRAG_SKIP_WORKFLOWS` | A comma-separated list of workflow names to skip. | `str` | optional | `None` |\n| `GRAPHRAG_UMAP_ENABLED` | Whether to enable UMAP layouts | `bool` | optional | False |", - "metadata": { - "title": "Default Configuration Mode (using Env Vars)", - "sourceURL": "https://microsoft.github.io/graphrag/posts/config/env_vars", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://learn.microsoft.com/en-us/azure/ai-ce", - "https://learn.microsoft.com/en-us/azure/ai-services/openai/reference", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Default Configuration Mode (using JSON/YAML)\n============================================\n\nThe default configuration mode may be configured by using a `config.json` or `config.yml` file in the data project root. If a `.env` file is present along with this config file, then it will be loaded, and the environment variables defined therein will be available for token replacements in your configuration document using `${ENV_VAR}` syntax.\n\nFor example:\n\n # .env\n API_KEY=some_api_key\n \n # config.json\n {\n \"llm\": {\n \"api_key\": \"${API_KEY}\"\n }\n }\n \n\nConfig Sections\n===============\n\ninput\n-----\n\n### Fields\n\n* `type` **file|blob** - The input type to use. Default=`file`\n* `file_type` **text|csv** - The type of input data to load. Either `text` or `csv`. Default is `text`\n* `file_encoding` **str** - The encoding of the input file. Default is `utf-8`\n* `file_pattern` **str** - A regex to match input files. Default is `.*\\.csv$` if in csv mode and `.*\\.txt$` if in text mode.\n* `source_column` **str** - (CSV Mode Only) The source column name.\n* `timestamp_column` **str** - (CSV Mode Only) The timestamp column name.\n* `timestamp_format` **str** - (CSV Mode Only) The source format.\n* `text_column` **str** - (CSV Mode Only) The text column name.\n* `title_column` **str** - (CSV Mode Only) The title column name.\n* `document_attribute_columns` **list\\[str\\]** - (CSV Mode Only) The additional document attributes to include.\n* `connection_string` **str** - (blob only) The Azure Storage connection string.\n* `container_name` **str** - (blob only) The Azure Storage container name.\n* `base_dir` **str** - The base directory to read input from, relative to the root.\n* `storage_account_blob_url` **str** - The storage account blob URL to use.\n\nllm\n---\n\nThis is the base LLM configuration section. Other steps may override this configuration with their own LLM configuration.\n\n### Fields\n\n* `api_key` **str** - The OpenAI API key to use.\n* `type` **openai\\_chat|azure\\_openai\\_chat|openai\\_embedding|azure\\_openai\\_embedding** - The type of LLM to use.\n* `model` **str** - The model name.\n* `max_tokens` **int** - The maximum number of output tokens.\n* `request_timeout` **float** - The per-request timeout.\n* `api_base` **str** - The API base url to use.\n* `api_version` **str** - The API version\n* `organization` **str** - The client organization.\n* `proxy` **str** - The proxy URL to use.\n* `cognitive_services_endpoint` **str** - The url endpoint for cognitive services.\n* `deployment_name` **str** - The deployment name to use (Azure).\n* `model_supports_json` **bool** - Whether the model supports JSON-mode output.\n* `tokens_per_minute` **int** - Set a leaky-bucket throttle on tokens-per-minute.\n* `requests_per_minute` **int** - Set a leaky-bucket throttle on requests-per-minute.\n* `max_retries` **int** - The maximum number of retries to use.\n* `max_retry_wait` **float** - The maximum backoff time.\n* `sleep_on_rate_limit_recommendation` **bool** - Whether to adhere to sleep recommendations (Azure).\n* `concurrent_requests` **int** The number of open requests to allow at once.\n* `temperature` **float** - The temperature to use.\n* `top_p` **float** - The top-p value to use.\n* `n` **int** - The number of completions to generate.\n\nparallelization\n---------------\n\n### Fields\n\n* `stagger` **float** - The threading stagger value.\n* `num_threads` **int** - The maximum number of work threads.\n\nasync\\_mode\n-----------\n\n**asyncio|threaded** The async mode to use. Either `asyncio` or \\`threaded.\n\nembeddings\n----------\n\n### Fields\n\n* `llm` (see LLM top-level config)\n* `parallelization` (see Parallelization top-level config)\n* `async_mode` (see Async Mode top-level config)\n* `batch_size` **int** - The maximum batch size to use.\n* `batch_max_tokens` **int** - The maximum batch #-tokens.\n* `target` **required|all** - Determines which set of embeddings to emit.\n* `skip` **list\\[str\\]** - Which embeddings to skip.\n* `strategy` **dict** - Fully override the text-embedding strategy.\n\nchunks\n------\n\n### Fields\n\n* `size` **int** - The max chunk size in tokens.\n* `overlap` **int** - The chunk overlap in tokens.\n* `group_by_columns` **list\\[str\\]** - group documents by fields before chunking.\n* `encoding_model` **str** - The text encoding model to use. Default is to use the top-level encoding model.\n* `strategy` **dict** - Fully override the chunking strategy.\n\ncache\n-----\n\n### Fields\n\n* `type` **file|memory|none|blob** - The cache type to use. Default=`file`\n* `connection_string` **str** - (blob only) The Azure Storage connection string.\n* `container_name` **str** - (blob only) The Azure Storage container name.\n* `base_dir` **str** - The base directory to write cache to, relative to the root.\n* `storage_account_blob_url` **str** - The storage account blob URL to use.\n\nstorage\n-------\n\n### Fields\n\n* `type` **file|memory|blob** - The storage type to use. Default=`file`\n* `connection_string` **str** - (blob only) The Azure Storage connection string.\n* `container_name` **str** - (blob only) The Azure Storage container name.\n* `base_dir` **str** - The base directory to write reports to, relative to the root.\n* `storage_account_blob_url` **str** - The storage account blob URL to use.\n\nreporting\n---------\n\n### Fields\n\n* `type` **file|console|blob** - The reporting type to use. Default=`file`\n* `connection_string` **str** - (blob only) The Azure Storage connection string.\n* `container_name` **str** - (blob only) The Azure Storage container name.\n* `base_dir` **str** - The base directory to write reports to, relative to the root.\n* `storage_account_blob_url` **str** - The storage account blob URL to use.\n\nentity\\_extraction\n------------------\n\n### Fields\n\n* `llm` (see LLM top-level config)\n* `parallelization` (see Parallelization top-level config)\n* `async_mode` (see Async Mode top-level config)\n* `prompt` **str** - The prompt file to use.\n* `entity_types` **list\\[str\\]** - The entity types to identify.\n* `max_gleanings` **int** - The maximum number of gleaning cycles to use.\n* `encoding_model` **str** - The text encoding model to use. By default, this will use the top-level encoding model.\n* `strategy` **dict** - Fully override the entity extraction strategy.\n\nsummarize\\_descriptions\n-----------------------\n\n### Fields\n\n* `llm` (see LLM top-level config)\n* `parallelization` (see Parallelization top-level config)\n* `async_mode` (see Async Mode top-level config)\n* `prompt` **str** - The prompt file to use.\n* `max_length` **int** - The maximum number of output tokens per summarization.\n* `strategy` **dict** - Fully override the summarize description strategy.\n\nclaim\\_extraction\n-----------------\n\n### Fields\n\n* `enabled` **bool** - Whether to enable claim extraction. default=False\n* `llm` (see LLM top-level config)\n* `parallelization` (see Parallelization top-level config)\n* `async_mode` (see Async Mode top-level config)\n* `prompt` **str** - The prompt file to use.\n* `description` **str** - Describes the types of claims we want to extract.\n* `max_gleanings` **int** - The maximum number of gleaning cycles to use.\n* `encoding_model` **str** - The text encoding model to use. By default, this will use the top-level encoding model.\n* `strategy` **dict** - Fully override the claim extraction strategy.\n\ncommunity\\_reports\n------------------\n\n### Fields\n\n* `llm` (see LLM top-level config)\n* `parallelization` (see Parallelization top-level config)\n* `async_mode` (see Async Mode top-level config)\n* `prompt` **str** - The prompt file to use.\n* `max_length` **int** - The maximum number of output tokens per report.\n* `max_input_length` **int** - The maximum number of input tokens to use when generating reports.\n* `strategy` **dict** - Fully override the community reports strategy.\n\ncluster\\_graph\n--------------\n\n### Fields\n\n* `max_cluster_size` **int** - The maximum cluster size to emit.\n* `strategy` **dict** - Fully override the cluster\\_graph strategy.\n\nembed\\_graph\n------------\n\n### Fields\n\n* `enabled` **bool** - Whether to enable graph embeddings.\n* `num_walks` **int** - The node2vec number of walks.\n* `walk_length` **int** - The node2vec walk length.\n* `window_size` **int** - The node2vec window size.\n* `iterations` **int** - The node2vec number of iterations.\n* `random_seed` **int** - The node2vec random seed.\n* `strategy` **dict** - Fully override the embed graph strategy.\n\numap\n----\n\n### Fields\n\n* `enabled` **bool** - Whether to enable UMAP layouts.\n\nsnapshots\n---------\n\n### Fields\n\n* `graphml` **bool** - Emit graphml snapshots.\n* `raw_entities` **bool** - Emit raw entity snapshots.\n* `top_level_nodes` **bool** - Emit top-level-node snapshots.\n\nencoding\\_model\n---------------\n\n**str** - The text encoding model to use. Default is `cl100k_base`.\n\nskip\\_workflows\n---------------\n\n**list\\[str\\]** - Which workflow names to skip.", - "markdown": "Default Configuration Mode (using JSON/YAML)\n============================================\n\nThe default configuration mode may be configured by using a `config.json` or `config.yml` file in the data project root. If a `.env` file is present along with this config file, then it will be loaded, and the environment variables defined therein will be available for token replacements in your configuration document using `${ENV_VAR}` syntax.\n\nFor example:\n\n # .env\n API_KEY=some_api_key\n \n # config.json\n {\n \"llm\": {\n \"api_key\": \"${API_KEY}\"\n }\n }\n \n\nConfig Sections\n===============\n\ninput\n-----\n\n### Fields\n\n* `type` **file|blob** - The input type to use. Default=`file`\n* `file_type` **text|csv** - The type of input data to load. Either `text` or `csv`. Default is `text`\n* `file_encoding` **str** - The encoding of the input file. Default is `utf-8`\n* `file_pattern` **str** - A regex to match input files. Default is `.*\\.csv$` if in csv mode and `.*\\.txt$` if in text mode.\n* `source_column` **str** - (CSV Mode Only) The source column name.\n* `timestamp_column` **str** - (CSV Mode Only) The timestamp column name.\n* `timestamp_format` **str** - (CSV Mode Only) The source format.\n* `text_column` **str** - (CSV Mode Only) The text column name.\n* `title_column` **str** - (CSV Mode Only) The title column name.\n* `document_attribute_columns` **list\\[str\\]** - (CSV Mode Only) The additional document attributes to include.\n* `connection_string` **str** - (blob only) The Azure Storage connection string.\n* `container_name` **str** - (blob only) The Azure Storage container name.\n* `base_dir` **str** - The base directory to read input from, relative to the root.\n* `storage_account_blob_url` **str** - The storage account blob URL to use.\n\nllm\n---\n\nThis is the base LLM configuration section. Other steps may override this configuration with their own LLM configuration.\n\n### Fields\n\n* `api_key` **str** - The OpenAI API key to use.\n* `type` **openai\\_chat|azure\\_openai\\_chat|openai\\_embedding|azure\\_openai\\_embedding** - The type of LLM to use.\n* `model` **str** - The model name.\n* `max_tokens` **int** - The maximum number of output tokens.\n* `request_timeout` **float** - The per-request timeout.\n* `api_base` **str** - The API base url to use.\n* `api_version` **str** - The API version\n* `organization` **str** - The client organization.\n* `proxy` **str** - The proxy URL to use.\n* `cognitive_services_endpoint` **str** - The url endpoint for cognitive services.\n* `deployment_name` **str** - The deployment name to use (Azure).\n* `model_supports_json` **bool** - Whether the model supports JSON-mode output.\n* `tokens_per_minute` **int** - Set a leaky-bucket throttle on tokens-per-minute.\n* `requests_per_minute` **int** - Set a leaky-bucket throttle on requests-per-minute.\n* `max_retries` **int** - The maximum number of retries to use.\n* `max_retry_wait` **float** - The maximum backoff time.\n* `sleep_on_rate_limit_recommendation` **bool** - Whether to adhere to sleep recommendations (Azure).\n* `concurrent_requests` **int** The number of open requests to allow at once.\n* `temperature` **float** - The temperature to use.\n* `top_p` **float** - The top-p value to use.\n* `n` **int** - The number of completions to generate.\n\nparallelization\n---------------\n\n### Fields\n\n* `stagger` **float** - The threading stagger value.\n* `num_threads` **int** - The maximum number of work threads.\n\nasync\\_mode\n-----------\n\n**asyncio|threaded** The async mode to use. Either `asyncio` or \\`threaded.\n\nembeddings\n----------\n\n### Fields\n\n* `llm` (see LLM top-level config)\n* `parallelization` (see Parallelization top-level config)\n* `async_mode` (see Async Mode top-level config)\n* `batch_size` **int** - The maximum batch size to use.\n* `batch_max_tokens` **int** - The maximum batch #-tokens.\n* `target` **required|all** - Determines which set of embeddings to emit.\n* `skip` **list\\[str\\]** - Which embeddings to skip.\n* `strategy` **dict** - Fully override the text-embedding strategy.\n\nchunks\n------\n\n### Fields\n\n* `size` **int** - The max chunk size in tokens.\n* `overlap` **int** - The chunk overlap in tokens.\n* `group_by_columns` **list\\[str\\]** - group documents by fields before chunking.\n* `encoding_model` **str** - The text encoding model to use. Default is to use the top-level encoding model.\n* `strategy` **dict** - Fully override the chunking strategy.\n\ncache\n-----\n\n### Fields\n\n* `type` **file|memory|none|blob** - The cache type to use. Default=`file`\n* `connection_string` **str** - (blob only) The Azure Storage connection string.\n* `container_name` **str** - (blob only) The Azure Storage container name.\n* `base_dir` **str** - The base directory to write cache to, relative to the root.\n* `storage_account_blob_url` **str** - The storage account blob URL to use.\n\nstorage\n-------\n\n### Fields\n\n* `type` **file|memory|blob** - The storage type to use. Default=`file`\n* `connection_string` **str** - (blob only) The Azure Storage connection string.\n* `container_name` **str** - (blob only) The Azure Storage container name.\n* `base_dir` **str** - The base directory to write reports to, relative to the root.\n* `storage_account_blob_url` **str** - The storage account blob URL to use.\n\nreporting\n---------\n\n### Fields\n\n* `type` **file|console|blob** - The reporting type to use. Default=`file`\n* `connection_string` **str** - (blob only) The Azure Storage connection string.\n* `container_name` **str** - (blob only) The Azure Storage container name.\n* `base_dir` **str** - The base directory to write reports to, relative to the root.\n* `storage_account_blob_url` **str** - The storage account blob URL to use.\n\nentity\\_extraction\n------------------\n\n### Fields\n\n* `llm` (see LLM top-level config)\n* `parallelization` (see Parallelization top-level config)\n* `async_mode` (see Async Mode top-level config)\n* `prompt` **str** - The prompt file to use.\n* `entity_types` **list\\[str\\]** - The entity types to identify.\n* `max_gleanings` **int** - The maximum number of gleaning cycles to use.\n* `encoding_model` **str** - The text encoding model to use. By default, this will use the top-level encoding model.\n* `strategy` **dict** - Fully override the entity extraction strategy.\n\nsummarize\\_descriptions\n-----------------------\n\n### Fields\n\n* `llm` (see LLM top-level config)\n* `parallelization` (see Parallelization top-level config)\n* `async_mode` (see Async Mode top-level config)\n* `prompt` **str** - The prompt file to use.\n* `max_length` **int** - The maximum number of output tokens per summarization.\n* `strategy` **dict** - Fully override the summarize description strategy.\n\nclaim\\_extraction\n-----------------\n\n### Fields\n\n* `enabled` **bool** - Whether to enable claim extraction. default=False\n* `llm` (see LLM top-level config)\n* `parallelization` (see Parallelization top-level config)\n* `async_mode` (see Async Mode top-level config)\n* `prompt` **str** - The prompt file to use.\n* `description` **str** - Describes the types of claims we want to extract.\n* `max_gleanings` **int** - The maximum number of gleaning cycles to use.\n* `encoding_model` **str** - The text encoding model to use. By default, this will use the top-level encoding model.\n* `strategy` **dict** - Fully override the claim extraction strategy.\n\ncommunity\\_reports\n------------------\n\n### Fields\n\n* `llm` (see LLM top-level config)\n* `parallelization` (see Parallelization top-level config)\n* `async_mode` (see Async Mode top-level config)\n* `prompt` **str** - The prompt file to use.\n* `max_length` **int** - The maximum number of output tokens per report.\n* `max_input_length` **int** - The maximum number of input tokens to use when generating reports.\n* `strategy` **dict** - Fully override the community reports strategy.\n\ncluster\\_graph\n--------------\n\n### Fields\n\n* `max_cluster_size` **int** - The maximum cluster size to emit.\n* `strategy` **dict** - Fully override the cluster\\_graph strategy.\n\nembed\\_graph\n------------\n\n### Fields\n\n* `enabled` **bool** - Whether to enable graph embeddings.\n* `num_walks` **int** - The node2vec number of walks.\n* `walk_length` **int** - The node2vec walk length.\n* `window_size` **int** - The node2vec window size.\n* `iterations` **int** - The node2vec number of iterations.\n* `random_seed` **int** - The node2vec random seed.\n* `strategy` **dict** - Fully override the embed graph strategy.\n\numap\n----\n\n### Fields\n\n* `enabled` **bool** - Whether to enable UMAP layouts.\n\nsnapshots\n---------\n\n### Fields\n\n* `graphml` **bool** - Emit graphml snapshots.\n* `raw_entities` **bool** - Emit raw entity snapshots.\n* `top_level_nodes` **bool** - Emit top-level-node snapshots.\n\nencoding\\_model\n---------------\n\n**str** - The text encoding model to use. Default is `cl100k_base`.\n\nskip\\_workflows\n---------------\n\n**list\\[str\\]** - Which workflow names to skip.", - "metadata": { - "title": "Default Configuration Mode (using JSON/YAML)", - "sourceURL": "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Custom Configuration Mode\n=========================\n\nThe primary configuration sections for Indexing Engine pipelines are described below. Each configuration section can be expressed in Python (for use in Python API mode) as well as YAML, but YAML is show here for brevity.\n\nUsing custom configuration is an advanced use-case. Most users will want to use the [Default Configuration](/graphrag/posts/config/overview)\n instead.\n\nIndexing Engine Examples\n------------------------\n\nThe [examples](https://github.com/microsoft/graphrag/blob/main/examples/)\n directory contains several examples of how to use the indexing engine with _custom configuration_.\n\nMost examples include two different forms of running the pipeline, both are contained in the examples `run.py`\n\n1. Using mostly the Python API\n2. Using mostly the a pipeline configuration file\n\nTo run an example:\n\n* Run `poetry shell` to activate a virtual environment with the required dependencies.\n* Run `PYTHONPATH=\"$(pwd)\" python examples/path_to_example/run.py` from the `root` directory.\n\nFor example to run the single\\_verb example, you would run the following commands:\n\n poetry shell\n\n PYTHONPATH=\"$(pwd)\" python examples/single_verb/run.py\n\nConfiguration Sections\n======================\n\n\\> extends\n==========\n\nThis configuration allows you to extend a base configuration file or files.\n\n # single base\n extends: ../base_config.yml\n\n # multiple bases\n extends:\n - ../base_config.yml\n - ../base_config2.yml\n\n\\> root\\_dir\n============\n\nThis configuration allows you to set the root directory for the pipeline. All data inputs and outputs are assumed to be relative to this path.\n\n root_dir: /workspace/data_project\n\n\\> storage\n==========\n\nThis configuration allows you define the output strategy for the pipeline.\n\n* `type`: The type of storage to use. Options are `file`, `memory`, and `blob`\n* `base_dir` (`type: file` only): The base directory to store the data in. This is relative to the config root.\n* `connection_string` (`type: blob` only): The connection string to use for blob storage.\n* `container_name` (`type: blob` only): The container to use for blob storage.\n\n\\> cache\n========\n\nThis configuration allows you define the cache strategy for the pipeline.\n\n* `type`: The type of cache to use. Options are `file` and `memory`, and `blob`.\n* `base_dir` (`type: file` only): The base directory to store the cache in. This is relative to the config root.\n* `connection_string` (`type: blob` only): The connection string to use for blob storage.\n* `container_name` (`type: blob` only): The container to use for blob storage.\n\n\\> reporting\n============\n\nThis configuration allows you define the reporting strategy for the pipeline. Report files are generated artifacts that summarize the performance metrics of the pipeline and emit any error messages.\n\n* `type`: The type of reporting to use. Options are `file`, `memory`, and `blob`\n* `base_dir` (`type: file` only): The base directory to store the reports in. This is relative to the config root.\n* `connection_string` (`type: blob` only): The connection string to use for blob storage.\n* `container_name` (`type: blob` only): The container to use for blob storage.\n\n\\> workflows\n============\n\nThis configuration section defines the workflow DAG for the pipeline. Here we define an array of workflows and express their inter-dependencies in steps:\n\n* `name`: The name of the workflow. This is used to reference the workflow in other parts of the config.\n* `steps`: The DataShaper steps that this workflow comprises. If a step defines an input in the form of `workflow:`, then it is assumed to have a dependency on the output of that workflow.\n\n workflows:\n - name: workflow1\n steps:\n - verb: derive\n args:\n column1: \"col1\"\n column2: \"col2\"\n - name: workflow2\n steps:\n - verb: derive\n args:\n column1: \"col1\"\n column2: \"col2\"\n input:\n # dependency established here\n source: workflow:workflow1\n\n\\> input\n========\n\n* `type`: The type of input to use. Options are `file` or `blob`.\n* `file_type`: The file type field discriminates between the different input types. Options are `csv` and `text`.\n* `base_dir`: The base directory to read the input files from. This is relative to the config file.\n* `file_pattern`: A regex to match the input files. The regex must have named groups for each of the fields in the file\\_filter.\n* `post_process`: A DataShaper workflow definition to apply to the input before executing the primary workflow.\n* `source_column` (`type: csv` only): The column containing the source/author of the data\n* `text_column` (`type: csv` only): The column containing the text of the data\n* `timestamp_column` (`type: csv` only): The column containing the timestamp of the data\n* `timestamp_format` (`type: csv` only): The format of the timestamp\n\n input:\n type: file\n file_type: csv\n base_dir: ../data/csv # the directory containing the CSV files, this is relative to the config file\n file_pattern: '.*[\\/](?P[^\\/]+)[\\/](?P\\d{4})-(?P\\d{2})-(?P\\d{2})_(?P[^_]+)_\\d+\\.csv$' # a regex to match the CSV files\n # An additional file filter which uses the named groups from the file_pattern to further filter the files\n # file_filter:\n # # source: (source_filter)\n # year: (2023)\n # month: (06)\n # # day: (22)\n source_column: \"author\" # the column containing the source/author of the data\n text_column: \"message\" # the column containing the text of the data\n timestamp_column: \"date(yyyyMMddHHmmss)\" # optional, the column containing the timestamp of the data\n timestamp_format: \"%Y%m%d%H%M%S\" # optional, the format of the timestamp\n post_process: # Optional, set of steps to process the data before going into the workflow\n - verb: filter\n args:\n column: \"title\",\n value: \"My document\"\n\n input:\n type: file\n file_type: csv\n base_dir: ../data/csv # the directory containing the CSV files, this is relative to the config file\n file_pattern: '.*[\\/](?P[^\\/]+)[\\/](?P\\d{4})-(?P\\d{2})-(?P\\d{2})_(?P[^_]+)_\\d+\\.csv$' # a regex to match the CSV files\n # An additional file filter which uses the named groups from the file_pattern to further filter the files\n # file_filter:\n # # source: (source_filter)\n # year: (2023)\n # month: (06)\n # # day: (22)\n post_process: # Optional, set of steps to process the data before going into the workflow\n - verb: filter\n args:\n column: \"title\",\n value: \"My document\"", - "markdown": "Custom Configuration Mode\n=========================\n\nThe primary configuration sections for Indexing Engine pipelines are described below. Each configuration section can be expressed in Python (for use in Python API mode) as well as YAML, but YAML is show here for brevity.\n\nUsing custom configuration is an advanced use-case. Most users will want to use the [Default Configuration](/graphrag/posts/config/overview)\n instead.\n\nIndexing Engine Examples\n------------------------\n\nThe [examples](https://github.com/microsoft/graphrag/blob/main/examples/)\n directory contains several examples of how to use the indexing engine with _custom configuration_.\n\nMost examples include two different forms of running the pipeline, both are contained in the examples `run.py`\n\n1. Using mostly the Python API\n2. Using mostly the a pipeline configuration file\n\nTo run an example:\n\n* Run `poetry shell` to activate a virtual environment with the required dependencies.\n* Run `PYTHONPATH=\"$(pwd)\" python examples/path_to_example/run.py` from the `root` directory.\n\nFor example to run the single\\_verb example, you would run the following commands:\n\n poetry shell\n\n PYTHONPATH=\"$(pwd)\" python examples/single_verb/run.py\n\nConfiguration Sections\n======================\n\n\\> extends\n==========\n\nThis configuration allows you to extend a base configuration file or files.\n\n # single base\n extends: ../base_config.yml\n\n # multiple bases\n extends:\n - ../base_config.yml\n - ../base_config2.yml\n\n\\> root\\_dir\n============\n\nThis configuration allows you to set the root directory for the pipeline. All data inputs and outputs are assumed to be relative to this path.\n\n root_dir: /workspace/data_project\n\n\\> storage\n==========\n\nThis configuration allows you define the output strategy for the pipeline.\n\n* `type`: The type of storage to use. Options are `file`, `memory`, and `blob`\n* `base_dir` (`type: file` only): The base directory to store the data in. This is relative to the config root.\n* `connection_string` (`type: blob` only): The connection string to use for blob storage.\n* `container_name` (`type: blob` only): The container to use for blob storage.\n\n\\> cache\n========\n\nThis configuration allows you define the cache strategy for the pipeline.\n\n* `type`: The type of cache to use. Options are `file` and `memory`, and `blob`.\n* `base_dir` (`type: file` only): The base directory to store the cache in. This is relative to the config root.\n* `connection_string` (`type: blob` only): The connection string to use for blob storage.\n* `container_name` (`type: blob` only): The container to use for blob storage.\n\n\\> reporting\n============\n\nThis configuration allows you define the reporting strategy for the pipeline. Report files are generated artifacts that summarize the performance metrics of the pipeline and emit any error messages.\n\n* `type`: The type of reporting to use. Options are `file`, `memory`, and `blob`\n* `base_dir` (`type: file` only): The base directory to store the reports in. This is relative to the config root.\n* `connection_string` (`type: blob` only): The connection string to use for blob storage.\n* `container_name` (`type: blob` only): The container to use for blob storage.\n\n\\> workflows\n============\n\nThis configuration section defines the workflow DAG for the pipeline. Here we define an array of workflows and express their inter-dependencies in steps:\n\n* `name`: The name of the workflow. This is used to reference the workflow in other parts of the config.\n* `steps`: The DataShaper steps that this workflow comprises. If a step defines an input in the form of `workflow:`, then it is assumed to have a dependency on the output of that workflow.\n\n workflows:\n - name: workflow1\n steps:\n - verb: derive\n args:\n column1: \"col1\"\n column2: \"col2\"\n - name: workflow2\n steps:\n - verb: derive\n args:\n column1: \"col1\"\n column2: \"col2\"\n input:\n # dependency established here\n source: workflow:workflow1\n\n\\> input\n========\n\n* `type`: The type of input to use. Options are `file` or `blob`.\n* `file_type`: The file type field discriminates between the different input types. Options are `csv` and `text`.\n* `base_dir`: The base directory to read the input files from. This is relative to the config file.\n* `file_pattern`: A regex to match the input files. The regex must have named groups for each of the fields in the file\\_filter.\n* `post_process`: A DataShaper workflow definition to apply to the input before executing the primary workflow.\n* `source_column` (`type: csv` only): The column containing the source/author of the data\n* `text_column` (`type: csv` only): The column containing the text of the data\n* `timestamp_column` (`type: csv` only): The column containing the timestamp of the data\n* `timestamp_format` (`type: csv` only): The format of the timestamp\n\n input:\n type: file\n file_type: csv\n base_dir: ../data/csv # the directory containing the CSV files, this is relative to the config file\n file_pattern: '.*[\\/](?P[^\\/]+)[\\/](?P\\d{4})-(?P\\d{2})-(?P\\d{2})_(?P[^_]+)_\\d+\\.csv$' # a regex to match the CSV files\n # An additional file filter which uses the named groups from the file_pattern to further filter the files\n # file_filter:\n # # source: (source_filter)\n # year: (2023)\n # month: (06)\n # # day: (22)\n source_column: \"author\" # the column containing the source/author of the data\n text_column: \"message\" # the column containing the text of the data\n timestamp_column: \"date(yyyyMMddHHmmss)\" # optional, the column containing the timestamp of the data\n timestamp_format: \"%Y%m%d%H%M%S\" # optional, the format of the timestamp\n post_process: # Optional, set of steps to process the data before going into the workflow\n - verb: filter\n args:\n column: \"title\",\n value: \"My document\"\n\n input:\n type: file\n file_type: csv\n base_dir: ../data/csv # the directory containing the CSV files, this is relative to the config file\n file_pattern: '.*[\\/](?P[^\\/]+)[\\/](?P\\d{4})-(?P\\d{2})-(?P\\d{2})_(?P[^_]+)_\\d+\\.csv$' # a regex to match the CSV files\n # An additional file filter which uses the named groups from the file_pattern to further filter the files\n # file_filter:\n # # source: (source_filter)\n # year: (2023)\n # month: (06)\n # # day: (22)\n post_process: # Optional, set of steps to process the data before going into the workflow\n - verb: filter\n args:\n column: \"title\",\n value: \"My document\"", - "metadata": { - "title": "Custom Configuration Mode", - "sourceURL": "https://microsoft.github.io/graphrag/posts/config/custom", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://microsoft.github.io/graphrag/posts/config/overview", - "https://github.com/microsoft/graphrag/blob/main/examples/", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Configuration Template\n======================\n\nThe following template can be used and stored as a `.env` in the the directory where you're are pointing the `--root` parameter on your Indexing Pipeline execution.\n\nFor details about how to run the Indexing Pipeline, refer to the [Index CLI](../../index/2-cli)\n documentation.\n\n.env File Template\n------------------\n\nRequired variables are uncommented. All the optional configuration can be turned on or off as needed.\n\n### Minimal Configuration\n\n # Base LLM Settings\n GRAPHRAG_API_KEY=\"your_api_key\"\n GRAPHRAG_API_BASE=\"http://.openai.azure.com\" # For Azure OpenAI Users\n GRAPHRAG_API_VERSION=\"api_version\" # For Azure OpenAI Users\n \n # Text Generation Settings\n GRAPHRAG_LLM_TYPE=\"azure_openai_chat\" # or openai_chat\n GRAPHRAG_LLM_DEPLOYMENT_NAME=\"gpt-4-turbo-preview\"\n GRAPHRAG_LLM_MODEL_SUPPORTS_JSON=True\n \n # Text Embedding Settings\n GRAPHRAG_EMBEDDING_TYPE=\"azure_openai_embedding\" # or openai_embedding\n GRAPHRAG_LLM_DEPLOYMENT_NAME=\"text-embedding-3-small\"\n \n # Data Mapping Settings\n GRAPHRAG_INPUT_TYPE=\"text\"\n \n\n### Full Configuration\n\n \n # Required LLM Config\n \n # Input Data Configuration\n GRAPHRAG_INPUT_TYPE=\"file\"\n \n # Plaintext Input Data Configuration\n # GRAPHRAG_INPUT_FILE_PATTERN=.*\\.txt\n \n # Text Input Data Configuration\n GRAPHRAG_INPUT_FILE_TYPE=\"text\"\n GRAPHRAG_INPUT_FILE_PATTERN=\".*\\.txt$\"\n GRAPHRAG_INPUT_SOURCE_COLUMN=source\n # GRAPHRAG_INPUT_TIMESTAMP_COLUMN=None\n # GRAPHRAG_INPUT_TIMESTAMP_FORMAT=None\n # GRAPHRAG_INPUT_TEXT_COLUMN=\"text\"\n # GRAPHRAG_INPUT_ATTRIBUTE_COLUMNS=id\n # GRAPHRAG_INPUT_TITLE_COLUMN=\"title\"\n # GRAPHRAG_INPUT_TYPE=\"file\"\n # GRAPHRAG_INPUT_CONNECTION_STRING=None\n # GRAPHRAG_INPUT_CONTAINER_NAME=None\n # GRAPHRAG_INPUT_BASE_DIR=None\n \n # Base LLM Settings\n GRAPHRAG_API_KEY=\"your_api_key\"\n GRAPHRAG_API_BASE=\"http://.openai.azure.com\" # For Azure OpenAI Users\n GRAPHRAG_API_VERSION=\"api_version\" # For Azure OpenAI Users\n # GRAPHRAG_API_ORGANIZATION=None\n # GRAPHRAG_API_PROXY=None\n \n # Text Generation Settings\n # GRAPHRAG_LLM_TYPE=openai_chat\n GRAPHRAG_LLM_API_KEY=\"your_api_key\" # If GRAPHRAG_API_KEY is not set\n GRAPHRAG_LLM_API_BASE=\"http://.openai.azure.com\" # For Azure OpenAI Users and if GRAPHRAG_API_BASE is not set\n GRAPHRAG_LLM_API_VERSION=\"api_version\" # For Azure OpenAI Users and if GRAPHRAG_API_VERSION is not set\n GRAPHRAG_LLM_MODEL_SUPPORTS_JSON=True # Suggested by default\n # GRAPHRAG_LLM_API_ORGANIZATION=None\n # GRAPHRAG_LLM_API_PROXY=None\n # GRAPHRAG_LLM_DEPLOYMENT_NAME=None\n # GRAPHRAG_LLM_MODEL=gpt-4-turbo-preview\n # GRAPHRAG_LLM_MAX_TOKENS=4000\n # GRAPHRAG_LLM_REQUEST_TIMEOUT=180\n # GRAPHRAG_LLM_THREAD_COUNT=50\n # GRAPHRAG_LLM_THREAD_STAGGER=0.3\n # GRAPHRAG_LLM_CONCURRENT_REQUESTS=25\n # GRAPHRAG_LLM_TPM=0\n # GRAPHRAG_LLM_RPM=0\n # GRAPHRAG_LLM_MAX_RETRIES=10\n # GRAPHRAG_LLM_MAX_RETRY_WAIT=10\n # GRAPHRAG_LLM_SLEEP_ON_RATE_LIMIT_RECOMMENDATION=True\n \n # Text Embedding Settings\n # GRAPHRAG_EMBEDDING_TYPE=openai_embedding\n GRAPHRAG_EMBEDDING_API_KEY=\"your_api_key\" # If GRAPHRAG_API_KEY is not set\n GRAPHRAG_EMBEDDING_API_BASE=\"http://.openai.azure.com\" # For Azure OpenAI Users and if GRAPHRAG_API_BASE is not set\n GRAPHRAG_EMBEDDING_API_VERSION=\"api_version\" # For Azure OpenAI Users and if GRAPHRAG_API_VERSION is not set\n # GRAPHRAG_EMBEDDING_API_ORGANIZATION=None\n # GRAPHRAG_EMBEDDING_API_PROXY=None\n # GRAPHRAG_EMBEDDING_DEPLOYMENT_NAME=None\n # GRAPHRAG_EMBEDDING_MODEL=text-embedding-3-small\n # GRAPHRAG_EMBEDDING_BATCH_SIZE=16\n # GRAPHRAG_EMBEDDING_BATCH_MAX_TOKENS=8191\n # GRAPHRAG_EMBEDDING_TARGET=required\n # GRAPHRAG_EMBEDDING_SKIP=None\n # GRAPHRAG_EMBEDDING_THREAD_COUNT=None\n # GRAPHRAG_EMBEDDING_THREAD_STAGGER=50\n # GRAPHRAG_EMBEDDING_CONCURRENT_REQUESTS=25\n # GRAPHRAG_EMBEDDING_TPM=0\n # GRAPHRAG_EMBEDDING_RPM=0\n # GRAPHRAG_EMBEDDING_MAX_RETRIES=10\n # GRAPHRAG_EMBEDDING_MAX_RETRY_WAIT=10\n # GRAPHRAG_EMBEDDING_SLEEP_ON_RATE_LIMIT_RECOMMENDATION=True\n \n # Data Mapping Settings\n # GRAPHRAG_INPUT_ENCODING=utf-8\n \n # Data Chunking\n # GRAPHRAG_CHUNK_SIZE=1200\n # GRAPHRAG_CHUNK_OVERLAP=100\n # GRAPHRAG_CHUNK_BY_COLUMNS=id\n \n # Prompting Overrides\n # GRAPHRAG_ENTITY_EXTRACTION_PROMPT_FILE=None\n # GRAPHRAG_ENTITY_EXTRACTION_MAX_GLEANINGS=1\n # GRAPHRAG_ENTITY_EXTRACTION_ENTITY_TYPES=organization,person,event,geo\n # GRAPHRAG_SUMMARIZE_DESCRIPTIONS_PROMPT_FILE=None\n # GRAPHRAG_SUMMARIZE_DESCRIPTIONS_MAX_LENGTH=500\n # GRAPHRAG_CLAIM_EXTRACTION_DESCRIPTION=\"Any claims or facts that could be relevant to threat analysis.\"\n # GRAPHRAG_CLAIM_EXTRACTION_PROMPT_FILE=None\n # GRAPHRAG_CLAIM_EXTRACTION_MAX_GLEANINGS=1\n # GRAPHRAG_COMMUNITY_REPORT_PROMPT_FILE=None\n # GRAPHRAG_COMMUNITY_REPORT_MAX_LENGTH=1500\n \n # Storage\n # GRAPHRAG_STORAGE_TYPE=file\n # GRAPHRAG_STORAGE_CONNECTION_STRING=None\n # GRAPHRAG_STORAGE_CONTAINER_NAME=None\n # GRAPHRAG_STORAGE_BASE_DIR=None\n \n # Cache\n # GRAPHRAG_CACHE_TYPE=file\n # GRAPHRAG_CACHE_CONNECTION_STRING=None\n # GRAPHRAG_CACHE_CONTAINER_NAME=None\n # GRAPHRAG_CACHE_BASE_DIR=None\n \n # Reporting\n # GRAPHRAG_REPORTING_TYPE=file\n # GRAPHRAG_REPORTING_CONNECTION_STRING=None\n # GRAPHRAG_REPORTING_CONTAINER_NAME=None\n # GRAPHRAG_REPORTING_BASE_DIR=None\n \n # Node2Vec Parameters\n # GRAPHRAG_NODE2VEC_ENABLED=False\n # GRAPHRAG_NODE2VEC_NUM_WALKS=10\n # GRAPHRAG_NODE2VEC_WALK_LENGTH=40\n # GRAPHRAG_NODE2VEC_WINDOW_SIZE=2\n # GRAPHRAG_NODE2VEC_ITERATIONS=3\n # GRAPHRAG_NODE2VEC_RANDOM_SEED=597832\n \n # Data Snapshotting\n # GRAPHRAG_SNAPSHOT_GRAPHML=False\n # GRAPHRAG_SNAPSHOT_RAW_ENTITIES=False\n # GRAPHRAG_SNAPSHOT_TOP_LEVEL_NODES=False\n \n # Miscellaneous Settings\n # GRAPHRAG_ASYNC_MODE=asyncio\n # GRAPHRAG_ENCODING_MODEL=cl100k_base\n # GRAPHRAG_MAX_CLUSTER_SIZE=10\n # GRAPHRAG_ENTITY_RESOLUTION_ENABLED=False\n # GRAPHRAG_SKIP_WORKFLOWS=None\n # GRAPHRAG_UMAP_ENABLED=False", - "markdown": "Configuration Template\n======================\n\nThe following template can be used and stored as a `.env` in the the directory where you're are pointing the `--root` parameter on your Indexing Pipeline execution.\n\nFor details about how to run the Indexing Pipeline, refer to the [Index CLI](../../index/2-cli)\n documentation.\n\n.env File Template\n------------------\n\nRequired variables are uncommented. All the optional configuration can be turned on or off as needed.\n\n### Minimal Configuration\n\n # Base LLM Settings\n GRAPHRAG_API_KEY=\"your_api_key\"\n GRAPHRAG_API_BASE=\"http://.openai.azure.com\" # For Azure OpenAI Users\n GRAPHRAG_API_VERSION=\"api_version\" # For Azure OpenAI Users\n \n # Text Generation Settings\n GRAPHRAG_LLM_TYPE=\"azure_openai_chat\" # or openai_chat\n GRAPHRAG_LLM_DEPLOYMENT_NAME=\"gpt-4-turbo-preview\"\n GRAPHRAG_LLM_MODEL_SUPPORTS_JSON=True\n \n # Text Embedding Settings\n GRAPHRAG_EMBEDDING_TYPE=\"azure_openai_embedding\" # or openai_embedding\n GRAPHRAG_LLM_DEPLOYMENT_NAME=\"text-embedding-3-small\"\n \n # Data Mapping Settings\n GRAPHRAG_INPUT_TYPE=\"text\"\n \n\n### Full Configuration\n\n \n # Required LLM Config\n \n # Input Data Configuration\n GRAPHRAG_INPUT_TYPE=\"file\"\n \n # Plaintext Input Data Configuration\n # GRAPHRAG_INPUT_FILE_PATTERN=.*\\.txt\n \n # Text Input Data Configuration\n GRAPHRAG_INPUT_FILE_TYPE=\"text\"\n GRAPHRAG_INPUT_FILE_PATTERN=\".*\\.txt$\"\n GRAPHRAG_INPUT_SOURCE_COLUMN=source\n # GRAPHRAG_INPUT_TIMESTAMP_COLUMN=None\n # GRAPHRAG_INPUT_TIMESTAMP_FORMAT=None\n # GRAPHRAG_INPUT_TEXT_COLUMN=\"text\"\n # GRAPHRAG_INPUT_ATTRIBUTE_COLUMNS=id\n # GRAPHRAG_INPUT_TITLE_COLUMN=\"title\"\n # GRAPHRAG_INPUT_TYPE=\"file\"\n # GRAPHRAG_INPUT_CONNECTION_STRING=None\n # GRAPHRAG_INPUT_CONTAINER_NAME=None\n # GRAPHRAG_INPUT_BASE_DIR=None\n \n # Base LLM Settings\n GRAPHRAG_API_KEY=\"your_api_key\"\n GRAPHRAG_API_BASE=\"http://.openai.azure.com\" # For Azure OpenAI Users\n GRAPHRAG_API_VERSION=\"api_version\" # For Azure OpenAI Users\n # GRAPHRAG_API_ORGANIZATION=None\n # GRAPHRAG_API_PROXY=None\n \n # Text Generation Settings\n # GRAPHRAG_LLM_TYPE=openai_chat\n GRAPHRAG_LLM_API_KEY=\"your_api_key\" # If GRAPHRAG_API_KEY is not set\n GRAPHRAG_LLM_API_BASE=\"http://.openai.azure.com\" # For Azure OpenAI Users and if GRAPHRAG_API_BASE is not set\n GRAPHRAG_LLM_API_VERSION=\"api_version\" # For Azure OpenAI Users and if GRAPHRAG_API_VERSION is not set\n GRAPHRAG_LLM_MODEL_SUPPORTS_JSON=True # Suggested by default\n # GRAPHRAG_LLM_API_ORGANIZATION=None\n # GRAPHRAG_LLM_API_PROXY=None\n # GRAPHRAG_LLM_DEPLOYMENT_NAME=None\n # GRAPHRAG_LLM_MODEL=gpt-4-turbo-preview\n # GRAPHRAG_LLM_MAX_TOKENS=4000\n # GRAPHRAG_LLM_REQUEST_TIMEOUT=180\n # GRAPHRAG_LLM_THREAD_COUNT=50\n # GRAPHRAG_LLM_THREAD_STAGGER=0.3\n # GRAPHRAG_LLM_CONCURRENT_REQUESTS=25\n # GRAPHRAG_LLM_TPM=0\n # GRAPHRAG_LLM_RPM=0\n # GRAPHRAG_LLM_MAX_RETRIES=10\n # GRAPHRAG_LLM_MAX_RETRY_WAIT=10\n # GRAPHRAG_LLM_SLEEP_ON_RATE_LIMIT_RECOMMENDATION=True\n \n # Text Embedding Settings\n # GRAPHRAG_EMBEDDING_TYPE=openai_embedding\n GRAPHRAG_EMBEDDING_API_KEY=\"your_api_key\" # If GRAPHRAG_API_KEY is not set\n GRAPHRAG_EMBEDDING_API_BASE=\"http://.openai.azure.com\" # For Azure OpenAI Users and if GRAPHRAG_API_BASE is not set\n GRAPHRAG_EMBEDDING_API_VERSION=\"api_version\" # For Azure OpenAI Users and if GRAPHRAG_API_VERSION is not set\n # GRAPHRAG_EMBEDDING_API_ORGANIZATION=None\n # GRAPHRAG_EMBEDDING_API_PROXY=None\n # GRAPHRAG_EMBEDDING_DEPLOYMENT_NAME=None\n # GRAPHRAG_EMBEDDING_MODEL=text-embedding-3-small\n # GRAPHRAG_EMBEDDING_BATCH_SIZE=16\n # GRAPHRAG_EMBEDDING_BATCH_MAX_TOKENS=8191\n # GRAPHRAG_EMBEDDING_TARGET=required\n # GRAPHRAG_EMBEDDING_SKIP=None\n # GRAPHRAG_EMBEDDING_THREAD_COUNT=None\n # GRAPHRAG_EMBEDDING_THREAD_STAGGER=50\n # GRAPHRAG_EMBEDDING_CONCURRENT_REQUESTS=25\n # GRAPHRAG_EMBEDDING_TPM=0\n # GRAPHRAG_EMBEDDING_RPM=0\n # GRAPHRAG_EMBEDDING_MAX_RETRIES=10\n # GRAPHRAG_EMBEDDING_MAX_RETRY_WAIT=10\n # GRAPHRAG_EMBEDDING_SLEEP_ON_RATE_LIMIT_RECOMMENDATION=True\n \n # Data Mapping Settings\n # GRAPHRAG_INPUT_ENCODING=utf-8\n \n # Data Chunking\n # GRAPHRAG_CHUNK_SIZE=1200\n # GRAPHRAG_CHUNK_OVERLAP=100\n # GRAPHRAG_CHUNK_BY_COLUMNS=id\n \n # Prompting Overrides\n # GRAPHRAG_ENTITY_EXTRACTION_PROMPT_FILE=None\n # GRAPHRAG_ENTITY_EXTRACTION_MAX_GLEANINGS=1\n # GRAPHRAG_ENTITY_EXTRACTION_ENTITY_TYPES=organization,person,event,geo\n # GRAPHRAG_SUMMARIZE_DESCRIPTIONS_PROMPT_FILE=None\n # GRAPHRAG_SUMMARIZE_DESCRIPTIONS_MAX_LENGTH=500\n # GRAPHRAG_CLAIM_EXTRACTION_DESCRIPTION=\"Any claims or facts that could be relevant to threat analysis.\"\n # GRAPHRAG_CLAIM_EXTRACTION_PROMPT_FILE=None\n # GRAPHRAG_CLAIM_EXTRACTION_MAX_GLEANINGS=1\n # GRAPHRAG_COMMUNITY_REPORT_PROMPT_FILE=None\n # GRAPHRAG_COMMUNITY_REPORT_MAX_LENGTH=1500\n \n # Storage\n # GRAPHRAG_STORAGE_TYPE=file\n # GRAPHRAG_STORAGE_CONNECTION_STRING=None\n # GRAPHRAG_STORAGE_CONTAINER_NAME=None\n # GRAPHRAG_STORAGE_BASE_DIR=None\n \n # Cache\n # GRAPHRAG_CACHE_TYPE=file\n # GRAPHRAG_CACHE_CONNECTION_STRING=None\n # GRAPHRAG_CACHE_CONTAINER_NAME=None\n # GRAPHRAG_CACHE_BASE_DIR=None\n \n # Reporting\n # GRAPHRAG_REPORTING_TYPE=file\n # GRAPHRAG_REPORTING_CONNECTION_STRING=None\n # GRAPHRAG_REPORTING_CONTAINER_NAME=None\n # GRAPHRAG_REPORTING_BASE_DIR=None\n \n # Node2Vec Parameters\n # GRAPHRAG_NODE2VEC_ENABLED=False\n # GRAPHRAG_NODE2VEC_NUM_WALKS=10\n # GRAPHRAG_NODE2VEC_WALK_LENGTH=40\n # GRAPHRAG_NODE2VEC_WINDOW_SIZE=2\n # GRAPHRAG_NODE2VEC_ITERATIONS=3\n # GRAPHRAG_NODE2VEC_RANDOM_SEED=597832\n \n # Data Snapshotting\n # GRAPHRAG_SNAPSHOT_GRAPHML=False\n # GRAPHRAG_SNAPSHOT_RAW_ENTITIES=False\n # GRAPHRAG_SNAPSHOT_TOP_LEVEL_NODES=False\n \n # Miscellaneous Settings\n # GRAPHRAG_ASYNC_MODE=asyncio\n # GRAPHRAG_ENCODING_MODEL=cl100k_base\n # GRAPHRAG_MAX_CLUSTER_SIZE=10\n # GRAPHRAG_ENTITY_RESOLUTION_ENABLED=False\n # GRAPHRAG_SKIP_WORKFLOWS=None\n # GRAPHRAG_UMAP_ENABLED=False", - "metadata": { - "title": "Configuration Template", - "sourceURL": "https://microsoft.github.io/graphrag/posts/config/template", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://microsoft.github.io/graphrag/posts/config/template/../../index/2-cli", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Prompt Tuning āš™ļø\n================\n\nThis page provides an overview of the prompt tuning options available for the GraphRAG indexing engine.\n\nDefault Prompts\n---------------\n\nThe default prompts are the simplest way to get started with the GraphRAG system. It is designed to work out-of-the-box with minimal configuration. You can find more detail about these prompts in the following links:\n\n* [Entity/Relationship Extraction](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/graph/prompts.py)\n \n* [Entity/Relationship Description Summarization](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/summarize/prompts.py)\n \n* [Claim Extraction](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/claims/prompts.py)\n \n* [Community Reports](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/community_reports/prompts.py)\n \n\nAuto Templating\n---------------\n\nAuto Templating leverages your input data and LLM interactions to create domain adaptive templates for the generation of the knowledge graph. It is highly encouraged to run it as it will yield better results when executing an Index Run. For more details about how to use it, please refer to the [Auto Templating](/graphrag/posts/prompt_tuning/auto_prompt_tuning)\n documentation.\n\nManual Configuration\n--------------------\n\nManual configuration is an advanced use-case. Most users will want to use the Auto Templating feature instead. Details about how to use manual configuration are available in the [Manual Prompt Configuration](/graphrag/posts/prompt_tuning/manual_prompt_tuning)\n documentation.", - "markdown": "Prompt Tuning āš™ļø\n================\n\nThis page provides an overview of the prompt tuning options available for the GraphRAG indexing engine.\n\nDefault Prompts\n---------------\n\nThe default prompts are the simplest way to get started with the GraphRAG system. It is designed to work out-of-the-box with minimal configuration. You can find more detail about these prompts in the following links:\n\n* [Entity/Relationship Extraction](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/graph/prompts.py)\n \n* [Entity/Relationship Description Summarization](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/summarize/prompts.py)\n \n* [Claim Extraction](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/claims/prompts.py)\n \n* [Community Reports](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/community_reports/prompts.py)\n \n\nAuto Templating\n---------------\n\nAuto Templating leverages your input data and LLM interactions to create domain adaptive templates for the generation of the knowledge graph. It is highly encouraged to run it as it will yield better results when executing an Index Run. For more details about how to use it, please refer to the [Auto Templating](/graphrag/posts/prompt_tuning/auto_prompt_tuning)\n documentation.\n\nManual Configuration\n--------------------\n\nManual configuration is an advanced use-case. Most users will want to use the Auto Templating feature instead. Details about how to use manual configuration are available in the [Manual Prompt Configuration](/graphrag/posts/prompt_tuning/manual_prompt_tuning)\n documentation.", - "metadata": { - "title": "Prompt Tuning āš™ļø", - "sourceURL": "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/graph/prompts.py", - "http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/summarize/prompts.py", - "http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/claims/prompts.py", - "http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/community_reports/prompts.py", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Prompt Tuning āš™ļø\n================\n\nGraphRAG provides the ability to create domain adaptive templates for the generation of the knowledge graph. This step is optional, though it is highly encouraged to run it as it will yield better results when executing an Index Run.\n\nThe templates are generated by loading the inputs, splitting them into chunks (text units) and then running a series of LLM invocations and template substitutions to generate the final prompts. We suggest using the default values provided by the script, but in this page you'll find the detail of each in case you want to further explore and tweak the template generation algorithm.\n\nPrerequisites\n-------------\n\nBefore running the automatic template generation make sure you have already initialized your workspace with the `graphrag.index --init` command. This will create the necessary configuration files and the default prompts. Refer to the [Init Documentation](/graphrag/posts/config/init)\n for more information about the initialization process.\n\nUsage\n-----\n\nYou can run the main script from the command line with various options:\n\n python -m graphrag.prompt_tune [--root ROOT] [--domain DOMAIN] [--method METHOD] [--limit LIMIT] [--language LANGUAGE] [--max-tokens MAX_TOKENS] [--chunk-size CHUNK_SIZE] [--no-entity-types] [--output OUTPUT]\n\nCommand-Line Options\n--------------------\n\n* `--root` (optional): The data project root directory, including the config files (YML, JSON, or .env). Defaults to the current directory.\n \n* `--domain` (optional): The domain related to your input data, such as 'space science', 'microbiology', or 'environmental news'. If left empty, the domain will be inferred from the input data.\n \n* `--method` (optional): The method to select documents. Options are all, random, or top. Default is random.\n \n* `--limit` (optional): The limit of text units to load when using random or top selection. Default is 15.\n \n* `--language` (optional): The language to use for input processing. If it is different from the inputs' language, the LLM will translate. Default is \"\" meaning it will be automatically detected from the inputs.\n \n* `--max-tokens` (optional): Maximum token count for prompt generation. Default is 2000.\n \n* `--chunk-size` (optional): The size in tokens to use for generating text units from input documents. Default is 200.\n \n* `--no-entity-types` (optional): Use untyped entity extraction generation. We recommend using this when your data covers a lot of topics or it is highly randomized.\n \n* `--output` (optional): The folder to save the generated prompts. Default is \"prompts\".\n \n\nExample Usage\n-------------\n\n python -m graphrag.prompt_tune --root /path/to/project --domain \"environmental news\" --method random --limit 10 --language English --max-tokens 2048 --chunk-size 256 --no-entity-types --output /path/to/output\n\nor, with minimal configuration (suggested):\n\n python -m graphrag.prompt_tune --root /path/to/project --no-entity-types\n\nDocument Selection Methods\n--------------------------\n\nThe auto template feature ingests the input data and then divides it into text units the size of the chunk size parameter. After that, it uses one of the following selection methods to pick a sample to work with for template generation:\n\n* `random`: Select text units randomly. This is the default and recommended option.\n* `top`: Select the head n text units.\n* `all`: Use all text units for the generation. Use only with small datasets; this option is not usually recommended.\n\nModify Env Vars\n---------------\n\nAfter running auto-templating, you should modify the following environment variables (or config variables) to pick up the new prompts on your index run. Note: Please make sure to update the correct path to the generated prompts, in this example we are using the default \"prompts\" path.\n\n* `GRAPHRAG_ENTITY_EXTRACTION_PROMPT_FILE` = \"prompts/entity\\_extraction.txt\"\n \n* `GRAPHRAG_COMMUNITY_REPORT_PROMPT_FILE` = \"prompts/community\\_report.txt\"\n \n* `GRAPHRAG_SUMMARIZE_DESCRIPTIONS_PROMPT_FILE` = \"prompts/summarize\\_descriptions.txt\"", - "markdown": "Prompt Tuning āš™ļø\n================\n\nGraphRAG provides the ability to create domain adaptive templates for the generation of the knowledge graph. This step is optional, though it is highly encouraged to run it as it will yield better results when executing an Index Run.\n\nThe templates are generated by loading the inputs, splitting them into chunks (text units) and then running a series of LLM invocations and template substitutions to generate the final prompts. We suggest using the default values provided by the script, but in this page you'll find the detail of each in case you want to further explore and tweak the template generation algorithm.\n\nPrerequisites\n-------------\n\nBefore running the automatic template generation make sure you have already initialized your workspace with the `graphrag.index --init` command. This will create the necessary configuration files and the default prompts. Refer to the [Init Documentation](/graphrag/posts/config/init)\n for more information about the initialization process.\n\nUsage\n-----\n\nYou can run the main script from the command line with various options:\n\n python -m graphrag.prompt_tune [--root ROOT] [--domain DOMAIN] [--method METHOD] [--limit LIMIT] [--language LANGUAGE] [--max-tokens MAX_TOKENS] [--chunk-size CHUNK_SIZE] [--no-entity-types] [--output OUTPUT]\n\nCommand-Line Options\n--------------------\n\n* `--root` (optional): The data project root directory, including the config files (YML, JSON, or .env). Defaults to the current directory.\n \n* `--domain` (optional): The domain related to your input data, such as 'space science', 'microbiology', or 'environmental news'. If left empty, the domain will be inferred from the input data.\n \n* `--method` (optional): The method to select documents. Options are all, random, or top. Default is random.\n \n* `--limit` (optional): The limit of text units to load when using random or top selection. Default is 15.\n \n* `--language` (optional): The language to use for input processing. If it is different from the inputs' language, the LLM will translate. Default is \"\" meaning it will be automatically detected from the inputs.\n \n* `--max-tokens` (optional): Maximum token count for prompt generation. Default is 2000.\n \n* `--chunk-size` (optional): The size in tokens to use for generating text units from input documents. Default is 200.\n \n* `--no-entity-types` (optional): Use untyped entity extraction generation. We recommend using this when your data covers a lot of topics or it is highly randomized.\n \n* `--output` (optional): The folder to save the generated prompts. Default is \"prompts\".\n \n\nExample Usage\n-------------\n\n python -m graphrag.prompt_tune --root /path/to/project --domain \"environmental news\" --method random --limit 10 --language English --max-tokens 2048 --chunk-size 256 --no-entity-types --output /path/to/output\n\nor, with minimal configuration (suggested):\n\n python -m graphrag.prompt_tune --root /path/to/project --no-entity-types\n\nDocument Selection Methods\n--------------------------\n\nThe auto template feature ingests the input data and then divides it into text units the size of the chunk size parameter. After that, it uses one of the following selection methods to pick a sample to work with for template generation:\n\n* `random`: Select text units randomly. This is the default and recommended option.\n* `top`: Select the head n text units.\n* `all`: Use all text units for the generation. Use only with small datasets; this option is not usually recommended.\n\nModify Env Vars\n---------------\n\nAfter running auto-templating, you should modify the following environment variables (or config variables) to pick up the new prompts on your index run. Note: Please make sure to update the correct path to the generated prompts, in this example we are using the default \"prompts\" path.\n\n* `GRAPHRAG_ENTITY_EXTRACTION_PROMPT_FILE` = \"prompts/entity\\_extraction.txt\"\n \n* `GRAPHRAG_COMMUNITY_REPORT_PROMPT_FILE` = \"prompts/community\\_report.txt\"\n \n* `GRAPHRAG_SUMMARIZE_DESCRIPTIONS_PROMPT_FILE` = \"prompts/summarize\\_descriptions.txt\"", - "metadata": { - "title": "Prompt Tuning āš™ļø", - "sourceURL": "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - }, - { - "content": "Prompt Tuningāš™ļø\n===============\n\nThe GraphRAG indexer, by default, will run with a handful of prompts that are designed to work well in the broad context of knowledge discovery. However, it is quite common to want to tune the prompts to better suit your specific use case. We provide a means for you to do this by allowing you to specify a custom prompt file, which will each use a series of token-replacements internally.\n\nEach of these prompts may be overridden by writing a custom prompt file in plaintext. We use token-replacements in the form of `{token_name}`, and the descriptions for the available tokens can be found below.\n\nEntity/Relationship Extraction\n------------------------------\n\n[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/graph/prompts.py)\n\n### Tokens (values provided by extractor)\n\n* **{input\\_text}** - The input text to be processed.\n* **{entity\\_types}** - A list of entity types\n* **{tuple\\_delimiter}** - A delimiter for separating values within a tuple. A single tuple is used to represent an individual entity or relationship.\n* **{record\\_delimiter}** - A delimiter for separating tuple instances.\n* **{completion\\_delimiter}** - An indicator for when generation is complete.\n\nSummarize Entity/Relationship Descriptions\n------------------------------------------\n\n[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/summarize/prompts.py)\n\n### Tokens (values provided by extractor)\n\n* **{entity\\_name}** - The name of the entity or the source/target pair of the relationship.\n* **{description\\_list}** - A list of descriptions for the entity or relationship.\n\nClaim Extraction\n----------------\n\n[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/claims/prompts.py)\n\n### Tokens (values provided by extractor)\n\n* **{input\\_text}** - The input text to be processed.\n* **{tuple\\_delimiter}** - A delimiter for separating values within a tuple. A single tuple is used to represent an individual entity or relationship.\n* **{record\\_delimiter}** - A delimiter for separating tuple instances.\n* **{completion\\_delimiter}** - An indicator for when generation is complete.\n\nNote: there is additional paramater for the `Claim Description` that is used in claim extraction. The default value is\n\n`\"Any claims or facts that could be relevant to information discovery.\"`\n\nSee the [configuration documentation](/graphrag/posts/config/overview/)\n for details on how to change this.\n\nGenerate Community Reports\n--------------------------\n\n[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/community_reports/prompts.py)\n\n### Tokens (values provided by extractor)\n\n* **{input\\_text}** - The input text to generate the report with. This will contain tables of entities and relationships.", - "markdown": "Prompt Tuningāš™ļø\n===============\n\nThe GraphRAG indexer, by default, will run with a handful of prompts that are designed to work well in the broad context of knowledge discovery. However, it is quite common to want to tune the prompts to better suit your specific use case. We provide a means for you to do this by allowing you to specify a custom prompt file, which will each use a series of token-replacements internally.\n\nEach of these prompts may be overridden by writing a custom prompt file in plaintext. We use token-replacements in the form of `{token_name}`, and the descriptions for the available tokens can be found below.\n\nEntity/Relationship Extraction\n------------------------------\n\n[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/graph/prompts.py)\n\n### Tokens (values provided by extractor)\n\n* **{input\\_text}** - The input text to be processed.\n* **{entity\\_types}** - A list of entity types\n* **{tuple\\_delimiter}** - A delimiter for separating values within a tuple. A single tuple is used to represent an individual entity or relationship.\n* **{record\\_delimiter}** - A delimiter for separating tuple instances.\n* **{completion\\_delimiter}** - An indicator for when generation is complete.\n\nSummarize Entity/Relationship Descriptions\n------------------------------------------\n\n[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/summarize/prompts.py)\n\n### Tokens (values provided by extractor)\n\n* **{entity\\_name}** - The name of the entity or the source/target pair of the relationship.\n* **{description\\_list}** - A list of descriptions for the entity or relationship.\n\nClaim Extraction\n----------------\n\n[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/claims/prompts.py)\n\n### Tokens (values provided by extractor)\n\n* **{input\\_text}** - The input text to be processed.\n* **{tuple\\_delimiter}** - A delimiter for separating values within a tuple. A single tuple is used to represent an individual entity or relationship.\n* **{record\\_delimiter}** - A delimiter for separating tuple instances.\n* **{completion\\_delimiter}** - An indicator for when generation is complete.\n\nNote: there is additional paramater for the `Claim Description` that is used in claim extraction. The default value is\n\n`\"Any claims or facts that could be relevant to information discovery.\"`\n\nSee the [configuration documentation](/graphrag/posts/config/overview/)\n for details on how to change this.\n\nGenerate Community Reports\n--------------------------\n\n[Prompt Source](http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/community_reports/prompts.py)\n\n### Tokens (values provided by extractor)\n\n* **{input\\_text}** - The input text to generate the report with. This will contain tables of entities and relationships.", - "metadata": { - "title": "Prompt Tuningāš™ļø", - "sourceURL": "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "pageStatusCode": 200, - "ogLocaleAlternate": [] - }, - "linksOnPage": [ - "https://microsoft.github.io/graphrag/", - "https://microsoft.github.io/graphrag/posts/get_started/", - "https://microsoft.github.io/graphrag/posts/developing/", - "https://microsoft.github.io/graphrag/posts/index/overview/", - "https://microsoft.github.io/graphrag/posts/index/0-architecture/", - "https://microsoft.github.io/graphrag/posts/index/1-default_dataflow/", - "https://microsoft.github.io/graphrag/posts/index/2-cli/", - "https://microsoft.github.io/graphrag/posts/config/overview/", - "https://microsoft.github.io/graphrag/posts/config/init", - "https://microsoft.github.io/graphrag/posts/config/env_vars", - "https://microsoft.github.io/graphrag/posts/config/json_yaml", - "https://microsoft.github.io/graphrag/posts/config/custom", - "https://microsoft.github.io/graphrag/posts/config/template", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/overview/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/auto_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/prompt_tuning/manual_prompt_tuning/", - "https://microsoft.github.io/graphrag/posts/query/overview/", - "https://microsoft.github.io/graphrag/posts/query/1-local_search/", - "https://microsoft.github.io/graphrag/posts/query/2-question_generation/", - "https://microsoft.github.io/graphrag/posts/query/0-global_search/", - "https://microsoft.github.io/graphrag/posts/query/3-cli/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/overview/", - "https://microsoft.github.io/graphrag/posts/query/notebooks/global_search_nb", - "https://microsoft.github.io/graphrag/posts/query/notebooks/local_search_nb", - "http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/graph/prompts.py", - "http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/summarize/prompts.py", - "http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/claims/prompts.py", - "http://github.com/microsoft/graphrag/blob/main/graphrag/index/graph/extractors/community_reports/prompts.py", - "https://go.microsoft.com/fwlink/?LinkId=521839", - "https://go.microsoft.com/fwlink/?LinkId=2259814", - "https://go.microsoft.com/fwlink/?LinkID=206977", - "https://www.microsoft.com/trademarks", - "https://www.microsoft.com", - "https://github.com/microsoft/graphrag", - "https://github.com/Azure-Samples/graphrag-accelerator" - ] - } -] \ No newline at end of file diff --git a/graphfleet/input/pdf/SELFCHECKGPT.pdf b/graphfleet/input/pdf/SELFCHECKGPT.pdf deleted file mode 100644 index 8cef32e0e..000000000 Binary files a/graphfleet/input/pdf/SELFCHECKGPT.pdf and /dev/null differ diff --git a/graphfleet/notebooks/Global Search Notebook.ipynb b/graphfleet/notebooks/Global Search Notebook.ipynb deleted file mode 100644 index db141e0fe..000000000 --- a/graphfleet/notebooks/Global Search Notebook.ipynb +++ /dev/null @@ -1,254 +0,0 @@ -{ - "cells": [ - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "# Copyright (c) 2024 Microsoft Corporation.\n", - "# Licensed under the MIT License." - ] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "import os\n", - "\n", - "import pandas as pd\n", - "import tiktoken\n", - "\n", - "from graphrag.query.indexer_adapters import read_indexer_entities, read_indexer_reports\n", - "from graphrag.query.llm.oai.chat_openai import ChatOpenAI\n", - "from graphrag.query.llm.oai.typing import OpenaiApiType\n", - "from graphrag.query.structured_search.global_search.community_context import (\n", - " GlobalCommunityContext,\n", - ")\n", - "from graphrag.query.structured_search.global_search.search import GlobalSearch" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "## Global Search example\n", - "\n", - "Global search method generates answers by searching over all AI-generated community reports in a map-reduce fashion. This is a resource-intensive method, but often gives good responses for questions that require an understanding of the dataset as a whole (e.g. What are the most significant values of the herbs mentioned in this notebook?)." - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "### LLM setup" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "from sys import api_version\n", - "\n", - "\n", - "api_key = os.environ[\"GRAPHRAG_API_KEY\"]\n", - "llm_model = os.environ[\"GRAPHRAG_LLM_MODEL\"]\n", - "api_version = \"2023-05-15\"\n", - "api_base = \"https://gpt-4o-fr.openai.azure.com\"\n", - "llm = ChatOpenAI(\n", - " api_key=api_key,\n", - " model=llm_model,\n", - " api_type=OpenaiApiType.AzureOpenAI, # OpenaiApiType.OpenAI or OpenaiApiType.AzureOpenAI\n", - " max_retries=20,\n", - " api_version=api_version,\n", - " api_base=api_base\n", - ")\n", - "\n", - "token_encoder = tiktoken.get_encoding(\"cl100k_base\")" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "### Load community reports as context for global search\n", - "\n", - "- Load all community reports in the `create_final_community_reports` table from the ire-indexing engine, to be used as context data for global search.\n", - "- Load entities from the `create_final_nodes` and `create_final_entities` tables from the ire-indexing engine, to be used for calculating community weights for context ranking. Note that this is optional (if no entities are provided, we will not calculate community weights and only use the `rank` attribute in the community reports table for context ranking)" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "# parquet files generated from indexing pipeline\n", - "INPUT_DIR = \"../graphfleet/output/usage_ai/artifacts\"\n", - "COMMUNITY_REPORT_TABLE = \"create_final_community_reports\"\n", - "ENTITY_TABLE = \"create_final_nodes\"\n", - "ENTITY_EMBEDDING_TABLE = \"create_final_entities\"\n", - "\n", - "# community level in the Leiden community hierarchy from which we will load the community reports\n", - "# higher value means we use reports from more fine-grained communities (at the cost of higher computation cost)\n", - "COMMUNITY_LEVEL = 2" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "entity_df = pd.read_parquet(f\"{INPUT_DIR}/{ENTITY_TABLE}.parquet\")\n", - "report_df = pd.read_parquet(f\"{INPUT_DIR}/{COMMUNITY_REPORT_TABLE}.parquet\")\n", - "entity_embedding_df = pd.read_parquet(f\"{INPUT_DIR}/{ENTITY_EMBEDDING_TABLE}.parquet\")\n", - "\n", - "reports = read_indexer_reports(report_df, entity_df, COMMUNITY_LEVEL)\n", - "entities = read_indexer_entities(entity_df, entity_embedding_df, COMMUNITY_LEVEL)\n", - "print(f\"Total report count: {len(report_df)}\")\n", - "print(\n", - " f\"Report count after filtering by community level {COMMUNITY_LEVEL}: {len(reports)}\"\n", - ")\n", - "report_df.head()" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "#### Build global context based on community reports" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "context_builder = GlobalCommunityContext(\n", - " community_reports=reports,\n", - " entities=entities, # default to None if you don't want to use community weights for ranking\n", - " token_encoder=token_encoder,\n", - ")" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "#### Perform global search" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "context_builder_params = {\n", - " \"use_community_summary\": False, # False means using full community reports. True means using community short summaries.\n", - " \"shuffle_data\": True,\n", - " \"include_community_rank\": True,\n", - " \"min_community_rank\": 0,\n", - " \"community_rank_name\": \"rank\",\n", - " \"include_community_weight\": True,\n", - " \"community_weight_name\": \"occurrence weight\",\n", - " \"normalize_community_weight\": True,\n", - " \"max_tokens\": 12_000, # change this based on the token limit you have on your model (if you are using a model with 8k limit, a good setting could be 5000)\n", - " \"context_name\": \"Reports\",\n", - "}\n", - "\n", - "map_llm_params = {\n", - " \"max_tokens\": 1000,\n", - " \"temperature\": 0.0,\n", - " \"response_format\": {\"type\": \"json_object\"},\n", - "}\n", - "\n", - "reduce_llm_params = {\n", - " \"max_tokens\": 2000, # change this based on the token limit you have on your model (if you are using a model with 8k limit, a good setting could be 1000-1500)\n", - " \"temperature\": 0.0,\n", - "}" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "search_engine = GlobalSearch(\n", - " llm=llm,\n", - " context_builder=context_builder,\n", - " token_encoder=token_encoder,\n", - " max_data_tokens=12_000, # change this based on the token limit you have on your model (if you are using a model with 8k limit, a good setting could be 5000)\n", - " map_llm_params=map_llm_params,\n", - " reduce_llm_params=reduce_llm_params,\n", - " allow_general_knowledge=False, # set this to True will add instruction to encourage the LLM to incorporate general knowledge in the response, which may increase hallucinations, but could be useful in some use cases.\n", - " json_mode=True, # set this to False if your LLM model does not support JSON mode.\n", - " context_builder_params=context_builder_params,\n", - " concurrent_coroutines=32,\n", - " response_type=\"multiple paragraphs\", # free form text describing the response type and format, can be anything, e.g. prioritized list, single paragraph, multiple paragraphs, multiple-page report\n", - ")" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "result = await search_engine.asearch(\n", - " \"What is the major insights to remember ?\"\n", - ")\n", - "\n", - "print(result.response)" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "# inspect the data used to build the context for the LLM responses\n", - "result.context_data[\"reports\"]" - ] - }, - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "# inspect number of LLM calls and tokens\n", - "print(f\"LLM calls: {result.llm_calls}. LLM tokens: {result.prompt_tokens}\")" - ] - } - ], - "metadata": { - "kernelspec": { - "display_name": "Python 3", - "language": "python", - "name": "python3" - }, - "language_info": { - "codemirror_mode": { - "name": "ipython", - "version": 3 - }, - "file_extension": ".py", - "mimetype": "text/x-python", - "name": "python", - "nbconvert_exporter": "python", - "pygments_lexer": "ipython3", - "version": "3.11.9" - } - }, - "nbformat": 4, - "nbformat_minor": 2 -} diff --git a/graphfleet/notebooks/Local Search Notebook.ipynb b/graphfleet/notebooks/Local Search Notebook.ipynb deleted file mode 100644 index 271a48800..000000000 --- a/graphfleet/notebooks/Local Search Notebook.ipynb +++ /dev/null @@ -1,1674 +0,0 @@ -{ - "cells": [ - { - "cell_type": "code", - "execution_count": null, - "metadata": {}, - "outputs": [], - "source": [ - "# Copyright (c) 2024 Microsoft Corporation.\n", - "# Licensed under the MIT License." - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, -======= - "execution_count": 33, ->>>>>>> origin/main - "metadata": {}, - "outputs": [], - "source": [ - "import os\n", - "\n", - "import pandas as pd\n", - "import tiktoken\n", - "\n", - "from graphrag.query.context_builder.entity_extraction import EntityVectorStoreKey\n", - "from graphrag.query.indexer_adapters import (\n", - " read_indexer_covariates,\n", - " read_indexer_entities,\n", - " read_indexer_relationships,\n", - " read_indexer_reports,\n", - " read_indexer_text_units,\n", - ")\n", - "from graphrag.query.input.loaders.dfs import (\n", - " store_entity_semantic_embeddings,\n", - ")\n", - "from graphrag.query.llm.oai.chat_openai import ChatOpenAI\n", - "from graphrag.query.llm.oai.embedding import OpenAIEmbedding\n", - "from graphrag.query.llm.oai.typing import OpenaiApiType\n", - "from graphrag.query.question_gen.local_gen import LocalQuestionGen\n", - "from graphrag.query.structured_search.local_search.mixed_context import (\n", - " LocalSearchMixedContext,\n", - ")\n", - "from graphrag.query.structured_search.local_search.search import LocalSearch\n", - "from graphrag.vector_stores.lancedb import LanceDBVectorStore" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "## Local Search Example\n", - "\n", - "Local search method generates answers by combining relevant data from the AI-extracted knowledge-graph with text chunks of the raw documents. This method is suitable for questions that require an understanding of specific entities mentioned in the documents (e.g. What are the healing properties of chamomile?)." - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "### Load text units and graph data tables as context for local search\n", - "\n", - "- In this test we first load indexing outputs from parquet files to dataframes, then convert these dataframes into collections of data objects aligning with the knowledge model." - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "### Load tables to dataframes" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, -======= - "execution_count": 34, ->>>>>>> origin/main - "metadata": {}, - "outputs": [], - "source": [ - "INPUT_DIR = \"../output/usage_ai/artifacts\"\n", - "LANCEDB_URI = f\"{INPUT_DIR}/lancedb\"\n", - "\n", - "COMMUNITY_REPORT_TABLE = \"create_final_community_reports\"\n", - "ENTITY_TABLE = \"create_final_nodes\"\n", - "ENTITY_EMBEDDING_TABLE = \"create_final_entities\"\n", - "RELATIONSHIP_TABLE = \"create_final_relationships\"\n", - "COVARIATE_TABLE = \"create_final_covariates\"\n", - "TEXT_UNIT_TABLE = \"create_final_text_units\"\n", - "COMMUNITY_LEVEL = 2" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "#### Read entities" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 35, - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Entity count: 1917\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "[2024-08-08T13:13:06Z WARN lance::dataset] No existing dataset at /Volumes/Samsung-SSD-T7/Qredence/GraphFleet/GraphFleet1/graphfleet/notebooks/../output/usage_ai/artifacts/lancedb/entity_description_embeddings.lance, it will be created\n" - ] - }, - { - "data": { - "text/html": [ - "
\n", - "\n", - "\n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - "
leveltitletypedescriptionsource_idcommunitydegreehuman_readable_ididsizegraph_embeddingentity_typetop_level_node_idxy
00MICROSOFTORGANIZATIONMicrosoft is a technology company that conduct...0323a135b8299e84a110f1aa06e6b4a6,204b433ad19e0...2250b45241d70f0e43fca764df95b2b81f7725.0[-0.06957332044839859, -0.04348175972700119, -...Noneb45241d70f0e43fca764df95b2b81f776.419162-6.675766
10SECOND MICROSOFT REPORT ON AI AND PRODUCTIVITY...DOCUMENTThe Second Microsoft Report on AI and Producti...644d7a7f0fa5b1428c1fafe65258b2c0135814119fd06010c494caa07f439b333f4c558.0[-0.12563732266426086, 0.07880357652902603, -0...None4119fd06010c494caa07f439b333f4c5-1.3105791.134900
20SONIA JAFFEPERSONSonia Jaffe is a researcher prominently involv...644d7a7f0fa5b1428c1fafe65258b2c0,9588396616caa...422d3835bf3dda84ead99deadbeac5d0d7d2.0[-0.08217236399650574, 0.0687311664223671, -0....Noned3835bf3dda84ead99deadbeac5d0d7d2.364835-5.030548
30NEHA PARIKH SHAHPERSONNeha Parikh Shah is one of the editors of the ...644d7a7f0fa5b1428c1fafe65258b2c01313077d2820ae1845bcbb1803379a3d1eae1.0[-0.07571760565042496, 0.05104828625917435, -0...None077d2820ae1845bcbb1803379a3d1eae-0.9220841.662963
40JENNA BUTLERPERSONJenna Butler is one of the editors of the Seco...644d7a7f0fa5b1428c1fafe65258b2c013143671ea0dd4e84c1a9b02c5ab2c8f4bac1.0[-0.07471488416194916, 0.06074602156877518, -0...None3671ea0dd4e84c1a9b02c5ab2c8f4bac-1.0440981.466650
\n", - "
" - ], - "text/plain": [ - " level title type \\\n", - "0 0 MICROSOFT ORGANIZATION \n", - "1 0 SECOND MICROSOFT REPORT ON AI AND PRODUCTIVITY... DOCUMENT \n", - "2 0 SONIA JAFFE PERSON \n", - "3 0 NEHA PARIKH SHAH PERSON \n", - "4 0 JENNA BUTLER PERSON \n", - "\n", - " description \\\n", - "0 Microsoft is a technology company that conduct... \n", - "1 The Second Microsoft Report on AI and Producti... \n", - "2 Sonia Jaffe is a researcher prominently involv... \n", - "3 Neha Parikh Shah is one of the editors of the ... \n", - "4 Jenna Butler is one of the editors of the Seco... \n", - "\n", - " source_id community degree \\\n", - "0 0323a135b8299e84a110f1aa06e6b4a6,204b433ad19e0... 2 25 \n", - "1 644d7a7f0fa5b1428c1fafe65258b2c0 13 58 \n", - "2 644d7a7f0fa5b1428c1fafe65258b2c0,9588396616caa... 4 2 \n", - "3 644d7a7f0fa5b1428c1fafe65258b2c0 13 1 \n", - "4 644d7a7f0fa5b1428c1fafe65258b2c0 13 1 \n", - "\n", - " human_readable_id id size \\\n", - "0 0 b45241d70f0e43fca764df95b2b81f77 25.0 \n", - "1 1 4119fd06010c494caa07f439b333f4c5 58.0 \n", - "2 2 d3835bf3dda84ead99deadbeac5d0d7d 2.0 \n", - "3 3 077d2820ae1845bcbb1803379a3d1eae 1.0 \n", - "4 4 3671ea0dd4e84c1a9b02c5ab2c8f4bac 1.0 \n", - "\n", - " graph_embedding entity_type \\\n", - "0 [-0.06957332044839859, -0.04348175972700119, -... None \n", - "1 [-0.12563732266426086, 0.07880357652902603, -0... None \n", - "2 [-0.08217236399650574, 0.0687311664223671, -0.... None \n", - "3 [-0.07571760565042496, 0.05104828625917435, -0... None \n", - "4 [-0.07471488416194916, 0.06074602156877518, -0... None \n", - "\n", - " top_level_node_id x y \n", - "0 b45241d70f0e43fca764df95b2b81f77 6.419162 -6.675766 \n", - "1 4119fd06010c494caa07f439b333f4c5 -1.310579 1.134900 \n", - "2 d3835bf3dda84ead99deadbeac5d0d7d 2.364835 -5.030548 \n", - "3 077d2820ae1845bcbb1803379a3d1eae -0.922084 1.662963 \n", - "4 3671ea0dd4e84c1a9b02c5ab2c8f4bac -1.044098 1.466650 " - ] - }, - "execution_count": 35, - "metadata": {}, - "output_type": "execute_result" - } - ], ->>>>>>> origin/main - "source": [ - "# read nodes table to get community and degree data\n", - "entity_df = pd.read_parquet(f\"{INPUT_DIR}/{ENTITY_TABLE}.parquet\")\n", - "entity_embedding_df = pd.read_parquet(f\"{INPUT_DIR}/{ENTITY_EMBEDDING_TABLE}.parquet\")\n", - "\n", - "entities = read_indexer_entities(entity_df, entity_embedding_df, COMMUNITY_LEVEL)\n", - "\n", - "# load description embeddings to an in-memory lancedb vectorstore\n", - "# to connect to a remote db, specify url and port values.\n", - "description_embedding_store = LanceDBVectorStore(\n", - " collection_name=\"entity_description_embeddings\",\n", - ")\n", - "description_embedding_store.connect(db_uri=LANCEDB_URI)\n", - "entity_description_embeddings = store_entity_semantic_embeddings(\n", - " entities=entities, vectorstore=description_embedding_store\n", - ")\n", - "\n", - "print(f\"Entity count: {len(entity_df)}\")\n", - "entity_df.head()" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "#### Read relationships" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 36, - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Relationship count: 869\n" - ] - }, - { - "data": { - "text/html": [ - "
\n", - "\n", - "\n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - "
sourcetargetweightdescriptiontext_unit_idsidhuman_readable_idsource_degreetarget_degreerank
0MICROSOFTSECOND MICROSOFT REPORT ON AI AND PRODUCTIVITY...2.0Microsoft released the Second Microsoft Report...[644d7a7f0fa5b1428c1fafe65258b2c0]b823ba1bfe944fa9887edd8faf8a5f170255883
1MICROSOFTAI AND PRODUCTIVITY REPORT1.0Microsoft released the AI and Productivity Report[204b433ad19e035991efb2df3a12772a]d0bfb473fdc64643954cdb4675e2f3891251641
2MICROSOFTSECOND MICROSOFT REPORT1.0Microsoft released the Second Microsoft Report...[204b433ad19e035991efb2df3a12772a]a4db1b2a9c3e4d2d838725f8166c36b42251136
3MICROSOFTSECOND MICROSOFT AI AND PRODUCTIVITY REPORT1.0Microsoft released the Second AI and Productiv...[bffea0de28de23b824ababad8ebd6b67]8dae140578c841ae9373cbc607c4a6e6325631
4MICROSOFTCOMMERCIAL INTEREST1.0Microsoft has a commercial interest in demonst...[bffea0de28de23b824ababad8ebd6b67]b215cc33cf40434f87f284ff8f3506a4425126
\n", - "
" - ], - "text/plain": [ - " source target weight \\\n", - "0 MICROSOFT SECOND MICROSOFT REPORT ON AI AND PRODUCTIVITY... 2.0 \n", - "1 MICROSOFT AI AND PRODUCTIVITY REPORT 1.0 \n", - "2 MICROSOFT SECOND MICROSOFT REPORT 1.0 \n", - "3 MICROSOFT SECOND MICROSOFT AI AND PRODUCTIVITY REPORT 1.0 \n", - "4 MICROSOFT COMMERCIAL INTEREST 1.0 \n", - "\n", - " description \\\n", - "0 Microsoft released the Second Microsoft Report... \n", - "1 Microsoft released the AI and Productivity Report \n", - "2 Microsoft released the Second Microsoft Report... \n", - "3 Microsoft released the Second AI and Productiv... \n", - "4 Microsoft has a commercial interest in demonst... \n", - "\n", - " text_unit_ids id \\\n", - "0 [644d7a7f0fa5b1428c1fafe65258b2c0] b823ba1bfe944fa9887edd8faf8a5f17 \n", - "1 [204b433ad19e035991efb2df3a12772a] d0bfb473fdc64643954cdb4675e2f389 \n", - "2 [204b433ad19e035991efb2df3a12772a] a4db1b2a9c3e4d2d838725f8166c36b4 \n", - "3 [bffea0de28de23b824ababad8ebd6b67] 8dae140578c841ae9373cbc607c4a6e6 \n", - "4 [bffea0de28de23b824ababad8ebd6b67] b215cc33cf40434f87f284ff8f3506a4 \n", - "\n", - " human_readable_id source_degree target_degree rank \n", - "0 0 25 58 83 \n", - "1 1 25 16 41 \n", - "2 2 25 11 36 \n", - "3 3 25 6 31 \n", - "4 4 25 1 26 " - ] - }, - "execution_count": 36, - "metadata": {}, - "output_type": "execute_result" - } - ], ->>>>>>> origin/main - "source": [ - "relationship_df = pd.read_parquet(f\"{INPUT_DIR}/{RELATIONSHIP_TABLE}.parquet\")\n", - "relationships = read_indexer_relationships(relationship_df)\n", - "\n", - "print(f\"Relationship count: {len(relationship_df)}\")\n", - "relationship_df.head()" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 37, - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Claim records: 174\n" - ] - } - ], ->>>>>>> origin/main - "source": [ - "# NOTE: covariates are turned off by default, because they generally need prompt tuning to be valuable\n", - "# Please see the GRAPHRAG_CLAIM_* settings\n", - "covariate_df = pd.read_parquet(f\"{INPUT_DIR}/{COVARIATE_TABLE}.parquet\")\n", - "\n", - "claims = read_indexer_covariates(covariate_df)\n", - "\n", - "print(f\"Claim records: {len(claims)}\")\n", - "covariates = {\"claims\": claims}" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "#### Read community reports" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 38, - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Report records: 106\n" - ] - }, - { - "data": { - "text/html": [ - "
\n", - "\n", - "\n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - "
communityfull_contentlevelranktitlerank_explanationsummaryfindingsfull_content_jsonid
0100# Random Forest and AI Power Users\\n\\nThe comm...28.5Random Forest and AI Power UsersThe impact rating is high due to the significa...The community revolves around the Random Fores...[{'explanation': 'The Random Forest model is t...{\\n \"title\": \"Random Forest and AI Power Us...3e0fa18f-216a-4095-a1d4-1bce2299771a
1101# Predictive Models and Model Performance in A...29.0Predictive Models and Model Performance in AIThe rating is high due to the significant impa...The community is centered around the use of pr...[{'explanation': 'Predictive models are centra...{\\n \"title\": \"Predictive Models and Model P...33c467c6-d236-4f62-b28b-9b8b5ac948dc
2102# Second Microsoft Report on AI and Productivi...29.0Second Microsoft Report on AI and Productivity...The report is highly significant as it provide...The community centers around the Second Micros...[{'explanation': 'The Second Microsoft Report ...{\\n \"title\": \"Second Microsoft Report on AI...8d54fa21-ec89-46da-b4ae-8ce2de2f885e
3103# AI and Productivity Research Community: Key ...29.0AI and Productivity Research Community: Key Co...The rating is high due to the significant impa...The community centers around the research cont...[{'explanation': 'Brian Houck is a contributin...{\\n \"title\": \"AI and Productivity Research ...360b2cc1-d0f8-456e-b0bf-aed193e8abd8
4104# Impact of Copilot for Security on Workplace ...29.0Impact of Copilot for Security on Workplace Pr...The rating is high due to the comprehensive re...The community centers around the research and ...[{'explanation': 'Ben Edelman, James Bono, Rob...{\\n \"title\": \"Impact of Copilot for Securit...824c64e1-ee24-4de9-9a7a-25e267092faf
\n", - "
" - ], - "text/plain": [ - " community full_content level rank \\\n", - "0 100 # Random Forest and AI Power Users\\n\\nThe comm... 2 8.5 \n", - "1 101 # Predictive Models and Model Performance in A... 2 9.0 \n", - "2 102 # Second Microsoft Report on AI and Productivi... 2 9.0 \n", - "3 103 # AI and Productivity Research Community: Key ... 2 9.0 \n", - "4 104 # Impact of Copilot for Security on Workplace ... 2 9.0 \n", - "\n", - " title \\\n", - "0 Random Forest and AI Power Users \n", - "1 Predictive Models and Model Performance in AI \n", - "2 Second Microsoft Report on AI and Productivity... \n", - "3 AI and Productivity Research Community: Key Co... \n", - "4 Impact of Copilot for Security on Workplace Pr... \n", - "\n", - " rank_explanation \\\n", - "0 The impact rating is high due to the significa... \n", - "1 The rating is high due to the significant impa... \n", - "2 The report is highly significant as it provide... \n", - "3 The rating is high due to the significant impa... \n", - "4 The rating is high due to the comprehensive re... \n", - "\n", - " summary \\\n", - "0 The community revolves around the Random Fores... \n", - "1 The community is centered around the use of pr... \n", - "2 The community centers around the Second Micros... \n", - "3 The community centers around the research cont... \n", - "4 The community centers around the research and ... \n", - "\n", - " findings \\\n", - "0 [{'explanation': 'The Random Forest model is t... \n", - "1 [{'explanation': 'Predictive models are centra... \n", - "2 [{'explanation': 'The Second Microsoft Report ... \n", - "3 [{'explanation': 'Brian Houck is a contributin... \n", - "4 [{'explanation': 'Ben Edelman, James Bono, Rob... \n", - "\n", - " full_content_json \\\n", - "0 {\\n \"title\": \"Random Forest and AI Power Us... \n", - "1 {\\n \"title\": \"Predictive Models and Model P... \n", - "2 {\\n \"title\": \"Second Microsoft Report on AI... \n", - "3 {\\n \"title\": \"AI and Productivity Research ... \n", - "4 {\\n \"title\": \"Impact of Copilot for Securit... \n", - "\n", - " id \n", - "0 3e0fa18f-216a-4095-a1d4-1bce2299771a \n", - "1 33c467c6-d236-4f62-b28b-9b8b5ac948dc \n", - "2 8d54fa21-ec89-46da-b4ae-8ce2de2f885e \n", - "3 360b2cc1-d0f8-456e-b0bf-aed193e8abd8 \n", - "4 824c64e1-ee24-4de9-9a7a-25e267092faf " - ] - }, - "execution_count": 38, - "metadata": {}, - "output_type": "execute_result" - } - ], ->>>>>>> origin/main - "source": [ - "report_df = pd.read_parquet(f\"{INPUT_DIR}/{COMMUNITY_REPORT_TABLE}.parquet\")\n", - "reports = read_indexer_reports(report_df, entity_df, COMMUNITY_LEVEL)\n", - "\n", - "print(f\"Report records: {len(report_df)}\")\n", - "report_df.head()" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "#### Read text units" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 39, - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Text unit records: 31\n" - ] - }, - { - "data": { - "text/html": [ - "
\n", - "\n", - "\n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - "
idtextn_tokensdocument_idsentity_idsrelationship_idscovariate_ids
0644d7a7f0fa5b1428c1fafe65258b2c0Generative AI in Real -World Workplaces \\nThe...512[ba2768bc8ec3c85ee75905597df151a2][b45241d70f0e43fca764df95b2b81f77, 4119fd06010...[b823ba1bfe944fa9887edd8faf8a5f17, bbdd53a15e9...[8c839725-9e50-435b-8397-f093e925ae69, f159c26...
1204b433ad19e035991efb2df3a12772atheir day -to-day jobs in significant \\nways....512[ba2768bc8ec3c85ee75905597df151a2][b45241d70f0e43fca764df95b2b81f77, 3b6cd96a273...[d0bfb473fdc64643954cdb4675e2f389, a4db1b2a9c3...[b95374b6-2443-4b90-bfb3-4812eb3e06df, c5ffa39...
2bffea0de28de23b824ababad8ebd6b67impact of these tools outside \\nof a lab sett...512[ba2768bc8ec3c85ee75905597df151a2][b45241d70f0e43fca764df95b2b81f77, 958beecdb5b...[8dae140578c841ae9373cbc607c4a6e6, b215cc33cf4...[248bf155-142e-4fc2-9e20-ea801f2b6891, 7e3f474...
392fe97fc068c126408255afd4e8f8eecon productivity in real-world contexts. This ...512[ba2768bc8ec3c85ee75905597df151a2][3b6cd96a27304614850709aba1c9598b, babe97e1d97...[49e24b5f2c1d40d7857afe327db4f554, 587f39a32e9...[6b66c008-6223-49ea-97c2-f7232f2d2378, 1f0eb3b...
4f4be7740240a148cb9de9642560bfa54but also changing which tasks people choose t...512[ba2768bc8ec3c85ee75905597df151a2][b45241d70f0e43fca764df95b2b81f77, d3835bf3dda...[c1ff9d8e1b8745d6860c34ce26122d79, 1d7b0deca76...[fbf94fad-2bf0-443b-9e55-c5dff0ce6c91, 3cfe2c0...
\n", - "
" - ], - "text/plain": [ - " id \\\n", - "0 644d7a7f0fa5b1428c1fafe65258b2c0 \n", - "1 204b433ad19e035991efb2df3a12772a \n", - "2 bffea0de28de23b824ababad8ebd6b67 \n", - "3 92fe97fc068c126408255afd4e8f8eec \n", - "4 f4be7740240a148cb9de9642560bfa54 \n", - "\n", - " text n_tokens \\\n", - "0 Generative AI in Real -World Workplaces \\nThe... 512 \n", - "1 their day -to-day jobs in significant \\nways.... 512 \n", - "2 impact of these tools outside \\nof a lab sett... 512 \n", - "3 on productivity in real-world contexts. This ... 512 \n", - "4 but also changing which tasks people choose t... 512 \n", - "\n", - " document_ids \\\n", - "0 [ba2768bc8ec3c85ee75905597df151a2] \n", - "1 [ba2768bc8ec3c85ee75905597df151a2] \n", - "2 [ba2768bc8ec3c85ee75905597df151a2] \n", - "3 [ba2768bc8ec3c85ee75905597df151a2] \n", - "4 [ba2768bc8ec3c85ee75905597df151a2] \n", - "\n", - " entity_ids \\\n", - "0 [b45241d70f0e43fca764df95b2b81f77, 4119fd06010... \n", - "1 [b45241d70f0e43fca764df95b2b81f77, 3b6cd96a273... \n", - "2 [b45241d70f0e43fca764df95b2b81f77, 958beecdb5b... \n", - "3 [3b6cd96a27304614850709aba1c9598b, babe97e1d97... \n", - "4 [b45241d70f0e43fca764df95b2b81f77, d3835bf3dda... \n", - "\n", - " relationship_ids \\\n", - "0 [b823ba1bfe944fa9887edd8faf8a5f17, bbdd53a15e9... \n", - "1 [d0bfb473fdc64643954cdb4675e2f389, a4db1b2a9c3... \n", - "2 [8dae140578c841ae9373cbc607c4a6e6, b215cc33cf4... \n", - "3 [49e24b5f2c1d40d7857afe327db4f554, 587f39a32e9... \n", - "4 [c1ff9d8e1b8745d6860c34ce26122d79, 1d7b0deca76... \n", - "\n", - " covariate_ids \n", - "0 [8c839725-9e50-435b-8397-f093e925ae69, f159c26... \n", - "1 [b95374b6-2443-4b90-bfb3-4812eb3e06df, c5ffa39... \n", - "2 [248bf155-142e-4fc2-9e20-ea801f2b6891, 7e3f474... \n", - "3 [6b66c008-6223-49ea-97c2-f7232f2d2378, 1f0eb3b... \n", - "4 [fbf94fad-2bf0-443b-9e55-c5dff0ce6c91, 3cfe2c0... " - ] - }, - "execution_count": 39, - "metadata": {}, - "output_type": "execute_result" - } - ], ->>>>>>> origin/main - "source": [ - "text_unit_df = pd.read_parquet(f\"{INPUT_DIR}/{TEXT_UNIT_TABLE}.parquet\")\n", - "text_units = read_indexer_text_units(text_unit_df)\n", - "\n", - "print(f\"Text unit records: {len(text_unit_df)}\")\n", - "text_unit_df.head()" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, -======= - "execution_count": 42, ->>>>>>> origin/main - "metadata": {}, - "outputs": [], - "source": [ - "api_key = os.environ[\"GRAPHRAG_API_KEY\"]\n", - "llm_model = os.environ[\"GRAPHRAG_LLM_MODEL\"]\n", - "embedding_model = os.environ[\"GRAPHRAG_EMBEDDING_MODEL\"]\n", - "api_version = \"2023-05-15\"\n", - "api_base = \"https://gpt-4o-fr.openai.azure.com\"\n", - "\n", - "llm = ChatOpenAI(\n", - " api_key=api_key,\n", - " model=llm_model,\n", - " api_type=OpenaiApiType.AzureOpenAI, # OpenaiApiType.OpenAI or OpenaiApiType.AzureOpenAI\n", - " max_retries=20,\n", - " api_base=api_base,\n", - " api_version=api_version\n", - ")\n", - "\n", - "token_encoder = tiktoken.get_encoding(\"cl100k_base\")\n", - "\n", - "text_embedder = OpenAIEmbedding(\n", - " api_key=api_key,\n", - " api_base=api_base,\n", - " api_type=OpenaiApiType.AzureOpenAI,\n", - " model=embedding_model,\n", - " deployment_name=embedding_model,\n", - " max_retries=20,\n", - " api_version=api_version\n", - ")" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "### Create local search context builder" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, -======= - "execution_count": 43, ->>>>>>> origin/main - "metadata": {}, - "outputs": [], - "source": [ - "context_builder = LocalSearchMixedContext(\n", - " community_reports=reports,\n", - " text_units=text_units,\n", - " entities=entities,\n", - " relationships=relationships,\n", - " # if you did not run covariates during indexing, set this to None\n", - " covariates=covariates,\n", - " entity_text_embeddings=description_embedding_store,\n", - " embedding_vectorstore_key=EntityVectorStoreKey.ID, # if the vectorstore uses entity title as ids, set this to EntityVectorStoreKey.TITLE\n", - " text_embedder=text_embedder,\n", - " token_encoder=token_encoder,\n", - ")" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "### Create local search engine" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, -======= - "execution_count": 44, ->>>>>>> origin/main - "metadata": {}, - "outputs": [], - "source": [ - "# text_unit_prop: proportion of context window dedicated to related text units\n", - "# community_prop: proportion of context window dedicated to community reports.\n", - "# The remaining proportion is dedicated to entities and relationships. Sum of text_unit_prop and community_prop should be <= 1\n", - "# conversation_history_max_turns: maximum number of turns to include in the conversation history.\n", - "# conversation_history_user_turns_only: if True, only include user queries in the conversation history.\n", - "# top_k_mapped_entities: number of related entities to retrieve from the entity description embedding store.\n", - "# top_k_relationships: control the number of out-of-network relationships to pull into the context window.\n", - "# include_entity_rank: if True, include the entity rank in the entity table in the context window. Default entity rank = node degree.\n", - "# include_relationship_weight: if True, include the relationship weight in the context window.\n", - "# include_community_rank: if True, include the community rank in the context window.\n", - "# return_candidate_context: if True, return a set of dataframes containing all candidate entity/relationship/covariate records that\n", - "# could be relevant. Note that not all of these records will be included in the context window. The \"in_context\" column in these\n", - "# dataframes indicates whether the record is included in the context window.\n", - "# max_tokens: maximum number of tokens to use for the context window.\n", - "\n", - "\n", - "local_context_params = {\n", - " \"text_unit_prop\": 0.5,\n", - " \"community_prop\": 0.1,\n", - " \"conversation_history_max_turns\": 5,\n", - " \"conversation_history_user_turns_only\": True,\n", - " \"top_k_mapped_entities\": 10,\n", - " \"top_k_relationships\": 10,\n", - " \"include_entity_rank\": True,\n", - " \"include_relationship_weight\": True,\n", - " \"include_community_rank\": False,\n", - " \"return_candidate_context\": False,\n", - " \"embedding_vectorstore_key\": EntityVectorStoreKey.ID, # set this to EntityVectorStoreKey.TITLE if the vectorstore uses entity title as ids\n", - " \"max_tokens\": 12_000, # change this based on the token limit you have on your model (if you are using a model with 8k limit, a good setting could be 5000)\n", - "}\n", - "\n", - "llm_params = {\n", - " \"max_tokens\": 2_000, # change this based on the token limit you have on your model (if you are using a model with 8k limit, a good setting could be 1000=1500)\n", - " \"temperature\": 0.0,\n", - "}" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, -======= - "execution_count": 45, ->>>>>>> origin/main - "metadata": {}, - "outputs": [], - "source": [ - "search_engine = LocalSearch(\n", - " llm=llm,\n", - " context_builder=context_builder,\n", - " token_encoder=token_encoder,\n", - " llm_params=llm_params,\n", - " context_builder_params=local_context_params,\n", - " response_type=\"multiple paragraphs\", # free form text describing the response type and format, can be anything, e.g. prioritized list, single paragraph, multiple paragraphs, multiple-page report\n", - ")" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "### Run local search on sample queries" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 46, - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "# How Companies Use AI\n", - "\n", - "Artificial Intelligence (AI) has become an integral part of modern business operations, offering a range of applications that enhance productivity, innovation, and overall efficiency. Companies across various industries are leveraging AI to automate routine tasks, analyze data, and improve decision-making processes. Below, we explore some of the key ways in which companies are utilizing AI.\n", - "\n", - "## Enhancing Developer Productivity\n", - "\n", - "One of the primary applications of AI in companies is to support software developers. AI tools can automate routine tasks, generate unit tests, write documentation, and analyze code for defects and optimizations. This support significantly enhances productivity by allowing developers to focus on more complex and creative aspects of their work. For instance, AI-driven code analysis can help identify and fix defects more efficiently, thereby improving the quality of the software [Data: AI Support for Developers and Its Impact on Productivity (39); Entities (299, 303, 304); Relationships (230, 616, 617, 618)].\n", - "\n", - "## Opportunities with Generative AI\n", - "\n", - "Generative AI presents numerous opportunities for companies to innovate and improve workflows. AI can generate new code snippets, suggest optimizations, and provide innovative solutions that developers might not have considered. This fosters a culture of innovation and continuous improvement within the organization. The relationship between generative AI and opportunities underscores the positive impact AI can have on the development process [Data: AI Support for Developers and Its Impact on Productivity (39); Entities (310); Relationships (206, 622, 628)].\n", - "\n", - "## Addressing Concerns and Challenges\n", - "\n", - "Despite the benefits, the integration of AI in companies comes with significant concerns. These include the potential for defects or vulnerabilities in AI-generated code, the need for validation and human oversight, and the impact on job security. Developers worry that over-reliance on AI could lead to a decrease in the quality of code and introduce security risks. Therefore, it is crucial for companies to implement robust validation processes and ensure that human oversight is maintained to mitigate these risks [Data: AI Support for Developers and Its Impact on Productivity (39); Entities (311, 320); Relationships (620, 623, 627)].\n", - "\n", - "## Importance of Training Programs\n", - "\n", - "Training programs are essential for helping employees effectively integrate AI into their workflows. These programs can address employees' concerns, improve their understanding of AI tools, and ensure that they are used effectively. By providing comprehensive training, companies can help employees build trust in AI tools and enhance their overall satisfaction and productivity. Training programs also help employees stay updated with the latest advancements in AI technology [Data: AI Support for Developers and Its Impact on Productivity (39); Entities (308); Relationships (621)].\n", - "\n", - "## Role of AI in Customer Service\n", - "\n", - "AI is also widely used in customer service to assist agents in resolving issues more efficiently. Conversational assistants, for example, can handle routine inquiries, allowing human agents to focus on more complex customer issues. This not only improves the efficiency of customer service operations but also enhances customer satisfaction by providing quicker and more accurate responses [Data: Entities (114)].\n", - "\n", - "## Building Trust in AI Tools\n", - "\n", - "Trust is a critical factor for the successful adoption of AI tools among employees. Companies need to ensure that AI-generated outputs are accurate and beneficial, addressing concerns about defects or vulnerabilities, and providing transparency in how AI tools operate. Building trust involves ensuring that AI tools are perceived as helpful and trustworthy, which is essential for their widespread adoption and effective use [Data: AI Support for Developers and Its Impact on Productivity (39); Entities (322); Relationships (629)].\n", - "\n", - "## Conclusion\n", - "\n", - "In summary, companies are leveraging AI to enhance productivity, foster innovation, and improve customer service. However, the successful integration of AI requires addressing concerns related to validation, human oversight, and job security. Comprehensive training programs and trust-building measures are essential to maximize the benefits of AI and ensure its effective use in the workplace. As AI technology continues to evolve, companies that effectively integrate these tools into their operations will likely see significant improvements in efficiency and innovation.\n" - ] - } - ], ->>>>>>> origin/main - "source": [ - "result = await search_engine.asearch(\"Tell me about how company use AI\")\n", - "print(result.response)" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 47, - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "# Key Focus Areas for AI Startups\n", - "\n", - "## Enhancing Developer Productivity\n", - "\n", - "AI startups should prioritize developing tools that enhance developer productivity. AI can automate routine tasks, generate unit tests, write documentation, and analyze code for defects and optimizations. These capabilities can significantly reduce the time developers spend on monotonous tasks, allowing them to focus on more complex and creative aspects of their work. For instance, automating routine tasks can free up valuable time, while AI-driven code analysis can help identify and fix defects more efficiently [Data: AI Support for Developers and Its Impact on Productivity (39); Entities (299, 303, 304); Relationships (230, 616, 617, 618)].\n", - "\n", - "## Leveraging Generative AI for Innovation\n", - "\n", - "Generative AI presents numerous opportunities for innovation. AI startups should focus on creating tools that can generate new code snippets or suggest optimizations that developers might not have considered. This can foster innovation and improve workflows by providing solutions and efficiencies that were previously unattainable. The relationship between generative AI and opportunities underscores the positive impact AI can have on the development process [Data: AI Support for Developers and Its Impact on Productivity (39); Entities (310); Relationships (206, 622, 628)].\n", - "\n", - "## Addressing Concerns and Ensuring Quality\n", - "\n", - "Despite the benefits, there are significant concerns regarding the use of AI support. Startups must address potential defects or vulnerabilities in AI-generated code, the need for validation and human oversight, and the impact on job security. Developers worry that over-reliance on AI could lead to a decrease in the quality of code and potentially introduce security risks. Implementing robust validation processes and ensuring human oversight are crucial to mitigate these risks [Data: AI Support for Developers and Its Impact on Productivity (39); Entities (311, 320); Relationships (620, 623, 627)].\n", - "\n", - "## Importance of Training Programs\n", - "\n", - "Training programs are essential for helping developers effectively integrate AI support into their workflows. These programs can address developers' concerns, improve their understanding of AI tools, and ensure that they are used effectively. By providing comprehensive training, organizations can help developers build trust in AI tools and enhance their overall satisfaction and productivity. Training programs can also help developers stay updated with the latest advancements in AI technology [Data: AI Support for Developers and Its Impact on Productivity (39); Entities (308); Relationships (621)].\n", - "\n", - "## Building Trust in AI Tools\n", - "\n", - "Trust is a critical factor for the successful adoption of AI tools among developers. Startups must ensure that AI-generated outputs are accurate and beneficial, address concerns about defects or vulnerabilities, and provide transparency in how AI tools operate. Building trust involves ensuring that AI tools are perceived as helpful and trustworthy to maximize their positive impact [Data: AI Support for Developers and Its Impact on Productivity (39); Entities (322); Relationships (629)].\n", - "\n", - "## Role of Virtual Pair-Programming Partners\n", - "\n", - "AI tools can act as virtual pair-programming partners, providing suggestions and feedback similar to a human partner. This can enhance the developer experience by offering real-time assistance and improving the quality of code. Virtual pair-programming partners can help developers identify potential issues early in the development process and suggest optimizations, thereby improving the overall efficiency and effectiveness of the development process [Data: AI Support for Developers and Its Impact on Productivity (39); Entities (317); Relationships (624)].\n", - "\n", - "## Conclusion\n", - "\n", - "In summary, AI startups should focus on enhancing developer productivity, leveraging generative AI for innovation, addressing concerns and ensuring quality, providing training programs, building trust in AI tools, and utilizing virtual pair-programming partners. By concentrating on these areas, AI startups can create tools that significantly improve the development process and foster innovation while addressing the concerns and needs of developers.\n" - ] - } - ], ->>>>>>> origin/main - "source": [ - "question = \"Tell me about what a startup that build IA should focus\"\n", - "result = await search_engine.asearch(question)\n", - "print(result.response)" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "#### Inspecting the context data used to generate the response" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 48, - "metadata": {}, - "outputs": [ - { - "data": { - "text/html": [ - "
\n", - "\n", - "\n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - "
identitydescriptionnumber of relationshipsin_context
0116ENTREPRENEURSHIP SUPPORT TOOLA generative AI-powered tool designed to assis...3True
1308TRAINING PROGRAMSTraining programs are initiatives that organiz...1True
2117KENYAN ENTREPRENEURSBusiness owners in Kenya who participated in a...1True
3579AI AT WORKAI at Work is a report by Microsoft and Linked...2True
4311CONCERNSConcerns refer to the potential drawbacks and ...1True
\n", - "
" - ], - "text/plain": [ - " id entity \\\n", - "0 116 ENTREPRENEURSHIP SUPPORT TOOL \n", - "1 308 TRAINING PROGRAMS \n", - "2 117 KENYAN ENTREPRENEURS \n", - "3 579 AI AT WORK \n", - "4 311 CONCERNS \n", - "\n", - " description number of relationships \\\n", - "0 A generative AI-powered tool designed to assis... 3 \n", - "1 Training programs are initiatives that organiz... 1 \n", - "2 Business owners in Kenya who participated in a... 1 \n", - "3 AI at Work is a report by Microsoft and Linked... 2 \n", - "4 Concerns refer to the potential drawbacks and ... 1 \n", - "\n", - " in_context \n", - "0 True \n", - "1 True \n", - "2 True \n", - "3 True \n", - "4 True " - ] - }, - "execution_count": 48, - "metadata": {}, - "output_type": "execute_result" - } - ], ->>>>>>> origin/main - "source": [ - "result.context_data[\"entities\"].head()" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 49, - "metadata": {}, - "outputs": [ - { - "data": { - "text/html": [ - "
\n", - "\n", - "\n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - "
idsourcetargetdescriptionweightranklinksin_context
0275SECOND MICROSOFT AI AND PRODUCTIVITY REPORTLEARNINGSThe report synthesizes learnings from various ...1.071True
1397ENTREPRENEURSHIP SUPPORT TOOLKENYAN ENTREPRENEURSThe generative AI-powered entrepreneurship sup...1.041True
2809KHEMKA, M.HOUCK, B.Khemka and Houck co-authored a survey on AI su...2.041True
3621AI SUPPORT FOR DEVELOPERSTRAINING PROGRAMSTraining programs can help developers effectiv...1.0174True
4623AI SUPPORT FOR DEVELOPERSCONCERNSConcerns refer to the potential drawbacks deve...1.0174True
\n", - "
" - ], - "text/plain": [ - " id source target \\\n", - "0 275 SECOND MICROSOFT AI AND PRODUCTIVITY REPORT LEARNINGS \n", - "1 397 ENTREPRENEURSHIP SUPPORT TOOL KENYAN ENTREPRENEURS \n", - "2 809 KHEMKA, M. HOUCK, B. \n", - "3 621 AI SUPPORT FOR DEVELOPERS TRAINING PROGRAMS \n", - "4 623 AI SUPPORT FOR DEVELOPERS CONCERNS \n", - "\n", - " description weight rank links \\\n", - "0 The report synthesizes learnings from various ... 1.0 7 1 \n", - "1 The generative AI-powered entrepreneurship sup... 1.0 4 1 \n", - "2 Khemka and Houck co-authored a survey on AI su... 2.0 4 1 \n", - "3 Training programs can help developers effectiv... 1.0 17 4 \n", - "4 Concerns refer to the potential drawbacks deve... 1.0 17 4 \n", - "\n", - " in_context \n", - "0 True \n", - "1 True \n", - "2 True \n", - "3 True \n", - "4 True " - ] - }, - "execution_count": 49, - "metadata": {}, - "output_type": "execute_result" - } - ], ->>>>>>> origin/main - "source": [ - "result.context_data[\"relationships\"].head()" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 50, - "metadata": {}, - "outputs": [ - { - "data": { - "text/html": [ - "
\n", - "\n", - "\n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - "
idtitlecontent
039AI Support for Developers and Its Impact on Pr...# AI Support for Developers and Its Impact on ...
139AI Support for Developers and Its Impact on Pr...# AI Support for Developers and Its Impact on ...
\n", - "
" - ], - "text/plain": [ - " id title \\\n", - "0 39 AI Support for Developers and Its Impact on Pr... \n", - "1 39 AI Support for Developers and Its Impact on Pr... \n", - "\n", - " content \n", - "0 # AI Support for Developers and Its Impact on ... \n", - "1 # AI Support for Developers and Its Impact on ... " - ] - }, - "execution_count": 50, - "metadata": {}, - "output_type": "execute_result" - } - ], ->>>>>>> origin/main - "source": [ - "result.context_data[\"reports\"].head()" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 51, - "metadata": {}, - "outputs": [ - { - "data": { - "text/html": [ - "
\n", - "\n", - "\n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - "
idtext
03on productivity in real-world contexts. This ...
11428 and 3.75, respectively. \\n \\nAs noted abov...
228Microsoft Copilot for Security . Available at...
32impact of these tools outside \\nof a lab sett...
425collect both survey \\nand telemetry measures ...
\n", - "
" - ], - "text/plain": [ - " id text\n", - "0 3 on productivity in real-world contexts. This ...\n", - "1 14 28 and 3.75, respectively. \\n \\nAs noted abov...\n", - "2 28 Microsoft Copilot for Security . Available at...\n", - "3 2 impact of these tools outside \\nof a lab sett...\n", - "4 25 collect both survey \\nand telemetry measures ..." - ] - }, - "execution_count": 51, - "metadata": {}, - "output_type": "execute_result" - } - ], ->>>>>>> origin/main - "source": [ - "result.context_data[\"sources\"].head()" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 52, - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Empty DataFrame\n", - "Columns: [in_context]\n", - "Index: []\n" - ] - } - ], ->>>>>>> origin/main - "source": [ - "if \"claims\" in result.context_data:\n", - " print(result.context_data[\"claims\"].head())" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "### Question Generation" - ] - }, - { - "cell_type": "markdown", - "metadata": {}, - "source": [ - "This function takes a list of user queries and generates the next candidate questions." - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, -======= - "execution_count": 53, ->>>>>>> origin/main - "metadata": {}, - "outputs": [], - "source": [ - "question_generator = LocalQuestionGen(\n", - " llm=llm,\n", - " context_builder=context_builder,\n", - " token_encoder=token_encoder,\n", - " llm_params=llm_params,\n", - " context_builder_params=local_context_params,\n", - ")" - ] - }, - { - "cell_type": "code", -<<<<<<< HEAD - "execution_count": null, - "metadata": {}, - "outputs": [], -======= - "execution_count": 54, - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "['- What are the main productivity gains associated with the use of AI tools like Microsoft Copilot in the workplace?', \"- How does the collaboration between Microsoft and LinkedIn contribute to understanding AI's impact on workplace environments?\", '- What are the challenges and concerns developers have regarding AI integration in their workflows?', '- How does the use of AI tools like Copilot vary across different roles, functions, and organizations?', '- What are the potential effects of AI tools on cognitive effort and collaboration within teams?']\n" - ] - } - ], ->>>>>>> origin/main - "source": [ - "question_history = [\n", - " \"Tell me about what are the most underestimate domain toward use of AI in company\",\n", - " \"How AI impact companies ?\",\n", - "]\n", - "candidate_questions = await question_generator.agenerate(\n", - " question_history=question_history, context_data=None, question_count=5\n", - ")\n", - "print(candidate_questions.response)" - ] - } - ], - "metadata": { - "kernelspec": { - "display_name": "Python 3", - "language": "python", - "name": "python3" - }, - "language_info": { - "codemirror_mode": { - "name": "ipython", - "version": 3 - }, - "file_extension": ".py", - "mimetype": "text/x-python", - "name": "python", - "nbconvert_exporter": "python", - "pygments_lexer": "ipython3", - "version": "3.11.9" - } - }, - "nbformat": 4, - "nbformat_minor": 2 -} diff --git a/graphfleet/output/graphindex/artifacts/clustered_graph.0.graphml b/graphfleet/output/graphindex/artifacts/clustered_graph.0.graphml new file mode 100644 index 000000000..efc51ecd6 --- /dev/null +++ b/graphfleet/output/graphindex/artifacts/clustered_graph.0.graphml @@ -0,0 +1,16324 @@ + + + + + + + + + + + + + + + + + + + PERSON + Darren Edge is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 0 + b45241d70f0e43fca764df95b2b81f77 + + + PERSON + Ha Trinh is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 1 + 4119fd06010c494caa07f439b333f4c5 + + + PERSON + Newman Cheng is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 2 + d3835bf3dda84ead99deadbeac5d0d7d + + + PERSON + Joshua Bradley is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 3 + 077d2820ae1845bcbb1803379a3d1eae + + + PERSON + Alex Chao is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 4 + 3671ea0dd4e84c1a9b02c5ab2c8f4bac + + + PERSON + Apurva Mody is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 5 + 19a7f254a5d64566ab5cc15472df02de + + + PERSON + Steven Truitt is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 6 + e7ffaee9d31d4d3c96e04f911d0a8f9e + + + PERSON + Jonathan Larson is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 7 + f7e11b0e297a44a896dc67928368f600 + + + ORGANIZATION + Microsoft Research is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + 3 + 8 + 1fd3fa8bb5a2408790042ab9573779ee + + + ORGANIZATION + Microsoft Strategic Missions and Technologies is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + 3 + 9 + 27f9fbe6ad8c4a8b9acee0d3596ed57c + + + ORGANIZATION + Microsoft Office of the CTO is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + 2 + 10 + e1fd0e904a53409aada44442f23a51cb + + + METHOD + RAG (Retrieval-Augmented Generation) is a developing research area with multiple established directions, including knowledge graph creation, completion, and extraction of causal graphs. It is a method used for generating responses in text generation tasks by retrieving relevant information from an external knowledge source to enable large language models to answer questions. This approach incorporates the retrieval of relevant data to augment text generation, producing direct responses in various text generation tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 2 + 0 + 31 + 11 + de988724cfdf45cebfba3b13c43ceede + + + TECHNOLOGY + LLM (Large Language Model) is a type of artificial intelligence model used for a variety of tasks in the field of Natural Language Processing and Information Retrieval. These tasks include generating and assessing text, entity extraction, summarization, understanding relationships in text, and automating human-like sensemaking and reasoning over large collections of documents. LLMs are also employed to generate intermediate answers and scores for text chunks, process these chunks to extract elements of a graph index, and automate the generation of questions for dataset evaluation. Additionally, LLMs can analyze and generate text based on retrieved information and queries, and they possess a context window that can be exceeded by external datasets. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,2c6ed90897310eea2f28e33fff1c32b0,6f33a085ff3304e5994f7fbb86c881a4,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 2 + 0 + 32 + 12 + 96aad7cb4b7d40e9b7e13b94a67af206 + + + METHOD + Graph RAG (Retrieval-Augmented Generation) is a sophisticated method that leverages the natural modularity of graphs to partition data for global summarization tasks. This approach integrates multiple stages and concepts, including knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS), to support human sensemaking over text corpora. It is particularly effective in providing comprehensive and structured overviews of public figures across various sectors of the entertainment industry, as well as generating answers for questions in the News article dataset. + +Graph RAG employs a high-level data flow and pipeline for processing and summarizing text, combining both global and local approaches to optimize token usage in text generation tasks. It uses community summaries to improve answer comprehensiveness and diversity while requiring fewer tokens compared to traditional source text summarization methods. This method has been shown to outperform naive RAG in terms of comprehensiveness and diversity in text generation tasks. + +A specific implementation of Graph RAG involves using four levels of graph communities, incorporating concepts from other systems such as self-memory and parallel generation of community answers. This multi-stage mechanism allows for the comparison of multiple conditions, enhancing the overall effectiveness of the summarization process. + +Graph RAG, launched by NebulaGraph, is a retrieval-augmented generation technology based on knowledge graphs. It combines global summarization with graph-based text indexing to answer questions over private text corpora, making it a versatile tool for various text analysis and summarization applications. + 086021a89900a39bcb62036981737bfa,21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,718017a4871c909420f84b85b8ba969d,833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19,fb3c48579608fa28be585ceb6cd2f0fe + 7 + 0 + 90 + 13 + c9632a35146940c2a86167c7726d35e9 + + + METHOD + QFS (Query-Focused Summarization) is a method used to generate summaries based on specific user queries + e8d83e6e7a7c0f57b218cef24976b745 + 7 + 0 + 1 + 14 + 9646481f66ce4fd2b08c2eddda42fc82 + + + CONCEPT + Community summaries are generated summaries of data clusters or communities, used to answer queries and tailored to the domain. These summaries are pre-generated for groups of closely-related entities, particularly in the Graph RAG approach, and are derived from community-generated content to compare with source texts. They are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks. Community summaries provide report-like insights into each community within a hierarchical structure, which is useful for understanding the dataset. They are generated from modular communities in the knowledge graph and cover different levels of each graph community hierarchy, including root-level communities in an entity-based graph index. Additionally, these summaries act as a kind of self-memory for generation-augmented retrieval. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + 12 + 0 + 26 + 15 + d91a266f766b4737a06b0fda588ba40b + + + CONCEPT + Global sensemaking questions are questions that require understanding and summarizing large datasets, often beyond the explicit content of the source texts + e8d83e6e7a7c0f57b218cef24976b745 + 1 + 0 + 3 + 16 + bc0e3f075a4c4ebbb7c7b152b65a5625 + + + METRIC + 1 million token range refers to the scale of datasets used in the evaluation of the Graph RAG approach + e8d83e6e7a7c0f57b218cef24976b745 + 1 + 0 + 1 + 17 + 254770028d7a4fa9877da4ba0ad5ad21 + + + TECHNOLOGY + Python is a programming language used for implementing both global and local Graph RAG approaches. Additionally, Python is utilized to implement the open-source version of the Graph RAG approach. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + 7 + 0 + 1 + 18 + 4a67211867e5464ba45126315a122a8a + + + URL + The URL "HTTPS://AKA.MS/GRAPHRAG" is the location where the open-source, Python-based implementation of Graph RAG approaches will be available. This URL serves as the repository for accessing the open-source implementation of the Graph RAG approach. + e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745 + 17 + 0 + 2 + 19 + 04dbbb2283b845baaeac0eaf0c34c9da + + + METHOD + Query-Focused Summarization (QFS) is a method used to generate summaries that are relevant to specific user queries. This summarization technique focuses on answering specific queries by utilizing the entire corpus of information available. It is designed to provide concise and relevant information based on the specific needs of the user, ensuring that the generated summaries are directly aligned with the queries posed. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + 12 + 0 + 5 + 20 + 1943f245ee4243bdbfbd2fd619ae824a + + + CONCEPT + An external knowledge source is a repository of information that can be accessed to retrieve relevant data for answering questions + e8d83e6e7a7c0f57b218cef24976b745 + 2 + 0 + 1 + 21 + 273daeec8cad41e6b3e450447db58ee7 + + + CONCEPT + A text corpus is a large collection of written texts used for analysis and research + e8d83e6e7a7c0f57b218cef24976b745 + 1 + 0 + 1 + 22 + e69dc259edb944ea9ea41264b9fcfe59 + + + CONCEPT + An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 0 + 2 + 23 + e2f5735c7d714423a2c4f61ca2644626 + + + CONCEPT + Source documents are the original texts from which information is extracted, retrieved, or summarized. These documents serve as the foundational input texts for various processing tasks, including the Graph RAG (Retrieval-Augmented Generation) approach. In this context, source documents are critical for extracting and analyzing information, ensuring that the data used in computational linguistics and information retrieval tasks is accurate and comprehensive. + bc9e2c9e369c4108cf4f6dd5f60960f4,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 5 + 24 + deece7e64b2a4628850d4bb6e394a9c3 + + + CONCEPT + A partial response is an intermediate answer generated from community summaries before being combined into a final response + e8d83e6e7a7c0f57b218cef24976b745 + 12 + 0 + 2 + 25 + e657b5121ff8456b9a610cfaead8e0cb + + + CONCEPT + A final response is the comprehensive answer generated after combining all partial responses + e8d83e6e7a7c0f57b218cef24976b745 + 12 + 0 + 1 + 26 + bf4e255cdac94ccc83a56435a5e4b075 + + + METRIC + COMPREHENSIVENESS is a metric used to evaluate the quality of generated responses by measuring how much detail an answer provides to cover all aspects and details of a question. It assesses the completeness and thoroughness of answers, ensuring that they encompass all relevant information. This metric is particularly important in evaluating the summarization approach, focusing on the completeness of the summary. In practical applications, such as evaluating Podcast transcripts and News articles, comprehensiveness has shown win rates between 72-83% and 72-80%, respectively. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + 7 + 0 + 9 + 27 + 3b040bcc19f14e04880ae52881a89c1c + + + METRIC + DIVERSITY is a metric used to evaluate the variety and richness of answers generated in response to a question. It measures how varied and rich an answer is in providing different perspectives and insights. This metric is particularly important in assessing the quality of summarization approaches, focusing on the variety of information included in the summary. DIVERSITY is applied to various types of content, including Podcast transcripts, where win rates range from 75-82%, and News articles, with win rates ranging from 62-71%. It is a crucial target quality for evaluating the effectiveness of different methods in generating diverse and informative responses. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + 7 + 0 + 7 + 28 + 3d6b216c14354332b1bf1927ba168986 + + + ACTIVITY + Human endeavors refer to activities and efforts across various domains that rely on reading and reasoning about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 2 + 29 + 1c109cfdc370463eb6d537e5b7b382fb + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models designed to understand and generate human-like text, playing a crucial role in automating sensemaking in complex domains. Modern language models, such as GPT, Llama, and Gemini, leverage in-context learning to effectively summarize content. These models are integral to the field of Natural Language Processing and Information Retrieval, enabling sophisticated text analysis and generation capabilities. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + 3 + 0 + 11 + 30 + 3d0dcbc8971b415ea18065edc4d8c8ef + + + DOMAIN + Scientific discovery is a complex domain where sensemaking is applied to understand and generate new knowledge from scientific texts + f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 1 + 31 + 68105770b523412388424d984e711917 + + + DOMAIN + Intelligence analysis is a complex domain where sensemaking is applied to understand and generate insights from intelligence data + f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 1 + 32 + 85c79fd84f5e4f918471c386852204c5 + + + PROCESS + SENSEMAKING is the process of understanding and making sense of complex information. It involves understanding connections among people, places, and events to anticipate their trajectories and act effectively. This process is crucial for navigating and interpreting intricate data landscapes, enabling individuals and organizations to make informed decisions based on the relationships and patterns identified within the information. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 7 + 33 + eae4259b19a741ab9f9f6af18c4a0470 + + + DATA + TEXT CHUNKS are segments of text that are embedded into a vector space for analysis. These segments are extracted from source documents and are used for processing in the Graph RAG (Retrieval-Augmented Generation) approach. By embedding these text chunks into a vector space, they can be analyzed more effectively, facilitating various natural language processing and information retrieval tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + 8 + 0 + 5 + 34 + 3138f39f2bcd43a69e0697cd3b05bc4d + + + DATA + Element instances are identified and extracted instances of graph nodes and edges from text chunks. They represent individual occurrences of entities, relationships, and claims extracted from source texts. These specific pieces of information are tailored to the domain, providing a structured representation of the underlying data. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 4 + 35 + dde131ab575d44dbb55289a6972be18f + + + DATA + Element summaries are concise representations of element instances, tailored to the domain. They are descriptive texts created by the LLM to summarize entities, relationships, and claims extracted from source texts. These summaries provide detailed descriptions of nodes, edges, and covariates within a community, and are used to understand the structure and semantics of the dataset. In essence, element summaries serve as a tool to encapsulate and convey the intricate details of elements within a graph, facilitating a deeper comprehension of the dataset's structural dynamics and semantic relationships. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 10 + 36 + de9e343f2e334d88a8ac7f8813a915e5 + + + DATA + Graph communities are groups of elements, including nodes, edges, and covariates, detected within a graph index, primarily used for summarization. These communities consist of groups of nodes that exhibit stronger connections to each other than to nodes outside the group. This structural characteristic allows for the identification and analysis of densely connected subgraphs, which can be crucial for understanding the underlying relationships and dynamics within complex networks. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 12 + 0 + 3 + 37 + e2bf260115514fb3b252fd879fb3e7be + + + DATA + COMMUNITY ANSWERS are query-focused summaries derived from community summaries. These answers are generated in parallel from the community summaries and are designed to respond to user queries effectively. Essentially, COMMUNITY ANSWERS serve as responses that synthesize information from community summaries to provide concise and relevant answers to specific questions posed by users. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + 12 + 0 + 3 + 38 + b462b94ce47a4b8c8fffa33f7242acec + + + DATA + GLOBAL ANSWER is a comprehensive response generated from multiple community summaries to answer a user query. It is the final query-focused summary produced from all relevant community summaries. The final answer is generated by combining intermediate community answers based on their helpfulness scores. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 12 + 0 + 7 + 39 + 17ed1d92075643579a712cc6c29e8ddb + + + TIME + Indexing time refers to the time when the graph index is created and elements are summarized + f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 1 + 40 + 3ce7c210a21b4deebad7cc9308148d86 + + + TIME + Query time refers to the time when a query is made and the relevant summaries are generated + f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 1 + 41 + d64ed762ea924caa95c8d06f072a9a96 + + + PROCESS + Graph RAG pipeline is a process using an LLM-derived graph index to detect, extract, and summarize nodes, edges, and covariates in source documents + f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 7 + 42 + adf4ee3fbe9b4d0381044838c4f889c8 + + + DATA + NODES are entities detected in the graph index of source documents. They represent the individual elements or points in a graph. For instance, in the Podcast dataset, there are 8,564 nodes, while the News dataset contains 15,754 nodes. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + 0 + 0 + 3 + 43 + 32ee140946e5461f9275db664dc541a5 + + + DATA + EDGES are relationships detected in the graph index of source documents. They represent the connections or links between nodes in a graph. For instance, in the Podcast dataset, there are 20,691 edges, while the News dataset contains 19,520 edges. These edges are crucial for understanding the structural dynamics and relationships within the datasets, providing insights into how different nodes (such as topics, entities, or documents) are interconnected. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + 0 + 0 + 3 + 44 + c160b9cb27d6408ba6ab20214a2f3f81 + + + DATA + Covariates are additional attributes associated with extracted node instances in the graph index. They represent claims or additional information detected in the graph index of source documents. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 2 + 45 + 23527cd679ff4d5a988d52e7cd056078 + + + METHOD + LEIDEN is a community detection algorithm renowned for its efficiency in recovering hierarchical community structures. It is widely used to partition graphs into modular communities, effectively grouping elements within a graph index. The algorithm's ability to identify and organize these communities makes it a valuable tool in the analysis of complex networks, particularly within the domains of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 7 + 46 + f1c6eed066f24cbdb376b910fce29ed4 + + + METHOD + Retrieval-Augmented Generation (RAG) is an established approach in the field of Natural Language Processing and Information Retrieval, designed to answer user questions over entire datasets. This method involves retrieving relevant text regions to provide grounding for the generation task, thereby enhancing the accuracy and relevance of the generated responses. By combining retrieval and generation processes, RAG effectively synthesizes and presents pertinent information, making it a powerful tool for handling complex queries and large datasets. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + 8 + 0 + 4 + 47 + 83a6cb03df6b41d8ad6ee5f6fef5f024 + + + ORGANIZATION + Microsoft is a technology company whose Chief Technology Officer, Kevin Scott, actively participates in podcast conversations. The organization is deeply involved in automating sensemaking in scientific discovery through the use of large language models (LLMs). Notably, Microsoft conducted a study examining the impact of large language models, specifically GPT-4, on scientific discovery. + 1d07b4248c2655081c7af0e373bd70c9,833e7d67dcd30790b26b71c9b5306f6b,f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 3 + 48 + 147c038aef3e4422acbbc5f7938c4ab8 + + + PERSON + Ranade is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 1 + 49 + b7702b90c7f24190b864e8c6e64612a5 + + + PERSON + Joshi is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 1 + 50 + de6fa24480894518ab3cbcb66f739266 + + + PERSON + Klein is an author who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 1 + 51 + 6fae5ee1a831468aa585a1ea09095998 + + + PERSON + Lewis is an author who contributed to the development of the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 1 + 52 + ef32c4b208d041cc856f6837915dc1b0 + + + PERSON + Traag is an author who contributed to the development of the Leiden community detection method + f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 1 + 53 + 07b2425216bd4f0aa4e079827cb48ef5 + + + PUBLICATION + arXiv is a preprint repository where several significant papers in the field of Natural Language Processing and Information Retrieval have been published. It serves as a platform for electronic preprints (known as e-prints) that are approved for publication after moderation, but not full peer review. Notable papers published on arXiv include "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models," "Lost in the middle: How language models use long contexts," "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," "Llama 2: Open foundation and fine-tuned chat models," "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy," "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries," "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions," "Enhancing knowledge graph construction using large language models," "Is chatgpt a good nlg evaluator? a preliminary study," "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt," "Causal graph discovery with retrieval-augmented generation based large language models," "Knowledge graph prompting for multi-document question answering," "Text summarization with latent queries," "Retrieval-augmented generation for large language models: A survey," and "Knowledge graph-augmented language models for knowledge-grounded dialogue generation." This repository is a crucial resource for researchers to disseminate their findings rapidly and access the latest advancements in their fields. + 00e8e4e881bd0862022f4dfc913b900b,086021a89900a39bcb62036981737bfa,58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035,f0306814bf64f5c9e79603fc6a52f4ea,fc4b27d64f055b7fc30176ba110dd02e + 4 + 0 + 39 + 54 + 2670deebfa3f4d69bb82c28ab250a209 + + + PUBLICATION + Preprint refers to the version of the research paper that is under review and available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 4 + 0 + 4 + 55 + 404309e89a5241d6bff42c05a45df206 + + + CATEGORY + cs.CL is the category under which the research paper is classified on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 4 + 0 + 1 + 56 + b785a9025069417f94950ad231bb1441 + + + DATE + 24 Apr 2024 is the date when the research paper was submitted to arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 4 + 0 + 1 + 57 + 3b6cd96a27304614850709aba1c9598b + + + IDENTIFIER + 2404.16130v1 is the identifier for the research paper on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 4 + 0 + 1 + 58 + d54956b79dd147f894b67a8b97dcbef0 + + + DATA + Document collections refer to large sets of documents that are analyzed for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 1 + 59 + 958beecdb5bb4060948415ffd75d2b03 + + + TECHNOLOGY + LLM PROMPTS are specific instructions given to large language models (LLMs) to tailor their responses to the domain of the dataset. These prompts are also used to extract elements from text chunks, ensuring that the LLMs provide relevant and precise information based on the given context. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 2 + 60 + b999ed77e19e4f85b7f1ae79af5c002a + + + METHOD + Community detection is a method used to identify groups of related elements within a graph. It involves the process of identifying communities within a graph, which are clusters of nodes that are more densely connected internally than with the rest of the network. This technique is crucial in understanding the structural dynamics and relationships within complex networks, such as those found in social networks, biological systems, and information retrieval systems. By uncovering these communities, researchers can gain insights into the underlying structure and function of the network, facilitating more effective analysis and interpretation of the data. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 12 + 0 + 3 + 61 + 48c0c4d72da74ff5bb926fa0c856d1a7 + + + METHOD + Domain-tailored summarization is a method used to create summaries that are specific to the domain of the dataset + f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 2 + 62 + 4f3c97517f794ebfb49c4c6315f9cf23 + + + PERSON + Klein et al. are authors who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 1 + 63 + 1745a2485a9443bab76587ad650e9be0 + + + PERSON + Ranade and Joshi are authors who discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + 3 + 0 + 1 + 64 + 32e6ccab20d94029811127dbbe424c64 + + + PERSON + Lewis et al. are authors who developed the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 1 + 65 + 94a964c6992945ebb3833dfdfdc8d655 + + + PERSON + Traag et al. are the authors who developed the Leiden algorithm, a method renowned for its efficiency in recovering hierarchical community structures. This algorithm is widely recognized in the field of Natural Language Processing and Information Retrieval for its ability to accurately detect and map out complex community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 8 + 0 + 1 + 66 + 1eb829d0ace042089f0746f78729696c + + + METHOD + QFS is a task framing that focuses on generating summaries based on specific queries, rather than just concatenating excerpts + fb3c48579608fa28be585ceb6cd2f0fe + 1 + 67 + 015e7b58d1a14b44beab3bbc9f912c18 + + + METHOD + A type of summarization that generates natural language summaries based on specific queries, rather than just extracting text + fb3c48579608fa28be585ceb6cd2f0fe + 3 + 68 + 26f88ab3e2e04c33a459ad6270ade565 + + + TECHNOLOGY + A neural network architecture that has shown substantial improvements in various summarization tasks + fb3c48579608fa28be585ceb6cd2f0fe + 3 + 69 + babe97e1d9784cffa1c85abc1e588126 + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + 3 + 0 + 3 + 70 + 1033a18c45aa4584b2aef6ab96890351 + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + 3 + 0 + 2 + 71 + c9b8ce91fc2945b4907fe35519339cac + + + TECHNOLOGY + GEMINI is a family of highly capable multimodal models, as described in an arXiv preprint. These models are known for their ability to perform in-context learning and summarization, making them a significant advancement in the field of Natural Language Processing and Information Retrieval. + 086021a89900a39bcb62036981737bfa,fb3c48579608fa28be585ceb6cd2f0fe + 3 + 0 + 2 + 72 + fa3c4204421c48609e52c8de2da4c654 + + + TECHNOLOGY + A knowledge graph is a structured representation of information, utilized in the Graph RAG approach for summarization. This structured representation of knowledge is specifically employed in the Graph RAG approach for global summarization, highlighting its role in organizing and integrating information to facilitate comprehensive and coherent summaries. + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + 7 + 0 + 2 + 73 + 53af055f068244d0ac861b2e89376495 + + + REFERENCE + Authors of a paper on Retrieval-augmented generation (RAG) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 8 + 0 + 1 + 74 + c03ab3ce8cb74ad2a03b94723bfab3c7 + + + REFERENCE + Author of a paper on query-focused summarization (QFS) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 75 + ed6d2eee9d7b4f5db466b1f6404d31cc + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 76 + fc01e9baa80e417c9206f941bb279407 + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 77 + 56d0e5ebe79e4814bd1463cf6ca21394 + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 78 + 7c49f2710e8b4d3b8dc9310834406ea5 + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 79 + c6d1e4f56c2843e89cf0b91c10bb6de2 + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 80 + 0adb2d9941f34ef7b2f7743cc6225844 + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 81 + 6b02373137fd438ba96af28f735cdbdb + + + REFERENCE + Authors of a paper on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 3 + 0 + 1 + 82 + 36a4fcd8efc144e6b8af9a1c7ab8b2ce + + + REFERENCE + "BROWN ET AL., 2020" refers to a publication by Brown et al. in 2020, which discusses in-context learning with few-shot examples. The authors of this paper are also known for their work on the GPT series of large language models. + bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 3 + 0 + 2 + 83 + fbeef791d19b413a9c93c6608286ab63 + + + REFERENCE + Authors of a paper on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 3 + 0 + 1 + 84 + d2b629c0396f4180a03e16ddf3818589 + + + REFERENCE + Authors of a paper on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 3 + 0 + 1 + 85 + 6102fc6619ed422ebc42588bfa97355d + + + REFERENCE + "KURATOV ET AL., 2024" refers to a publication by Kuratov and colleagues in 2024. The study discusses the recall degradation and potential for information loss in longer context windows of Large Language Models (LLMs). The authors explore the limitations of these extended context windows, providing insights into how the performance of LLMs can be affected when dealing with longer sequences of text. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 3 + 0 + 3 + 86 + 8d141c0b80f74b79a05eed7fe161fe49 + + + REFERENCE + "LIU ET AL., 2023" refers to a publication by Liu et al. in 2023, which discusses the recall degradation and potential for information loss in longer context windows of large language models (LLMs). The authors explore the limitations of LLM context windows, highlighting how extended contexts can lead to decreased recall accuracy and information retention. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 3 + 0 + 3 + 87 + e22d1d1cd8d14f12b81828d940f40d70 + + + TECHNOLOGY + COMMUNITY DETECTION ALGORITHMS are algorithms used to partition a graph into communities of nodes with stronger connections to one another. These algorithms are designed to identify modular communities of closely-related nodes within a graph, thereby revealing the underlying structure and relationships within the network. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + 8 + 0 + 6 + 88 + 9ab48505fb1b487babd0d1f6d3a3f980 + + + ALGORITHM + Louvain is a community detection algorithm used to partition graphs into modular communities + 21e52bc06a82796b1f4bcd73edda1f2a + 8 + 0 + 1 + 89 + 148fffeb994541b2b4b6dcefda7001a8 + + + DATASET + HOTPOTQA is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical resource for evaluating entity extraction prompts, particularly with advanced models like GPT-4-turbo. Additionally, HotPotQA is utilized to observe the behavior of text chunk extraction within the Graph RAG (Retrieval-Augmented Generation) approach, making it a versatile tool in the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4 + 3 + 0 + 3 + 90 + 89c08e793298442686292454a1abff31 + + + TECHNOLOGY + GPT-4-Turbo is a version of the GPT-4 model characterized by its large context size of 128k tokens, which is utilized in various analytical tasks. Specifically, GPT-4-Turbo is employed for entity extraction in evaluations, leveraging its extensive context capacity to enhance the accuracy and comprehensiveness of the analysis. This model is particularly suited for tasks within the Natural Language Processing and Information Retrieval domain, where handling large volumes of text and extracting relevant entities are critical. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + 3 + 0 + 2 + 91 + 0467928aa65e4a4fba62bdb1467e3a54 + + + DATASET + The "PODCAST TRANSCRIPTS" dataset is a comprehensive collection of compiled transcripts from podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders. This dataset is used for analysis and consists of 1669 text chunks, each containing 600 tokens with 100-token overlaps between chunks, amounting to approximately 1 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620,ebf5249c888e07fedce6572a4c03f88c + 0 + 0 + 19 + 92 + 43c3390303c6476cb65f584e37c3e81c + + + DATASET + The "NEWS ARTICLES" dataset is a comprehensive collection of news articles used for analysis. It serves as a benchmark dataset comprising news articles published from September 2013 to December 2023. The dataset spans a range of categories, including entertainment, business, sports, technology, health, and science. It consists of 3197 text chunks, each containing 600 tokens, with a 100-token overlap between chunks, amounting to approximately 1.7 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620 + 0 + 0 + 13 + 93 + fa14b16c17e3417dba5a4b473ea5b18d + + + METHOD + MAP-REDUCE is a method employed for text summarization by applying a map-reduce approach directly to source texts. It is particularly utilized for query-focused summarization of an entire corpus, enabling efficient processing and extraction of relevant information from large datasets. This technique leverages the map-reduce paradigm to distribute the computational workload, making it suitable for handling extensive text collections in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,973164fa90bf2b4ee267f4fd795916bf + 12 + 0 + 2 + 94 + 7cc3356d38de4328a51a5cbcb187dac3 + + + METRIC + "EMPOWERMENT" is a concept and metric used in the evaluation of various methods, with an average win rate of 51.3%. It measures how well an answer helps the reader understand and make informed judgments about a topic. Specifically, it evaluates the effectiveness of generated answers in empowering users by developing their understanding of broad issues and themes. Empowerment is a target quality in summarization approaches, focusing on the ability to help users reach an informed understanding. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,ebf5249c888e07fedce6572a4c03f88c + 0 + 0 + 6 + 95 + bef16fb5fd7344cca5e295b13ef3e0cd + + + METHOD + Naive RAG is a basic retrieval-augmented generation (RAG) method used as a baseline for comparison in text generation tasks. It converts documents to text, splits them into chunks, and embeds these chunks into a vector space for query matching. While it produces the most direct responses, it is outperformed by global approaches in terms of comprehensiveness and diversity. Naive RAG is also noted for listing public figures mentioned in entertainment articles, focusing on their professional achievements and personal lives. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 13 + 96 + bb9e01bc171d4326a29afda59ece8d17 + + + METHOD + A method for summarizing source texts using a map-reduce approach + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 97 + 3c063eea52e94164b70c99431ea30bae + + + OUTPUT + Questions generated to evaluate the summarization approach, focusing on understanding activities + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 98 + 252cc8452bfc4c2aa58cab68d8b61879 + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 99 + 7e2c84548fb94ee395ba8588d8f2a006 + + + METRIC + TOKEN COSTS refer to the computational cost measured in tokens used in the summarization process. Specifically, in the context of the Graph RAG (Retrieval-Augmented Generation) approach, token costs denote the number of tokens required for processing text. This metric is crucial for evaluating the efficiency and scalability of text processing methods within the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 7 + 0 + 1 + 100 + f034618dde7948beb6dab30176d0fc87 + + + PROCESS + The high-level process of the Graph RAG approach and pipeline + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 101 + 5c41f96be13e49dba649454297834546 + + + PARAMETER + Design parameters are key settings and configurations in the Graph RAG approach. These parameters are crucial as they influence the design of the Graph RAG approach and pipeline, determining the effectiveness and efficiency of the overall system. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + PARAMETER + 7 + 0 + 1 + 102 + 7ea4afbf8a264f29af29950ce98105ba + + + METHOD + GLOBAL SUMMARIZATION is a method for summarizing information on a global scale. It aims to encapsulate the overall structure and semantics of a dataset, providing a comprehensive overview of information from large datasets or corpora. This technique is particularly useful in the field of Natural Language Processing and Information Retrieval, where it helps in distilling vast amounts of data into coherent and concise summaries, facilitating better understanding and analysis of the underlying information. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 0 + 5 + 103 + 91ff849d12b24574b0691dbddf44968b + + + ATTRIBUTE + Modularity is an inherent quality of graphs that allows them to be partitioned into communities of closely-related nodes + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 104 + d73c1f2fb3094d8dace42ad2a76e9a52 + + + OUTPUT + Descriptions generated from modular communities in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + 12 + 0 + 1 + 105 + cdc8901e668749889bd49bebdc4ff1f6 + + + INPUT + A specific question or request for information that the summarization methods aim to answer + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 106 + 36084a9fab53433493f079e97e68bf65 + + + DATASET + A large collection of texts or documents used for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 107 + eebcc7ec8e3e4df7aea83659bbdc2199 + + + OUTPUT + Intermediate answers generated from community summaries before being combined into a final global answer + 21e52bc06a82796b1f4bcd73edda1f2a + 12 + 0 + 2 + 108 + ceadf262ef834e9ab146b20650912cae + + + OUTPUT + The comprehensive answer generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + 12 + 0 + 1 + 109 + 7f65feab75424b53b24470d305ba331a + + + METHOD + A method that focuses on generating questions to understand activities from datasets + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 2 + 110 + fd9cb733b28d420cb5cef01e545a132c + + + INPUT + Brief descriptions of datasets used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 111 + 0fbcca3f17c649a08aea64b5a7d9ef36 + + + DATASET + Datasets that represent real-world information, such as podcast transcripts and news articles + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 112 + 482027a59f32484c9c44fd700615c1b6 + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 113 + de837ff3d626451282ff6ac77a82216d + + + METHOD + A method that summarizes the original source texts directly + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 114 + 460295fed3ae4cd39f9f274cec9c2506 + + + OUTPUT + LOW-LEVEL COMMUNITY SUMMARIES are a type of community summary used in the News dataset for analysis. These summaries provide a detailed overview of the source text and are generated from lower hierarchical levels of the community in the knowledge graph. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,bc9e2c9e369c4108cf4f6dd5f60960f4 + 8 + 0 + 3 + 115 + 553b285bba60460ab1ed8341ae61282b + + + OUTPUT + INTERMEDIATE-LEVEL COMMUNITY SUMMARIES are summaries that provide a mid-level overview of the source text. These summaries are generated from intermediate hierarchical levels of the community in the knowledge graph, offering a balanced perspective that captures essential details without overwhelming the reader with excessive information. This approach ensures that the summaries are both informative and concise, making them valuable for understanding the structural dynamics and relationships within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 8 + 0 + 2 + 116 + cec95bf17e7e4c939b56c9c6f402a29f + + + OUTPUT + Summaries generated from higher hierarchical levels of the community in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + 7 + 0 + 1 + 117 + 599164aead034bc19446efacc77554d2 + + + PROCESS, SYSTEM + The entity "PIPELINE" refers to a series of processes or steps used to analyze and summarize a dataset. Specifically, in the context of the Graph RAG approach, the pipeline denotes the sequence of steps and processes involved. This structured sequence is essential for systematically handling data, ensuring that each stage of the analysis is methodically executed to achieve accurate and comprehensive results. + 7fb7d9ce2da9c940a32afdd87d1d9e56,bc9e2c9e369c4108cf4f6dd5f60960f4 + 12 + 0 + 2 + 118 + bbf148ae4d48422f8fdef754cfa2b9e4 + + + DATA STRUCTURE, OUTPUT + The "GRAPH INDEX" is a data structure used in Retrieval-Augmented Generation (RAG) systems to organize and retrieve information. It is a self-generated index that enables Graph RAG by utilizing a graph structure to organize and retrieve data. This index is created from a graph structure and is employed for tasks such as query-focused summarization. The graph index includes various elements extracted from text chunks using Large Language Model (LLM) prompts. Additionally, it supports conditions C0-C3 and is created using generic prompts for entity and relationship extraction. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + 0 + 0 + 18 + 119 + de61b2670999433f807a6a1dc2b81e43 + + + DATA, UNIT + Entity references are mentions of entities within text chunks, extracted during the processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + 8 + 0 + 3 + 120 + 3e95dacfe57b4d57b5da4310ef2e157f + + + METRIC + Recall is a metric used to measure the completeness of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 8 + 0 + 1 + 121 + 1f1545308e9347af91fd03b94aadc21f + + + METRIC + Precision is a metric used to measure the accuracy of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 8 + 0 + 1 + 122 + 6ea81acaf232485e94fff638e03336e1 + + + TECHNIQUE, METHOD + FEW-SHOT EXAMPLES are specialized instances provided to the Large Language Model (LLM) to improve its performance in domains with specialized knowledge such as science, medicine, and law. These examples are tailored to the domain of the data used in the graph indexing process and serve as sample inputs for in-context learning. By tailoring the extraction prompt to the document corpus domain, few-shot examples enhance the LLM's ability to understand and process domain-specific information effectively. + 2c6ed90897310eea2f28e33fff1c32b0,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4 + 2 + 0 + 8 + 123 + d136b08d586d488f9e4188b524c85a29 + + + DATA, UNIT + Named entities are specific types of entities such as people, places, and organizations, extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 2 + 0 + 1 + 124 + cccfa151fedc4b218a8d96adc7dceabe + + + REFERENCE, PUBLICATION + A reference to a publication by Yang et al. in 2018, introducing the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + 3 + 0 + 1 + 125 + ce54725672a74ebcabe6127577dacb2b + + + METHOD, APPROACH + Techniques refer to the specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 7 + 0 + 1 + 126 + ea2b28ca1a974ffab4517811dc1d1e5c + + + ATTRIBUTE, CONFIGURATION + Implementation details are specific configurations and settings used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 7 + 0 + 1 + 127 + aff21f1da1654e7babdcf3fb0e4a75fc + + + PROCESS, METHOD + A single extraction round refers to one complete cycle of extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + 2 + 0 + 1 + 128 + dc2cc9016e3f49dbac7232f05cce794d + + + ATTRIBUTE, CONFIGURATION + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + 8 + 0 + 1 + 129 + 6ea0cef05f694dcea455478f40674e45 + + + METRIC, ISSUE + Recall degradation refers to the decrease in recall performance when using longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + 2 + 0 + 1 + 130 + 7ab5d53a872f4dfc98f3d386879f3c75 + + + PROCESS, METHOD + The extraction process involves identifying and extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + 2 + 0 + 2 + 131 + af1d0fec22114a3398b8016f5225f9ed + + + ATTRIBUTE, CONFIGURATION + Domain refers to the specific area of knowledge or field to which the document corpus belongs + bc9e2c9e369c4108cf4f6dd5f60960f4 + 8 + 0 + 1 + 132 + b07a7f088364459098cd8511ff27a4c8 + + + DATA, INPUT + Document corpus refers to the collection of documents being processed in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 8 + 0 + 2 + 133 + 8870cf2b5df64d2cab5820f67e29b9f1 + + + TECHNIQUE, METHOD + Default prompt is the standard set of instructions given to the LLM for extracting named entities + bc9e2c9e369c4108cf4f6dd5f60960f4 + 2 + 0 + 2 + 134 + cd130938a2844050be991af70baf5ee0 + + + TECHNIQUE, METHOD + Secondary extraction prompt is an additional set of instructions given to the LLM for extracting covariates + bc9e2c9e369c4108cf4f6dd5f60960f4 + 2 + 0 + 2 + 135 + 43544b99c3b04b059546198a0ae6366d + + + METHOD + A covariate prompt is used to extract additional attributes associated with detected entities, including claims linked to the entities + 2c6ed90897310eea2f28e33fff1c32b0 + 9 + 0 + 2 + 136 + a671bf7fea2f4514b6e96ba99127fafd + + + CONCEPT + Claims are statements or assertions linked to detected entities, including details such as subject, object, type, description, source text span, and start and end dates + 2c6ed90897310eea2f28e33fff1c32b0 + 9 + 0 + 7 + 137 + 525f41ea20274a05af4e52b625b473f3 + + + METHOD + Gleanings refer to multiple rounds of entity extraction to ensure that no entities are missed in the process + 2c6ed90897310eea2f28e33fff1c32b0 + 2 + 0 + 1 + 138 + 071a416efbec4f0886c19ac68f6d43cb + + + TECHNIQUE + Logit bias is a technique used to force a yes/no decision from the LLM during the entity extraction process + 2c6ed90897310eea2f28e33fff1c32b0 + 2 + 0 + 1 + 139 + 6d8473ef3b1042bf87178a611e3dbcc6 + + + CONCEPT + An entity node is a representation of an entity in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 8 + 0 + 1 + 140 + 30c9641543c24773938bd8ec57ea98ab + + + CONCEPT + A relationship edge is a representation of a relationship between entities in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 8 + 0 + 1 + 141 + 18b839da898e4026b81727d759d95c6a + + + CONCEPT + A claim covariate is an additional attribute or variable associated with a claim in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 8 + 0 + 1 + 142 + eeef6ae5c464400c8755900b4f1ac37a + + + CONCEPT + Communities of entities are groups of closely-related entities detected and summarized by the LLM + 2c6ed90897310eea2f28e33fff1c32b0 + 2 + 0 + 2 + 143 + 422433aa45804c7ebb973b2fafce5da6 + + + CONCEPT + The "NOISY GRAPH STRUCTURE" refers to a graph structure that may contain inconsistencies or errors, making it challenging to analyze. This type of graph often includes duplicate or inconsistent entity elements due to variations in text format. These inconsistencies can arise from various sources, such as data entry errors, differing data formats, or incomplete information, which complicate the process of extracting meaningful insights and relationships from the graph. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56 + 2 + 0 + 1 + 144 + 86505bca739d4bccaaa1a8e0f3baffdc + + + DOMAIN + Science is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + 2 + 0 + 1 + 145 + 1af9faf341e14a5bbf4ddc9080e8dc0b + + + DOMAIN + Medicine is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + 2 + 0 + 1 + 146 + 353d91abc68648639d65a549e59b5cf3 + + + DOMAIN + Law is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + 2 + 0 + 1 + 147 + 7ce637e4f35b42e3a9f8272cab69cd22 + + + ATTRIBUTE + Source text span is an attribute of a claim that indicates the specific portion of text from which the claim was extracted + 2c6ed90897310eea2f28e33fff1c32b0 + 9 + 0 + 1 + 148 + 4d999d7744b04a998475f8f8531589f0 + + + ATTRIBUTE + Start date is an attribute of a claim that indicates when the event or fact described in the claim began + 2c6ed90897310eea2f28e33fff1c32b0 + 9 + 0 + 1 + 149 + 9a6f414210e14841a5b0e661aedc898d + + + ATTRIBUTE + End date is an attribute of a claim that indicates when the event or fact described in the claim ended + 2c6ed90897310eea2f28e33fff1c32b0 + 9 + 0 + 1 + 150 + db541b7260974db8bac94e953009f60e + + + ATTRIBUTE + Description is an attribute of a claim that provides a detailed explanation of the claim + 2c6ed90897310eea2f28e33fff1c32b0 + 9 + 0 + 1 + 151 + f2ff8044718648e18acef16dd9a65436 + + + ATTRIBUTE + Subject is an attribute of a claim that indicates the main entity involved in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + 9 + 0 + 1 + 152 + 00d785e7d76b47ec81b508e768d40584 + + + ATTRIBUTE + Object is an attribute of a claim that indicates the entity that is affected by the subject in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + 9 + 0 + 1 + 153 + 87915637da3e474c9349bd0ae604bd95 + + + CONCEPT + A concept referring to an entity that has multiple name variations but is resilient to such variations due to sufficient connectivity to closely-related entities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1 + 154 + 8f1eba29f39e411188200bf0d14628ec + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models capable of understanding and generating human-like text. They are utilized for a variety of tasks, including the creation and completion of knowledge graphs, which are essential for structuring and interlinking information in a meaningful way. Additionally, LLMs serve as evaluators of natural language generation, assessing the quality and coherence of text produced by other AI systems. These models play a crucial role in the field of Natural Language Processing and Information Retrieval, contributing significantly to advancements in how machines comprehend and interact with human language. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf + 2 + 0 + 5 + 155 + 7282c73622b8408e97289d959faff483 + + + TECHNOLOGY + Structured representations of knowledge in the form of triples (subject, predicate, object) used for reasoning tasks + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 0 + 0 + 1 + 156 + 3deb220d31f74103aa44870a36a63220 + + + CONCEPT + Nodes in a graph that are of the same type and are described using rich descriptive text + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 2 + 157 + af7a1584dd15492cb9a4940e285f57fc + + + CONCEPT + Edges in a graph that represent relationships between entity nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 2 + 158 + 6e8d9029ce4e4ea182367173ab2c7bbf + + + METRIC + Weights assigned to edges in a graph, representing the normalized counts of detected relationship instances + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1 + 159 + cbf232211e7d4eb6abdbe182f71c2cf0 + + + CONCEPT + The "HIERARCHICAL COMMUNITY STRUCTURE" is a framework in which communities are organized in a hierarchy, with each level providing a partition of the graph nodes. This structure organizes data into a hierarchy of communities, facilitating a multi-level clustering approach. Hierarchical community structure is utilized to generate community summaries, offering a comprehensive method for understanding the relationships and structural dynamics within specialized communities. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39 + 12 + 0 + 7 + 160 + bb0cff774a4440b289cc6f3b929fe13c + + + CONCEPT + A division of graph nodes into mutually-exclusive, collectively-exhaustive communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 7 + 0 + 2 + 161 + ce55841ebfdd47008bab8c258f10372e + + + TECHNOLOGY + MULTIHOP-RAG is a benchmark dataset comprising news articles published from September 2013 to December 2023, covering a range of categories including entertainment, business, sports, technology, health, and science. It is specifically designed for open-domain question answering, targeting explicit fact retrieval. Additionally, MULTIHOP-RAG represents a specific implementation or variant of Retrieval-Augmented Generation (RAG) used in the context of graph communities. This dataset is also utilized for community detection and analysis, making it a versatile tool in the field of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,922778ce1cb2fdd6dbab1746c8795620 + 8 + 0 + 8 + 162 + 6090e736374d45fd84f0e4610a314f8f + + + PERSON + An author who has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 8 + 0 + 1 + 163 + 0e8d921ccd8d4a8594b65b7fd19f7120 + + + PERSON + Authors who have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 8 + 0 + 1 + 164 + 59c726a8792d443e84ab052cb7942b4a + + + CONCEPT + The entity "DATASET" refers to a collection of data used for various purposes such as analysis, summarization, and evaluation. This can include diverse types of data like podcast transcripts and news articles. Specifically, the term encompasses datasets used for evaluation purposes, including notable examples like the Podcast and News datasets. + 1d07b4248c2655081c7af0e373bd70c9,7fb7d9ce2da9c940a32afdd87d1d9e56,973164fa90bf2b4ee267f4fd795916bf + 16 + 0 + 3 + 165 + 4f2c665decf242b0bfcaf7350b0e02ed + + + CONCEPT + GLOBAL QUERIES refer to questions or inquiries that require comprehensive answers derived from multiple sources or datasets. These queries aim to retrieve information from a global perspective, covering the entire dataset. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 2 + 166 + 66cdf168f36d4a57a505028c97dc06e0 + + + CONCEPT + ROOT COMMUNITIES are the top-level clusters in a hierarchical community structure. These communities represent the highest level of organization within the hierarchy, serving as the primary divisions from which more specific sub-communities branch out. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 4 + 167 + 38f51478f41f48db9bee570859b6f43e + + + CONCEPT + SUB-COMMUNITIES are lower-level clusters within root communities in a hierarchical community structure, providing more detailed information. These sub-communities play a crucial role in breaking down the larger, more general root communities into more specific and focused groups, thereby facilitating a deeper and more granular understanding of the overall community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 5 + 168 + 896d2a51e8de47de85ba8ced108c3d53 + + + TECHNOLOGY + Detailed documents that provide information about specific subtopics within a community + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 12 + 0 + 1 + 169 + 14555b518e954637b83aa762dc03164e + + + CONCEPT + The division of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 12 + 0 + 2 + 170 + b1f6164116d44fe8b8f135d7f65b9e58 + + + CONCEPT + A system in which elements are ranked or organized in levels + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 12 + 0 + 3 + 171 + c8b2408617804483b620e1a6691ac90d + + + CONCEPT + LEVEL 0 represents the root-level communities in the hierarchical clustering with maximum modularity. It serves as the foundational layer in a hierarchical community structure, indicating the initial and most significant division of the dataset into distinct groups. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 2 + 172 + a5e0d1644eb547ba9a5c3211aac4631a + + + CONCEPT + LEVEL 1 is a sub-level in a hierarchical community structure, providing more detailed information about the internal organization. Specifically, Level 1 represents sub-communities within the root-level communities, thereby revealing the internal structure and dynamics of these larger groups. This level of granularity helps in understanding the intricate relationships and specialized interactions that occur within the broader community framework. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 2 + 173 + 5a28b94bc63b44edb30c54748fd14f15 + + + CONCEPT + A visual representation of graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 8 + 0 + 1 + 174 + f97011b2a99d44648e18d517e1eae15c + + + METHOD + The Leiden algorithm is a method used for detecting communities in large networks + 843fc5421e086120ffa1c75856ecf6cd + 8 + 0 + 1 + 175 + 35489ca6a63b47d6a8913cf333818bc1 + + + TOOL + OpenORD is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + 8 + 0 + 1 + 176 + 5d3344f45e654d2c808481672f2f08dd + + + TOOL + Force Atlas 2 is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + 8 + 0 + 1 + 177 + 6fb57f83baec45c9b30490ee991f433f + + + ELEMENT + Nodes represent entities in a graph, with size proportional to their degree + 843fc5421e086120ffa1c75856ecf6cd + 8 + 0 + 2 + 178 + 68762e6f0d1c41cd857c6b964a8e76c3 + + + ELEMENT + Edges represent connections between nodes in a graph + 843fc5421e086120ffa1c75856ecf6cd + 8 + 0 + 2 + 179 + 70634e10a5e845aa8c6a32fe7e8eb2b2 + + + ELEMENT + Covariates are variables that are linked to nodes and edges in a graph + 843fc5421e086120ffa1c75856ecf6cd + 8 + 0 + 2 + 180 + 04085f7cf46544b79597fc49286ff84d + + + CONCEPT + The LLM context window is the token limit within which summaries are added for processing by a language model + 843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 2 + 181 + d203efdbfb2f4b2a899abfb31cf72e82 + + + METHOD + Hierarchical clustering is a method of clustering data into a tree-like structure with multiple levels + 843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 2 + 182 + 6731a665561840c2898ce8c9788e4c88 + + + CONCEPT + The token limit is the maximum number of tokens that can be processed in a single context window by a language model + 843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 1 + 183 + 4026806fa92f4e849a59a7f5c9a45c79 + + + CONCEPT + Summary detail refers to the level of detail provided in a summary + 843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 1 + 184 + 68e0c60d2e8845d89d9d0ad397833648 + + + CONCEPT + Scope refers to the range or extent of information covered in a summary + 843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 1 + 185 + 101572f552b54e529fe7765c05168981 + + + CONCEPT + A "USER QUERY" is a question or inquiry posed by a user seeking information, which the system aims to answer. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd + CONCEPT + 12 + 0 + 2 + 186 + 60c58026b2764b40adffca6eaa31d6d9 + + + ELEMENT + Chunks are segments of community summaries divided into pre-specified token sizes + 843fc5421e086120ffa1c75856ecf6cd + ELEMENT + 12 + 0 + 1 + 187 + ad1595a78935472999444c9330e7730e + + + METRIC + Prominence is a metric used to prioritize community edges based on the combined degree of source and target nodes + 843fc5421e086120ffa1c75856ecf6cd + 2 + 188 + 735d19aea0744b2295556841c5c4c3fd + + + METRIC + Combined source and target node degree is a metric used to measure the overall prominence of community edges + 843fc5421e086120ffa1c75856ecf6cd + 1 + 189 + c725babdb14a485582f8fbdf95429030 + + + ELEMENT + Community edges are connections between nodes within a community, prioritized based on prominence + 843fc5421e086120ffa1c75856ecf6cd + 1 + 190 + a0047221896d418d849847d422fa4bb8 + + + CONCEPT + Sub-community summaries are shorter summaries of sub-communities used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + 8 + 0 + 1 + 191 + 98fc2ee593184c5a839454db4eec7013 + + + CONCEPT + Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 1 + 192 + 80020a1da63042459e00266b2a605452 + + + CATEGORY + Community level refers to the different levels in the hierarchical community structure used to generate summaries + 843fc5421e086120ffa1c75856ecf6cd + 12 + 0 + 1 + 193 + 31a7e680c4d54101afe4c8d52d246913 + + + DATA + Chunks are segments of community summaries divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + 12 + 0 + 3 + 194 + 351abba16e5c448994c6daf48121b14d + + + METRIC + A score between 0-100 generated by the LLM to indicate how helpful an answer is in addressing the target question + 1d07b4248c2655081c7af0e373bd70c9 + 12 + 0 + 3 + 195 + 50ea7d3b69614bcdbfbff7ddbfbf3d34 + + + USER + A user looking for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + 10 + 0 + 6 + 196 + 004f40a5aeca48a1879db728eb12bcba + + + USER + A user incorporating current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + 14 + 0 + 6 + 197 + 4465efb7f6ed4dedad72a658184addd2 + + + TOPIC + A topic dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + 10 + 0 + 1 + 198 + b0dd60e11dad4ff782623acf039b3948 + + + TOPIC + A topic discussing the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + 10 + 0 + 1 + 199 + db8c43fa4df947b09e5754d3b1393ead + + + TOPIC + A topic discussing the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + 10 + 0 + 1 + 200 + 5dabc4cd05da425cb194a04482bf0c29 + + + TOPIC + A topic discussing suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + 10 + 0 + 1 + 201 + 9d08f285a7be4c79b8f359c51d51db37 + + + TOPIC + A topic discussing collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + 10 + 0 + 1 + 202 + adffed660d154b519c1817e514e83096 + + + TOPIC + Current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + 14 + 0 + 1 + 203 + b7e9c9ef572c445a9574ca571e41fb96 + + + TOPIC + A topic addressing the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + 14 + 0 + 1 + 204 + dcb9f281cd6248c699e0ebb285a42a5e + + + TOPIC + Examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + 14 + 0 + 1 + 205 + 072cdee531b74513984f49d99a8d64a0 + + + TOPIC + Insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + 14 + 0 + 1 + 206 + 5ae335d9210a45fda3f92a9a028d6d9b + + + TOPIC + The importance of health literacy highlighted through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + 14 + 0 + 1 + 207 + 5ac60a941a5b4934bdc43d2f87de601c + + + OUTPUT + Answers generated for each chunk of community summaries + 1d07b4248c2655081c7af0e373bd70c9 + 12 + 0 + 3 + 208 + d405c3154d0e48ce96fad4c28fe20590 + + + METRIC + The pre-specified size of tokens used to divide community summaries into chunks + 1d07b4248c2655081c7af0e373bd70c9 + 12 + 0 + 1 + 209 + 7923d8521c744bd9aab131c1aea91ffd + + + TECHNOLOGY + The "CONTEXT WINDOW" refers to a window of text used to generate answers, constrained by token size. The size of the context window is consistent across all conditions, ensuring uniformity in answer generation processes. + 1d07b4248c2655081c7af0e373bd70c9,973164fa90bf2b4ee267f4fd795916bf + 12 + 0 + 2 + 210 + 5bd156c87ec44e19ae6f8f62e6e50b9d + + + PERSON + Kevin Scott is the Chief Technology Officer (CTO) of Microsoft and actively participates in podcast conversations. His involvement in these discussions is documented and compiled in the dataset, highlighting his contributions to the field of technology and his role in shaping Microsoft's strategic direction. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + 0 + 0 + 2 + 211 + c1a146d7fb16429ea6d0aa2a55ee597f + + + PERSON + Individuals who are leaders in the technology industry and participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + 0 + 0 + 1 + 212 + ede9350632084da5b0b577ff799ab14b + + + INPUT + A specific activity or goal that the user aims to achieve using the datasets + 1d07b4248c2655081c7af0e373bd70c9 + 16 + 0 + 2 + 213 + ed559fb4ebde45518849ec803b350fa3 + + + INPUT + QUESTIONS refer to specific inquiries generated by the Large Language Model (LLM) based on the user's task and the target datasets. These questions are utilized in the analysis to evaluate the performance of different methods within the domain of Natural Language Processing and Information Retrieval. The generation and subsequent use of these questions are crucial for assessing the effectiveness and accuracy of various computational techniques and models. + 1d07b4248c2655081c7af0e373bd70c9,4c855404ee3d3c94aa2136f1513c666f + 16 + 0 + 4 + 214 + f422035f8b78417f98e4d116971cf9f3 + + + + + 1d07b4248c2655081c7af0e373bd70c9 + 16 + 0 + 1 + 215 + c79d686eba044c5586c706cdc096817d + + + DATASET + MT-BENCH is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical tool in evaluating the performance of models in retrieving accurate and relevant information from a broad range of topics. MT-Bench is prominently featured in the academic paper titled "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," where it is utilized to assess the capabilities of large language models in a structured and rigorous manner. This dataset plays a significant role in advancing the field of Natural Language Processing and Information Retrieval by providing a standardized metric for model comparison and evaluation. + 922778ce1cb2fdd6dbab1746c8795620,b1bbda43309e8e0e2175ea034aa88e13 + DATASET + 6 + 0 + 12 + 216 + 0f70db1e598d463fbbcdd1e288bd9490 + + + PROCESS + The process through which people inspect, engage with, and contextualize data within the broader scope of real-world activities + 922778ce1cb2fdd6dbab1746c8795620 + PROCESS + 1 + 217 + b35c3d1a7daa4924b6bdb58bc69c354d + + + TECHNOLOGY + Retrieval-Augmented Generation systems used for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + TECHNOLOGY + 0 + 0 + 3 + 218 + a97e2ecd870944cfbe71c79bc0fcc752 + + + AUTHORS + Authors of a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 1 + 219 + 3e1b063bbfa9423d84e50311296d2f3c + + + AUTHORS + Authors of a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 1 + 220 + 9a8ce816ee954bdabd01ea2081538009 + + + AUTHORS + Authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 8 + 0 + 1 + 221 + 09f18f81442d4d6d93a90f0fac683f9b + + + AUTHORS + Authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 3 + 0 + 1 + 222 + e02be3e37ca0454883a4c1fd859c24bb + + + AUTHORS + Authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 6 + 0 + 1 + 223 + 6e0c81bef5364c988b21bf9b709d9861 + + + + + 922778ce1cb2fdd6dbab1746c8795620 + 1 + 224 + 1dbc51475cb04dafa4a8833a8378635e + + + PODCAST + "BEHIND THE TECH" is a podcast series featuring conversations between Kevin Scott and other technology leaders. It serves as a media platform associated with Kevin Scott, providing insights and discussions on various technological advancements and industry trends. + 833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620 + 0 + 225 + c12b9ebd8b4e42b7896822a32e3fa6eb + + + PERSON + Kevin Scott, Microsoft CTO, who participates in the podcast conversations compiled in the dataset + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 226 + 27505f6ade4b4e5f9316ffe9c34821f7 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 227 + 0ee7db2c6bea4630ba9f0c25e8a967ad + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 228 + 5a6c1d15424149f69052cd8d91fbff75 + + + DATASET + A benchmark dataset for open-domain question answering, targeting explicit fact retrieval + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 229 + d005bf75c31d4848ad7041f39651e59c + + + METRIC + N represents the number of test questions per dataset used in the evaluation + 973164fa90bf2b4ee267f4fd795916bf + 16 + 0 + 1 + 230 + 9b3eef8f3a3a45e6873838db95295b8a + + + METHOD + A method applying a map-reduce approach directly to source texts for summarization + 973164fa90bf2b4ee267f4fd795916bf + 12 + 0 + 3 + 231 + fdc954b454744820804d7798f3e0b5de + + + METHOD + A na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached + 973164fa90bf2b4ee267f4fd795916bf + 0 + 0 + 2 + 232 + 49c1383836934ec495c3b35769100a73 + + + CATEGORY + C0 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a root-level community summary, which is utilized to answer user queries by providing the fewest number of summaries. This category is essential for understanding the structural dynamics within the community, particularly in the domain of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 0 + 0 + 8 + 233 + 859dedcc3736439a8a563419f16cb3d8 + + + CATEGORY + C1 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a high-level community summary used to answer user queries, effectively representing sub-communities of C0. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 0 + 0 + 9 + 234 + 6078b9980a6c4dcd9198d151b833ead7 + + + CATEGORY + C2 is a category or condition used in the analysis, representing a specific subset of the data. It functions as an intermediate-level community summary used to answer user queries, representing sub-communities of C1. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 0 + 0 + 9 + 235 + f93cd6b8213e46dda67af7e5382e1bd2 + + + CATEGORY + C3 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a category or condition that encapsulates low-level community summaries, which are instrumental in answering user queries. These summaries represent sub-communities of C2, providing detailed insights into the structural dynamics and relationships within the broader community. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 0 + 0 + 9 + 236 + 496f17c2f74244c681db1b23c7a39c0c + + + METHOD + TS, or "Text Summarization," is a category or condition used in the analysis, representing a specific subset of the data. It is particularly focused on source text summarization within the analysis. TS employs a text summarization method that applies a map-reduce approach directly to source texts, facilitating efficient and scalable summarization processes. This category is integral to understanding and processing large volumes of text data, making it a crucial component in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 0 + 0 + 10 + 237 + da1684437ab04f23adac28ff70bd8429 + + + METHOD + "SS" is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a baseline condition and is associated with a na¨ıve RAG (Retrieval-Augmented Generation) approach. In this context, text chunks are retrieved and added to the context window until the token limit is reached. + 4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 0 + 0 + 4 + 238 + 4517768fc4e24bd2a790be0e08a7856e + + + CONCEPT + The prompts used for answer generation, which are the same across all conditions with minor modifications + 973164fa90bf2b4ee267f4fd795916bf + 12 + 0 + 1 + 239 + 545edff337344e518f68d1301d745455 + + + DATASET + The "PODCAST DATASET" is a collection of podcast transcripts utilized for both analysis and evaluation purposes. This dataset is specifically designed to support various analytical tasks, providing a rich source of textual data for researchers and practitioners in the field of Natural Language Processing and Information Retrieval. The transcripts within the dataset offer valuable insights and serve as a critical resource for evaluating different computational models and techniques. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + 0 + 0 + 15 + 240 + 9376ce8940e647a99e5e087514b88fa4 + + + DATASET + The "NEWS DATASET" is a collection of news articles utilized for both analysis and evaluation purposes. This dataset serves as a valuable resource for examining and assessing various aspects of news content, making it an essential tool in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + 0 + 0 + 15 + 241 + b38a636e86984600bb4b57c2e2df9747 + + + CONCEPT + METRICS in the context of Natural Language Processing and Information Retrieval are essential tools used to evaluate the performance of natural language generation. These metrics include both reference-based metrics, which compare generated texts to a set of reference texts, and qualities of the generated texts themselves. They are crucial in the analysis to assess the effectiveness of different methods in generating natural language, ensuring that the outputs are both accurate and of high quality. + 4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + 16 + 0 + 3 + 242 + 4bc7440b8f4b4e4cae65a5c49defa923 + + + REFERENCE + "WANG ET AL., 2023A" refers to a study conducted by Wang and colleagues in 2023, which highlights the effectiveness of Large Language Models (LLMs) in evaluation. This study is a significant contribution to the field, providing insights into the capabilities and performance of LLMs in various evaluative tasks. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 2 + 0 + 2 + 243 + 5d1b038ce8be4533b54dd79d6496de9b + + + REFERENCE + "ZHENG ET AL., 2024" refers to a study conducted by Zheng and colleagues in 2024. This study highlights the effectiveness of Large Language Models (LLMs) in evaluation processes. The research, authored by Zheng et al., provides significant insights into the capabilities and applications of LLMs within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 2 + 0 + 3 + 244 + ac6e5a44e0c04a4fa93589376fde4c34 + + + REFERENCE + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + 16 + 0 + 1 + 245 + 40e4ef7dbc98473ba311bd837859a62a + + + CONCEPT + The entity "CONDITIONS" refers to the different scenarios or variables that are compared in an experiment. Specifically, in the context of the analysis, these conditions include Graph RAG, text summarization, and semantic search RAG. These conditions are used to evaluate and compare various aspects of performance and effectiveness within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 7 + 0 + 3 + 246 + 222f0ea8a5684123a7045986640ec844 + + + CONCEPT + USER QUERIES refer to the inquiries made by users to retrieve information. These queries are answered using different methods and conditions, depending on the context and the specific requirements of the information retrieval process. + 973164fa90bf2b4ee267f4fd795916bf,e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 0 + 6 + 247 + 668cf1fdfd644d39acc6350b86117ea2 + + + CONCEPT + Types of entities extracted during the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + 0 + 0 + 1 + 248 + 478e4c72d8fb46dd8cc9f0691c9878fd + + + METRIC + The "CONTEXT WINDOW SIZE" refers to the fixed size of the context window used in various stages of natural language processing and information retrieval tasks. For the final evaluation, the context window size is set to 8k tokens. During the analysis phase, different context window sizes are tested, including 8k, 16k, 32k, and 64k tokens. Additionally, in the graph indexing process, the context window size is set to 600 tokens. This variability in context window sizes highlights the importance of adapting the window size to the specific requirements of different tasks within the domain. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + 3 + 0 + 9 + 249 + 82b0446e7c9d4fc793f7b97f890e9049 + + + CONCEPT + The process of extracting information, with 1 gleaning for the Podcast dataset and 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + 0 + 0 + 2 + 250 + 8169efeea3ce473d9fd2f1c688126a1c + + + TECHNOLOGY + Natural Language Generation (NLG) is a subfield of artificial intelligence that focuses on generating human-like text from data + 322e02986c8724eedbcf3ebfa20b989c + 2 + 0 + 7 + 251 + c2d48b75af6a4d7989ccf9eceabd934e + + + METHOD + A method where a Large Language Model (LLM) is used to compare and evaluate competing outputs + 322e02986c8724eedbcf3ebfa20b989c + 2 + 0 + 1 + 252 + 5f1fc373a8f34050a5f7dbd8ac852c1b + + + METHOD + A method for evaluating the performance of Retrieval-Augmented Generation (RAG) systems, focusing on context relevance, faithfulness, and answer relevance + 322e02986c8724eedbcf3ebfa20b989c + 4 + 253 + 0c010fa3aeac4b28b2fbb8c2339c2521 + + + PUBLICATION + A reference to a study or paper authored by Es and others in 2023 + 322e02986c8724eedbcf3ebfa20b989c + 1 + 254 + c2999bdca08a478b84b10219875b285e + + + TOOL + A Large Language Model (LLM) used to evaluate and compare generated texts based on specific metrics + 322e02986c8724eedbcf3ebfa20b989c + 15 + 0 + 11 + 255 + 263d07354a1b4336b462024288f9bcd3 + + + METRIC + DIRECTNESS is a metric that measures how specifically and clearly an answer addresses a question. It is used to evaluate the straightforwardness of the generated answers. Additionally, it serves as a validity test metric to measure the directness of responses, with naive RAG (Retrieval-Augmented Generation) producing the most direct responses. + 322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 15 + 0 + 3 + 256 + f9005e5c01b44bb489f7112322fd1162 + + + DATA + An example of LLM-generated assessment shown in a table format + 322e02986c8724eedbcf3ebfa20b989c + 15 + 0 + 1 + 257 + d9ef017549724f4fbc4ff4ba6701dac0 + + + DATA + The entity "QUESTION" refers to a specific query used in the evaluation process, particularly as a metric to evaluate the generated responses by asking specific questions. This approach is commonly employed in the domain of Natural Language Processing and Information Retrieval to assess the quality and relevance of responses generated by various models or systems. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 2 + 258 + 33b9e826af3f43838c07c847b6349497 + + + ENTITY + Public figures are individuals who have gained fame or notoriety in various sectors such as entertainment, sports, and digital media. These individuals are well-known in the entertainment industry and are frequently mentioned across various articles. Their prominence in public discourse spans multiple domains, reflecting their influence and recognition in society. + 322e02986c8724eedbcf3ebfa20b989c,718017a4871c909420f84b85b8ba969d + 7 + 0 + 5 + 259 + dbe9063124d047dc8d6fcaeadcda038f + + + DATASET + ENTERTAINMENT ARTICLES is a collection of articles focused on the entertainment industry. This dataset consists of articles related to various aspects of the entertainment sector, providing a comprehensive resource for understanding trends, developments, and key topics within this field. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 6 + 260 + c885166d0c454a748376b56279f96408 + + + DOMAIN + The **ENTERTAINMENT INDUSTRY** is a multifaceted sector that encompasses various forms of entertainment, including movies, music, television, sports, and digital media. This industry is characterized by its diverse range of content and mediums, which collectively contribute to its broad appeal and significant cultural impact. The entertainment industry plays a crucial role in shaping public opinion, trends, and cultural norms through its extensive reach and influence across different platforms and genres. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 19 + 261 + 586bccefb1e344289c1ee984e165de9c + + + METRIC + A metric indicating the highest level of development or achievement in a particular field + 322e02986c8724eedbcf3ebfa20b989c + 2 + 0 + 1 + 262 + a2201b8753ba4847ab0b22054e27d2c0 + + + METRIC + A metric indicating results that are comparable to or better than those of others in the same field + 322e02986c8724eedbcf3ebfa20b989c + 2 + 0 + 1 + 263 + b5ecd0553dd742f5813c9b855d548a41 + + + METRIC + A metric based on evaluations made by humans + 322e02986c8724eedbcf3ebfa20b989c + 2 + 0 + 1 + 264 + 89b2003e97804961805ea1886d078ebd + + + METRIC + Metrics that require a gold standard or reference answers for evaluation + 322e02986c8724eedbcf3ebfa20b989c + 2 + 0 + 2 + 265 + 6dd7f5f6b4544271a97f6a136f82fc3d + + + METHOD + An evaluation method that does not require reference answers + 322e02986c8724eedbcf3ebfa20b989c + 2 + 0 + 1 + 266 + eb01db8435554f2cbafe39a50f62f20a + + + METRIC + A metric that measures how relevant the generated text is to the given context + 322e02986c8724eedbcf3ebfa20b989c + 1 + 267 + 3d175ad1f0014cd4871eff4e86db9f88 + + + METRIC + A metric that measures how accurately the generated text reflects the source information + 322e02986c8724eedbcf3ebfa20b989c + 1 + 268 + c8e706fbdc90420d952deed03c4f04b4 + + + METRIC + A metric that measures how relevant the generated answer is to the question + 322e02986c8724eedbcf3ebfa20b989c + 1 + 269 + cf6115e69d6649cc99ef2bd11854ccfb + + + METHOD + A method involving multiple stages or steps + 322e02986c8724eedbcf3ebfa20b989c + 7 + 0 + 1 + 270 + 9ed7e3d187b94ab0a90830b17d66615e + + + DATA + The correct or ideal answers used as a benchmark in evaluations + 322e02986c8724eedbcf3ebfa20b989c + 2 + 0 + 2 + 271 + b4c7432f712849d7aba9dccbb77471ef + + + DATA + "SENSEMAKING QUESTIONS" are a class of questions used to evaluate the performance of Retrieval-Augmented Generation (RAG) systems. These questions are specifically designed to help users understand and make sense of complex information, as well as to validate the understanding and interpretation of data. By employing sensemaking questions, researchers and practitioners can assess how effectively a RAG system can retrieve and generate relevant information, thereby ensuring that the system aids in the comprehension and accurate interpretation of intricate datasets. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 0 + 3 + 272 + 434e752b992c4e6a812557529315c5b9 + + + METHOD + A method where two items are directly compared against each other + 322e02986c8724eedbcf3ebfa20b989c + 15 + 0 + 1 + 273 + df79a27b9a4f42fd839c90bb8a79ad91 + + + DATA + TARGET METRICS are specific measures used to evaluate the performance of RAG systems. These metrics are aimed to be achieved or measured in the analysis and are the focus of an evaluation. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 0 + 3 + 274 + 8f140fd7126f47b6b00307b0181509f9 + + + DATA + A metric used as a baseline or standard for comparison + 322e02986c8724eedbcf3ebfa20b989c + 15 + 0 + 2 + 275 + 40450f2c91944a81944621b94f190b49 + + + METRIC + A metric that measures the accuracy and reliability of a method or result + 322e02986c8724eedbcf3ebfa20b989c + 15 + 0 + 1 + 276 + 5b9fa6a959294dc29c8420b2d7d3096f + + + METRIC + A metric that measures the randomness or variability in a process + 322e02986c8724eedbcf3ebfa20b989c + 15 + 0 + 1 + 277 + b84d71ed9c3b45819eb3205fd28e13a0 + + + DATA + The average scores obtained from multiple evaluations + 322e02986c8724eedbcf3ebfa20b989c + 15 + 0 + 1 + 278 + b0b464bc92a541e48547fe9738378dab + + + PERSON + Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 6 + 279 + 44c65dda6fb7472dae36f6eea720ab47 + + + PERSON + Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 6 + 280 + 5d97ff82691c4482973d73d1860e4757 + + + PERSON + Britney Spears is a public figure frequently mentioned in entertainment articles, known for her significant contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 6 + 281 + 2567445079794d1e84f17abc48776002 + + + PERSON + Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his significant contributions to the music industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 6 + 282 + 392be891f8b649fabdc20e7bf549f669 + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in film and television + e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 1 + 283 + 0111777c4e9e4260ab2e5ddea7cbcf58 + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in music + e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 1 + 284 + 785f7f32471c439e89601ab81c828d1d + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in sports + e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 1 + 285 + 6768339b54084020aec27adcef8994ff + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in digital media and business + e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 1 + 286 + f09f381c319f4251847d1a4bb8cdcac1 + + + CATEGORY + A category of public figures in the entertainment industry who are involved in controversies + e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 1 + 287 + eec11f567e7f4943b157c3a657eb9a46 + + + METRIC + A metric used to determine the winner in the comparison of generated responses + e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 3 + 288 + efef117839b64ce9adf614a461d41ba6 + + + METRIC + A metric used to evaluate the quality of LLM-generated responses + e8c8f911135faf3ff35f24107eb3f99c + 2 + 0 + 1 + 289 + 2171091ada0942d8ae7944df11659f6e + + + SECTOR + The entity "FILM" refers to a sector within the entertainment industry that encompasses movies and cinema. This sector includes public figures involved in the movie industry, such as actors, directors, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 2 + 290 + bcfdc48e5f044e1d84c5d217c1992d4b + + + SECTOR + The entity "TELEVISION" refers to a sector within the entertainment industry that encompasses TV shows and series. This sector includes public figures involved in TV shows, such as actors, hosts, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 2 + 291 + b232fb0f2ac14790b931d1e7fcddd8ad + + + SECTOR + MUSIC is a sector within the entertainment industry that encompasses musical performances and recordings. This sector includes public figures involved in the music industry, such as singers, musicians, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 6 + 292 + 1c16b22e18d3483b8d41b284754274e2 + + + SECTOR + The entity "SPORTS" refers to a sector within the entertainment industry that encompasses athletic events and competitions. This sector includes public figures involved in sports, such as athletes, coaches, and sports commentators. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 4 + 293 + 0080f96708cd4054a5f0986ca86889f4 + + + SECTOR + DIGITAL MEDIA is a sector within the entertainment industry that encompasses online content and social media. This sector includes public figures involved in online platforms, such as influencers, content creators, and digital marketers. These individuals play a significant role in shaping digital landscapes through their engagement with audiences and their ability to leverage various online tools and platforms for content dissemination and marketing purposes. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 2 + 294 + e683130322ac47708a852a5e51abb7c5 + + + CATEGORY + A category within the entertainment industry that includes stories and themes that shape culture + e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 1 + 295 + 71a0a8c1beb64da08124205e9a803d98 + + + CATEGORY + A category within the entertainment industry that includes popular movements and styles + e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 1 + 296 + f84314943bee4c859c9a62f268c9c216 + + + CATEGORY + A category within the entertainment industry that includes public conversations and debates + e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 1 + 297 + ba481175ee1d4329bf07757a30abd3a1 + + + CATEGORY + A category within the entertainment industry that includes formal discussions and communications + e8c8f911135faf3ff35f24107eb3f99c + 7 + 0 + 1 + 298 + 8d8da35190bf43c5878fa38f3eb4f3d2 + + + RESPONSE + Answer 1 provides a varied and rich response by covering a wide range of public figures from different sectors of the entertainment industry, including film, television, music, sports, gaming, and digital media. It offers insights into the contributions and influence of these figures, as well as controversies and their impact on public discourse. The answer also cites specific data sources for each mentioned figure, indicating a diverse range of evidence to support the claims. + 718017a4871c909420f84b85b8ba969d + 7 + 0 + 11 + 299 + 2fb7e14a3f124526bd7b24867fc18e81 + + + RESPONSE + "ANSWER 2" is a generated answer for the example question in the News article dataset. It focuses on a smaller group of public figures, primarily from the music industry and sports, and relies heavily on a single source for data. "ANSWER 2" provides concise explanations for the frequent mentions of specific public figures such as Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + 7 + 0 + 10 + 300 + 5c13c7d61e6c4bfe839f21e7ad3530a7 + + + METHOD + Naïve RAG is a baseline method for retrieval-augmented generation (RAG) that does not use advanced techniques. It is a basic form of RAG with certain drawbacks that advanced RAG systems aim to overcome. Naïve RAG is used to generate answers for questions in the News article dataset and to generate responses that directly list specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d,e4d9b12cf2b4c691c74019eefff4fb39,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19 + 2 + 0 + 4 + 301 + a621663edba64d99b7e50f1e53f32ee7 + + + DATASET + The "NEWS ARTICLE DATASET" is a collection of news articles utilized for various analytical purposes. This dataset is specifically employed for generating responses to questions about public figures in the entertainment industry, making it a valuable resource for both analysis and information retrieval tasks within this domain. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + 0 + 0 + 16 + 302 + 42be4e140061482ea509dd3e26189480 + + + TOPIC + Controversies are events or issues involving public figures that generate public debate and impact public discourse. + 718017a4871c909420f84b85b8ba969d + 7 + 0 + 2 + 303 + 4da4ef951ff340f1a3dd679de4be3341 + + + SECTOR + The gaming sector includes public figures involved in the gaming industry, including gamers, developers, and streamers. + 718017a4871c909420f84b85b8ba969d + 7 + 0 + 1 + 304 + 2f05fcce857e4a499ca4e89a3cefbcb3 + + + RESOURCE + Data sources are references or reports used to support claims about public figures and their influence. + 718017a4871c909420f84b85b8ba969d + 7 + 0 + 2 + 305 + b3aeb7ae009a4f52ae3ae4586e32fe11 + + + METHOD + Assessments generated by large language models (LLMs) to evaluate the answers produced by different methods + ebf5249c888e07fedce6572a4c03f88c + 0 + 0 + 1 + 306 + 089b9b9841714b8da043777e2cda3767 + + + DATASET + An example question used in the News article dataset for analysis + ebf5249c888e07fedce6572a4c03f88c + 0 + 0 + 1 + 307 + 38f1e44579d0437dac1203c34678d3c3 + + + DATA + The datasets used in the analysis, consisting of various text sources + 4c855404ee3d3c94aa2136f1513c666f + 16 + 0 + 2 + 308 + 1ca24718a96b47f3a8855550506c4b41 + + + METRIC + A metric used to compare the performance of different conditions in the analysis + 4c855404ee3d3c94aa2136f1513c666f + 13 + 0 + 1 + 309 + 9c980dfe3cab44b7a83408405edab0b6 + + + CATEGORY + A specific setup or scenario used in the analysis, such as C0, C1, C2, C3, TS, and SS + 4c855404ee3d3c94aa2136f1513c666f + 13 + 0 + 4 + 310 + f23484b1b45d44c3b7847e1906dddd37 + + + METRIC + WIN RATE is a measure used to evaluate the success rate of different methods in providing comprehensive and diverse answers. It represents the percentage of times a particular approach or method achieves a win in a given context. Specifically, it quantifies the percentage of times a condition outperformed another in the analysis. This metric is crucial in assessing the effectiveness of various strategies within the domain of Natural Language Processing and Information Retrieval, offering insights into the comparative performance of different techniques. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4 + 13 + 0 + 3 + 311 + 929f30875e1744b49e7b416eaf5a790c + + + METRIC + The condition that performed the best across all comparisons in the analysis + 4c855404ee3d3c94aa2136f1513c666f + 13 + 0 + 1 + 312 + 4920fda031804ce8a1073ace8e061ed6 + + + METRIC + The expected win rate of a condition when compared to itself, shown as 50% for reference + 4c855404ee3d3c94aa2136f1513c666f + 13 + 0 + 1 + 313 + 4b8aa4587c7344adac2cbfa69d5e40fa + + + METHOD + The use of large language models (LLMs) at the time of querying, evaluated in the analysis + 4c855404ee3d3c94aa2136f1513c666f + 3 + 0 + 1 + 314 + 52701d941dfb45359693baae8f267056 + + + METHOD + The "FINAL EVALUATION" is the last stage of the analysis where the best performing context window size was used. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + 3 + 0 + 3 + 315 + 31499ee6277a4d71b19cb5b6be554c69 + + + PROCESS + The process that resulted in the creation of graphs for the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + 0 + 0 + 1 + 316 + d99eabad5dfd47278692569d2a9395b1 + + + STRUCTURE + A data structure consisting of nodes and edges, used to represent the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + 0 + 0 + 4 + 317 + d53f15cb7f7845de91cc44ad44ff9f6e + + + METHOD + Approaches that consistently outperformed the naive RAG in comprehensiveness and diversity metrics + 36db32c37e1987e2c5863898ad882190 + 7 + 0 + 3 + 318 + 23becf8c6fca4f47a53ec4883d4bf63f + + + METRIC + The number of context units, such as community summaries or text chunks, used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + 0 + 0 + 4 + 319 + d0ffa3bcd1234258953ff4956d19f561 + + + METRIC + The term "TOKENS" refers to the number of individual words used in the analysis. The evaluation typically focuses on corpora in the region of 1 million tokens. This metric is crucial for understanding the scope and scale of the text data being analyzed, particularly in the fields of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,92e93fc6449756c0a60200636b297f65 + METRIC + 0 + 0 + 7 + 320 + ac41b77ba33c4c84877eb425aba03aa1 + + + METRIC + The percentage of the maximum token count used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + 0 + 0 + 4 + 321 + 5d3184dabfd647a5a7e565f72c60ff24 + + + METHOD + MAP-REDUCE SUMMARIZATION is a method for summarizing source texts using a map-reduce approach. This summarization technique is notably resource-intensive, necessitating the highest number of context tokens compared to other methods. The map-reduce framework, originally popularized for its efficiency in processing large-scale data, is adapted here to handle the complexities of text summarization, ensuring comprehensive and accurate extraction of key information from extensive source texts. + 36db32c37e1987e2c5863898ad882190,e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 0 + 2 + 322 + 0ec262c2cfef4dd581f3655e5e496e31 + + + DATA + Summaries at the root level of the community hierarchy, requiring dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + 0 + 0 + 1 + 323 + 100c2fccd7f74d9281707082f062ba72 + + + DATASET + SOURCE TEXTS are the original texts from which summaries or analyses are derived. These texts serve as the foundational material used for comparison with community summaries in the analysis. + 6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 0 + 2 + 324 + 378fc7636eeb4aabbfd40995a6960c64 + + + REFERENCE + A reference to a paper by Ram et al. in 2023 discussing RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + 2 + 0 + 1 + 325 + 80a04aa18cd649d584292f23b10c0727 + + + REFERENCE + "GAO ET AL., 2023" is a paper published in 2023 by Gao et al. that delves into advanced Retrieval-Augmented Generation (RAG) techniques, specifically where the index is a knowledge graph. The publication also touches upon naive RAG approaches, providing a comprehensive examination of both advanced and basic methodologies within the domain of Natural Language Processing and Information Retrieval. + 6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + 7 + 0 + 3 + 326 + 4e9ca18ccc1d4527a3bc035d07f5e162 + + + CATEGORY + Intermediate-level summaries are a type of community summary used in the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 0 + 0 + 1 + 327 + 5564257e89f1428486a64fcf52f49490 + + + CATEGORY + Root-level summaries are a type of community summary used in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 7 + 0 + 1 + 328 + 83c76fbd2a004d90a5b0a6736ffed61d + + + METRIC + Answer comprehensiveness is a measure used to evaluate the completeness of answers provided by different methods + 6f33a085ff3304e5994f7fbb86c881a4 + 7 + 0 + 1 + 329 + d9779c41e3c74fe0b26e23822a4b995b + + + TECHNOLOGY + Ad-hoc LLM use refers to the spontaneous use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + 2 + 0 + 1 + 330 + 9d7a563b3b2d405092c31f1fe08cff77 + + + TECHNOLOGY + Element extraction prompts are used to extract specific details in the Graph RAG index + 6f33a085ff3304e5994f7fbb86c881a4 + 7 + 0 + 1 + 331 + bd43f3d439a54781bd4b721a9a269b92 + + + CONCEPT, TECHNOLOGY + A mathematical space in which text chunks and queries are embedded to represent similar semantics + f35de4d9fb65f1d5a392064b20545c19 + 1 + 332 + adc0f95733e74351a891c4dadf650a52 + + + CONCEPT, DATA + Search inputs that are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + 1 + 333 + 225105a7be14447cb03186bd40756059 + + + TECHNOLOGY, METHOD + A type of RAG system that includes patterns for iterative and dynamic cycles of interleaved retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 2 + 0 + 1 + 334 + efce8a9d61254447a26aee99e53f0398 + + + TECHNOLOGY, CONCEPT + A concept related to generation-augmented retrieval that facilitates future generation cycles + f35de4d9fb65f1d5a392064b20545c19 + 7 + 0 + 2 + 335 + 4a75a9f0b18a48bea9c0601c0fc395c4 + + + TECHNOLOGY, METHOD + A method that facilitates future generation cycles by using self-memory + f35de4d9fb65f1d5a392064b20545c19 + 7 + 0 + 1 + 336 + e19287afe00a431f9a593a4827d1b448 + + + TECHNOLOGY, METHOD + A strategy for iterative retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 7 + 0 + 1 + 337 + f2c06f3a0c704296bf3353b91ee8af47 + + + TECHNOLOGY, METHOD + A federated strategy for retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 7 + 0 + 1 + 338 + f512103ed4624accac6cbbf90d7d250a + + + TECHNOLOGY, METHOD + A method that combines multiple concepts for summarizing multiple documents + f35de4d9fb65f1d5a392064b20545c19 + 7 + 0 + 2 + 339 + 2325dafe50d1435cbee8ebcaa69688df + + + TECHNOLOGY, METHOD + A method for answering questions that require multiple steps or "hops" to gather information + f35de4d9fb65f1d5a392064b20545c19 + 19 + 0 + 4 + 340 + 469aeef98cd1421fa123277b93d7b83a + + + TECHNOLOGY, METHOD + An approach that involves generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 7 + 0 + 2 + 341 + 2fb66f9a0de6406d83b61742a3b52cd6 + + + TECHNOLOGY, METHOD + A method for generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 7 + 0 + 1 + 342 + b0e6cfd979ea48b997019b059999d3c2 + + + TECHNOLOGY, METHOD + A method for answering multiple interpretations of ambiguous questions by generating a hierarchical structure + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 7 + 0 + 1 + 343 + ef00ec3a324f4f5986141401002af3f6 + + + TECHNOLOGY, METHOD + A process that involves using LLMs to create knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 2 + 0 + 1 + 344 + a542fd7aed7341468028928937ea2983 + + + TECHNOLOGY, METHOD + A process that involves using LLMs to complete existing knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 2 + 0 + 1 + 345 + 1c5e296a5ac541c1b5cac4357537c22d + + + TECHNOLOGY, METHOD + Graphs that represent causal relationships, which can be extracted using LLMs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 2 + 0 + 1 + 346 + 5ecf534a9ffe46e0b1c2144110c691c0 + + + REFERENCE, PUBLICATION + A reference to a publication by Cheng et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 7 + 0 + 1 + 347 + 4d183e7007624fcd98af96b9d752c16d + + + REFERENCE, PUBLICATION + A reference to a publication by Mao et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 7 + 0 + 1 + 348 + 718c507cb8ac49e6a35c251ac951b5ca + + + REFERENCE, PUBLICATION + A reference to a publication by Shao et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 7 + 0 + 1 + 349 + b45ef27279c043269b23b894461d7d8c + + + REFERENCE, PUBLICATION + A reference to a publication by Wang et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 7 + 0 + 1 + 350 + 10983a248cc448c59c94df4d1d0898f0 + + + REFERENCE, PUBLICATION + A reference to a publication by Su et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 7 + 0 + 1 + 351 + e2ec7d3cdbeb4dd086ae6eb399332363 + + + REFERENCE, PUBLICATION + A reference to a publication by Feng et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 7 + 0 + 1 + 352 + 67f10971666240ea930f3b875aabdc1a + + + REFERENCE, PUBLICATION + A reference to a publication by Trivedi et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 7 + 0 + 1 + 353 + 8b95083939ad4771b57a97c2d5805f36 + + + REFERENCE, PUBLICATION + A reference to a publication by Khattab et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 7 + 0 + 1 + 354 + 3c4062de44d64870a3cc5913d5769244 + + + REFERENCE, PUBLICATION + A reference to a publication by Sarthi et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 7 + 0 + 1 + 355 + 24652fab20d84381b112b8491de2887e + + + REFERENCE, PUBLICATION + A reference to a publication by Kim et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 7 + 0 + 1 + 356 + d4602d4a27b34358baa86814a3836d68 + + + REFERENCE, PUBLICATION + "TRAJANOSKA ET AL., 2023" refers to a paper by Trajanoska et al. published in 2023, which focuses on using Large Language Models (LLMs) for knowledge graph creation. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting innovative methodologies for leveraging advanced language models to construct and enhance knowledge graphs. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 2 + 0 + 2 + 357 + 36be44627ece444284f9e759b8cd25c6 + + + REFERENCE, PUBLICATION + "Yao et al., 2023" refers to a paper published by Yao and colleagues in 2023. The study focuses on the application of large language models (LLMs) for the task of knowledge graph completion. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting the potential of advanced LLMs to enhance the accuracy and efficiency of knowledge graph completion processes. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 2 + 0 + 2 + 358 + a64b4b17b07a44e4b1ac33580d811936 + + + REFERENCE, PUBLICATION + A reference to a publication by Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 2 + 0 + 1 + 359 + 423b72bbd56f4caa98f3328202c1c3c9 + + + TECHNOLOGY, METHOD + A system that combines multiple concepts for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 7 + 0 + 1 + 360 + 5c7ef01f46a94641bf1ae5cd25f8a538 + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 19 + 0 + 1 + 361 + aefde1f7617f4c0e9aed31db77f6d862 + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 19 + 0 + 1 + 362 + ad52ba79a84748a49067e53b1d5095f9 + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 19 + 0 + 1 + 363 + 289616058bf4495887292003b27ba216 + + + TECHNOLOGY, METHOD + Strategies used before the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + 0 + 364 + 7ffa3a064bce468082739c5a164df5a3 + + + TECHNOLOGY, METHOD + Strategies used during the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + 0 + 365 + ce36d1d637cf4a4e93f5e37ffbc6bd76 + + + TECHNOLOGY, METHOD + Strategies used after the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + 0 + 366 + eeb9c02c0efa4131b9e95d33c31019fc + + + TECHNOLOGY, METHOD + A pattern in Modular RAG systems for iterative and dynamic cycles of retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 0 + 367 + 7b2472c5dd9949c58828413387b94659 + + + TECHNOLOGY, METHOD + Cycles of generation that are facilitated by self-memory in Graph RAG + f35de4d9fb65f1d5a392064b20545c19 + 0 + 368 + bdddcb17ba6c408599dd395ce64f960a + + + PUBLICATION + A paper by Ban et al. published in 2023, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 2 + 0 + 1 + 369 + bc70fee2061541148833d19e86f225b3 + + + PUBLICATION + A paper by Zhang et al. published in 2024, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 2 + 0 + 1 + 370 + 0fc15cc3b44c4142a770feb4c037a6f7 + + + METHOD + A method where the index is a knowledge graph, developed by Baek et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 2 + 0 + 2 + 371 + a24e9df02e1b4b43bf6324b039e28285 + + + PUBLICATION + A paper by Baek et al. published in 2023, focusing on the KAPING method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 2 + 0 + 1 + 372 + ab3a5a6713244fd595a1ace978c3d960 + + + METHOD + A method where subsets of the graph structure are the objects of enquiry, developed by He et al. in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + 2 + 0 + 2 + 373 + 02a88c0d128e4586b2f1f64329786d3c + + + PUBLICATION + A paper by He et al. published in 2024, focusing on the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 2 + 0 + 1 + 374 + 1ca41537c47c4752a17a44d1d7086d96 + + + METHOD + A method where derived graph metrics are the objects of enquiry, developed by Zhang in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 2 + 0 + 2 + 375 + 7e0d14ca308b4796bdc675a64bd3a36e + + + PUBLICATION + A paper by Zhang published in 2023, focusing on the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 2 + 0 + 1 + 376 + 8323efc8e539419e9ca3c98e758f6609 + + + METHOD + A method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, developed by Kang et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 2 + 0 + 2 + 377 + a80c7c98c0b647f8b9f6f8cc09168e44 + + + PUBLICATION + A paper by Kang et al. published in 2023, focusing on the SURGE method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 2 + 0 + 1 + 378 + 2d66a15939294d21b83b3e277f0a4e46 + + + METHOD + A method where retrieved event-plot subgraphs are serialized using narrative templates, developed by Ranade and Joshi in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 2 + 0 + 2 + 379 + 47f6d6573cf34e1096c95e36251dd60c + + + PUBLICATION + A paper by Ranade and Joshi published in 2023, focusing on the FABULA method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 2 + 0 + 1 + 380 + 2fbd74d5ccca4be99c5257b3ac95cfba + + + PUBLICATION + A paper by Wang et al. published in 2023, focusing on a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 2 + 0 + 1 + 381 + a2b1621a3e424ae29a6a73f00edbeca3 + + + ORGANIZATION + LangChain is an organization that developed Langchain graphs and supports a variety of graph databases. + 71f6daf11e64e5273a3847d46bf228e1,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + 18 + 0 + 5 + 382 + ec45e1c400654c4f875046926486ded7 + + + ORGANIZATION + LlamaIndex is an organization that developed the LlamaIndex Knowledge Graph Index and supports a variety of graph databases. + 6cd82819982879bd164547d2773ba5c7,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + 18 + 0 + 3 + 383 + 047cd93e9d704c7d8dadb6e79f9458df + + + TECHNOLOGY + Neo4J is both a graph database format supported by various Retrieval-Augmented Generation (RAG) applications and an organization that developed Project NaLLM. The graph database format of Neo4J is widely recognized for its efficiency in handling complex relationships and structures, making it a valuable tool in the field of Natural Language Processing and Information Retrieval. As an organization, Neo4J has contributed significantly to the advancement of these domains through innovative projects like NaLLM, which further underscores its pivotal role in the community. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + 18 + 0 + 4 + 384 + 5b71ee73a5b6484495b2a0a75219426c + + + METHOD + A method that can create and reason over knowledge graphs in Neo4J format, developed by Neo4J in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + 18 + 0 + 1 + 385 + e1f524d4b9754ce2b64a0a4c8f73b854 + + + TECHNOLOGY + NebulaGraph is both a graph database format and an organization that has made significant contributions to the field of graph databases and retrieval-augmented generation (RAG) applications. As a graph database format, NebulaGraph is supported by various RAG applications, facilitating the efficient handling and querying of complex graph data structures. Additionally, NebulaGraph, as an organization, has pioneered the industry-first graph RAG, which integrates retrieval-augmented generation with large language models (LLMs) based on knowledge graphs. This innovation underscores NebulaGraph's role in advancing the capabilities of knowledge graph-based applications and enhancing the performance of LLMs in generating contextually relevant information. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + 18 + 0 + 4 + 386 + ae1fe1c014c54ec4bcdf10dbdaed5068 + + + METHOD + A method that can create and reason over knowledge graphs in NebulaGraph format, developed by NebulaGraph in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + 18 + 0 + 1 + 387 + 92646910ee624bd7909fac2b5c0232e3 + + + METHOD + A method for comparing fabrication rates, developed by Manakul et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 1 + 388 + 05913bee89a94bca88449249e35ba74d + + + PUBLICATION + "MANAKUL ET AL., 2023" refers to a paper by Manakul et al. published in 2023, which focuses on the SelfCheckGPT method. This work by Manakul and colleagues is centered around the development and application of SelfCheckGPT, a technique likely aimed at enhancing the performance and reliability of GPT models. The paper contributes to the field of Natural Language Processing and Information Retrieval by addressing specific challenges and proposing innovative solutions through the SelfCheckGPT method. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + PUBLICATION + 2 + 389 + 57b8930790c34dcba4a32c6be703ed78 + + + STAKEHOLDER + END USERS are individuals who are the final users of the system or analysis. They play a crucial role in validating sensemaking questions and target metrics, ensuring that the system or analysis meets the intended objectives and provides meaningful insights. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + STAKEHOLDER + 2 + 0 + 2 + 390 + 838c4498bc3c437f8d65428b580766a2 + + + CONCEPT + Considerations and compromises involved in building a graph index + 92e93fc6449756c0a60200636b297f65 + CONCEPT + 2 + 0 + 1 + 391 + 1b893f24eb98477aad6ce49c0f26737e + + + METRIC + The effectiveness of RAG systems, which varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + METRIC + 2 + 0 + 1 + 392 + 6573bc2af4f94596a3f4452a602d6fc4 + + + CONCEPT + Various forms of data used in RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + 2 + 0 + 1 + 393 + 0dddcca0e5df4b16bc03a51a2d2d8e16 + + + METRIC + The scale of datasets used in RAG systems, which affects performance + 92e93fc6449756c0a60200636b297f65 + METRIC + 2 + 0 + 1 + 394 + df40ad480a3c47299a6c8fad05349304 + + + PROCESS + The process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + PROCESS + 2 + 0 + 1 + 395 + fe98fb197d294b0b837aee8d5a98dfb1 + + + DATASET + Collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + DATASET + 2 + 0 + 1 + 396 + feb9ddd0ac2949178f26a36949aa5422 + + + CONCEPT + Different categories of questions used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + 2 + 0 + 1 + 397 + b4e4fa2e3dfc46e68d532d659b18d17d + + + METHOD + SelfCheckGPT is an approach used to compare fabrication rates in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 398 + f58813d090b947a48c1b4614b92c3ec3 + + + METHOD + A method for global summarization of source texts that does not use a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 0 + 2 + 399 + 30a251bc3d04430d82b5a1a98c7b8c75 + + + RESOURCE + The amount of computational resources allocated for a task + e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 400 + 93e1d19f9bfa4c6b8962d56d10ea9483 + + + METRIC + The expected number of queries over the lifetime of a dataset + e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 401 + 8046335ba70b434aa3188392a746fd78 + + + DATA + Annotations that provide detailed information about the text + e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 0 + 1 + 402 + 5c02b1ab32064c64a0f8b27b219e358a + + + METHOD + A method that uses embeddings to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 0 + 3 + 403 + c5f77ba0c261408780db3d50346f16b7 + + + METHOD + RAG schemes that combine embedding-based matching with other approaches + e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 0 + 2 + 404 + 453ecf5476f64f4a8d5020b95baf1314 + + + METHOD + Mechanisms used in map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 0 + 2 + 405 + 6a1d83c9ce2b483dbd7de5ab3ae2487d + + + DATA + A hierarchical organization of communities + e4d9b12cf2b4c691c74019eefff4fb39 + 11 + 0 + 2 + 406 + 66c3dffb7d7a4fa8bb6b48a22ca917a6 + + + METHOD + A method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) for global text summarization + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 0 + 2 + 407 + 6f3dd1fd6d7f4df4af0656ed0525c92e + + + METRIC + The cost associated with the number of tokens used in a text generation task + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 0 + 1 + 408 + 711eb39432794b0a91110358dd536517 + + + TECHNOLOGY + An implementation of Graph RAG approaches using the Python programming language + e4d9b12cf2b4c691c74019eefff4fb39 + 17 + 0 + 2 + 409 + 0e00585b08044954a254116665400463 + + + PERSON + A person who contributed to the work mentioned in the acknowledgements + e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 0 + 1 + 410 + db0147eff2204a20b5e5e6bec7a8bae5 + + + METRIC + The rates at which fabrications occur in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 411 + 67bb4f4678284819add02ba04f3b1103 + + + METRIC + The expected number of queries over the lifetime of a specific dataset + e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 0 + 1 + 412 + 2033ec0487f04240abb3bdbe77b39087 + + + METRIC + The benefits or value obtained from using a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 0 + 1 + 413 + f026fab8fec948ae9e7baa2ad715e6ef + + + METHOD + Different methods related to retrieval-augmented generation that utilize graph structures + e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 0 + 1 + 414 + d0d7ed36d6f54b5d986dfd854096b728 + + + METHOD + Graph RAG approaches that operate in a more localized manner + e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 0 + 1 + 415 + bf6a4c18f44042799eb7456a6b85b54a + + + DATA + Annotations made on the graph to provide additional information + e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 0 + 1 + 416 + fac4a59c2278498d83f9f1b4231ad62e + + + DATA + Reports generated from community summaries + e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 0 + 1 + 417 + d6d2b5862ddc4c4d87deee3423506817 + + + METHOD + An operation that aggregates information across multiple levels of a hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 0 + 1 + 418 + 47d588d26e2b4cccb68fe2af4c147c8f + + + METHOD + A mechanism that allows for exploring detailed information by following higher-level summaries + e4d9b12cf2b4c691c74019eefff4fb39 + 11 + 0 + 2 + 419 + c0f2dc03d8df400db4997c1a0babd6ad + + + DATA + The trail of information that guides users to more detailed data + e4d9b12cf2b4c691c74019eefff4fb39 + 11 + 0 + 1 + 420 + 0211d61aae834229a3a1e004ff5cc658 + + + DATA + The top-level communities in a hierarchical structure + e4d9b12cf2b4c691c74019eefff4fb39 + 12 + 0 + 1 + 421 + ccbbbcc055c34709abcf103208c2c299 + + + DATA + A graph index organized around entities + e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 0 + 1 + 422 + 989add81cf874018a569239b68d17ff2 + + + TECHNOLOGY + A publicly available implementation of a technology + e4d9b12cf2b4c691c74019eefff4fb39 + 17 + 0 + 1 + 423 + fd7d94fbab084bc380480abeef6bfade + + + PERSON + Alonso Guevara Fernández is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 21 + 424 + cfb915c95caf41c6a25e99a9f37f03a2 + + + PERSON + Amber Hoak is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 19 + 425 + 8815ed80f9b741dbb458d902024f34a4 + + + PERSON + Andrés Morales Esquivel is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 426 + dddb831546354e088d29aebd154e3a31 + + + PERSON + Ben Cutler is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 427 + 005d2154da754b21adcd90ac921bd5f7 + + + PERSON + Billie Rinaldi is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 428 + 711ba818354546cea69f1532b92a2f26 + + + PERSON + Chris Sanchez is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 429 + 5c4d8a8f9c104176b87d2bfdf04ae0bd + + + PERSON + Chris Trevino is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 430 + 5a781604f1fb4719b730f43f534627f6 + + + PERSON + Christine Caggiano is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 431 + ecdc1020b10e49ca869d399825e16fa3 + + + PERSON + David Tittsworth is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 432 + 0d8fde01d7234726a00d7e73e2e01d66 + + + PERSON + Dayenne de Souza is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 433 + 9c4bd60958fd4e09a6d5b9e2ab163b5a + + + PERSON + Douglas Orbaker is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 434 + 39d31f770cf740e78d526a2e1101a1db + + + PERSON + Ed Clark is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 435 + 9d282b2250f7408888504f1f93c202a8 + + + PERSON + Gabriel Nieves-Ponce is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 436 + c063484895794a0eaae1b0ff070ad4c9 + + + PERSON + Gaudy Blanco Meneses is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 437 + e8868920e21b4431aad16e86db977ecb + + + PERSON + Kate Lytvynets is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 438 + aea3378bfff842e5b3f4b7a4b55b3879 + + + PERSON + Katy Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 439 + d562223c17d948bf98e34b4d97dde932 + + + PERSON + Mónica Carvajal is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 440 + cde2d75c51d245879265b79d14b8699b + + + PERSON + Nathan Evans is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 441 + 44594467054849d4a1fadb46ddd51641 + + + PERSON + Richard Ortega is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 442 + 2918130221f94f4387da049b647bfe6a + + + PERSON + Rodrigo Racanicci is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 443 + fd139ac75b0e4777ab67b7423eaaa37f + + + PERSON + Sarah Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1 + 444 + a701c349eb7142d48ba7efad89caf9d2 + + + PERSON + Shane Solomon is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1 + 445 + e5d40a1b17f74b1db5d18279caedb04a + + + PUBLICATION + A technical report on GPT-4 published as an arXiv preprint + 086021a89900a39bcb62036981737bfa + 0 + 446 + de25d06733d04385825ee082792f5e52 + + + METHOD + A method for zero-shot knowledge graph question answering described in an arXiv preprint + 086021a89900a39bcb62036981737bfa + 0 + 447 + 32f6f11a7845416b8c6eb9fb0b382140 + + + METHOD + A method for harnessing large language models for advanced causal discovery from data + 086021a89900a39bcb62036981737bfa + 0 + 448 + 91407be8c3e54e23918d3a7183d962db + + + METHOD + A method incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models + 086021a89900a39bcb62036981737bfa + 0 + 449 + 3831134696584d83bbf676a6b3bfa8f9 + + + PERSON + J. Achiam is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 450 + 50e512a5dbe941f5af68bfdf74b1c3c0 + + + PERSON + S. Adler is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 451 + edc717747e904728b57185f5013461f9 + + + PERSON + S. Agarwal is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 452 + 8fba1fea719d49d380ac2d9c310d68b3 + + + PERSON + L. Ahmad is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 453 + 532da08f04f645708e747c57e9c4ee05 + + + PERSON + I. Akkaya is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 454 + 3cf0ab4cf14e47ddabd49d500a3dc488 + + + PERSON + F. L. Aleman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 455 + a39b72f8921f43ef8ef295c7cc8f7294 + + + PERSON + D. Almeida is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 456 + 9f5adbeb6cf04f089abe78d86cfa6aba + + + PERSON + J. Altenschmidt is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 457 + efb6350e65964659bc20396c0166b296 + + + PERSON + S. Altman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 458 + e095cc36da784300b27c6f8c60a96440 + + + PERSON + S. Anadkat is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 459 + c68893ca39d74ba08c6eb138f24441e1 + + + PERSON + R. Anil is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 460 + 472b23bb92834173b4118d101040c726 + + + PERSON + S. Borgeaud is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 461 + 81869985b45a4fefbbbb23ea118a3de4 + + + PERSON + Y. Wu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 462 + 42b8584c5a874eb08fbd61f0c18f3ca0 + + + PERSON + J.-B. Alayrac is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 463 + 824d93d9840a4b7c8b1f31bc6816b497 + + + PERSON + J. Yu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 464 + f209a808f1f04a5699601e672f4abd06 + + + PERSON + R. Soricut is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 465 + ccb335166f6c4564ac1c61549d8ded50 + + + PERSON + J. Schalkwyk is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 466 + cbe1a41a82aa4f268e8264568b25938f + + + PERSON + A. M. Dai is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 467 + 28e7639f55ce464c8a080cbb2c745fa2 + + + PERSON + A. Hauth is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 468 + 3f3a2d7aa1294116814f0b4d89baa23d + + + PERSON + J. Baek is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 2 + 469 + 3073b33926bd4f33807ffa3befacefaf + + + PERSON + A. F. Aji is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 2 + 470 + 2b916117691c4872a9c4e4888d4fe4ab + + + PERSON + A. Saffari is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 2 + 471 + 1f7b02bf486e4f42b23e9cb1a63207f3 + + + PERSON + T. Ban is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 1 + 472 + e744c118ae7f4638a01d060bbaedd6e9 + + + PERSON + L. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 1 + 473 + e1c1080c717d437996def1a41772d179 + + + PERSON + X. Wang is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 0 + 474 + 63fba9a7c47a4f14ac0bee6bc90d0fea + + + PERSON + H. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 0 + 475 + 6bfc2395b4f54a528a1ebac94a43acb8 + + + PERSON + T. Baumel is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 476 + 1cce5cebf437428eb1a60dffbdfa603f + + + PERSON + M. Eyal is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 477 + dc94039d6643460ca3c66150b9087129 + + + PERSON + M. Elhadad is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 478 + f197d75f159943f8a3ff441199790bc7 + + + PUBLICATION + The arXiv preprint identifier for the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0 + 479 + 4d8890c699684c9381105b03b0b41b03 + + + PUBLICATION + The arXiv preprint identifier for the Gemini paper + 086021a89900a39bcb62036981737bfa + 0 + 480 + b1658adfa43847eabad1437db235e858 + + + PUBLICATION + The arXiv preprint identifier for the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 0 + 481 + a1773cac7d4c4939aec965660e5015fe + + + PUBLICATION + The arXiv preprint identifier for the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 482 + 6a054cb59fb44cf494b93988b5f88833 + + + PERSON + Baumel, T. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 4 + 483 + e7b103a52e384e3e8bf14105223e7e82 + + + PERSON + Eyal, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 4 + 484 + 3f1042452c254cecaf7189e89162adc8 + + + PERSON + Elhadad, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 4 + 485 + fd31d549420744d1bd1a6b1112a9a6ba + + + PERSON + Blondel, V. D. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 486 + f7ab348030714072a277682b51f7c588 + + + PERSON + Guillaume, J.-L. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 487 + 2139b0906dc541e094138a978d070416 + + + PERSON + Lambiotte, R. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 488 + ff5466607e5d4453b1d833629292f664 + + + PERSON + Lefebvre, E. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 489 + 71f95003936e46a98d90757ffd845d40 + + + PUBLICATION + The journal where the paper "Fast unfolding of communities in large networks" was published + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 490 + bada987ea7da4c939393ee1c3d08ccd4 + + + PERSON + Brown, T. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9 + 491 + d0a274e7934d446fb91847bb53a961a6 + + + PERSON + Mann, B. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9 + 492 + 0a799eab61bc4e6b884db6689f9c2c4a + + + PERSON + Ryder, N. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9 + 493 + 8c34cd494a63438dac219c1dc0f73100 + + + PERSON + Subbiah, M. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 8 + 494 + c6f428af0c5e4f629902fd5455bf19ac + + + PERSON + Kaplan, J. D. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 495 + d1fd271d16c348019c2fcced762b35a2 + + + PERSON + Dhariwal, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 496 + ffa128c9c0c84d39bad1bba8cfa4adc5 + + + PERSON + Neelakantan, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 497 + 058f66cc356b43cc9433bd3c8d57fa46 + + + PERSON + Shyam, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 498 + ff74091eaba246698fcae59c21eec828 + + + PERSON + Sastry, G. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 499 + f6cbbf1b8f4b48a28a16e4dd8976b9bb + + + PERSON + Askell, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 500 + 757ca40654d5476aa949a26b733be8d4 + + + PUBLICATION + "ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS" is a prominent conference where significant papers in the field of Natural Language Processing and Information Retrieval are presented. Notable papers presented at this conference include "Language models are few-shot learners" and "Retrieval-augmented generation for knowledge-intensive NLP tasks." Additionally, it is also the journal where the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" was published. This conference and journal serve as key platforms for disseminating cutting-edge research in neural information processing systems. + 58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,b1bbda43309e8e0e2175ea034aa88e13 + 0 + 501 + 539d55e7c42e44b59d98f59fae3e0ee1 + + + PERSON + Cheng, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 502 + 3785eeadea9042bfb2e50f16c0397a12 + + + PERSON + Luo, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 503 + 48cd97f2297143e09d61ff2a8542c0c5 + + + PERSON + Chen, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 504 + ff95eb0d5f7f49b782027d5c7ae3c3fe + + + PERSON + Liu, L. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 505 + 086da554db5b4ad5806aedeb0024197c + + + PERSON + Zhao, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory"Zhao, D. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + PERSON + 4 + 0 + 2 + 506 + 216ee8a907a0466a88b27f8ada19ffa0 + + + PERSON + Yan, R. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 507 + 6fefb317687d4ac98efe39a52f3e190f + + + PERSON + Dang, H. T. is an author of the paper "Duc 2005: Evaluation of question-focused summarization systems" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 508 + 320d9d91238948a8be67972ccceab878 + + + PUBLICATION + The conference where the paper "Duc 2005: Evaluation of question-focused summarization systems" was presented + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 509 + bdcbcccadd474b3bbe9a8f56c811bab4 + + + PERSON + Es, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 2 + 510 + f127fc4d87f94794be89134406ba0694 + + + PERSON + James, J. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 2 + 511 + c27966a4e3be434686454204ac7b3ab4 + + + PERSON + Espinosa-Anke, L. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 2 + 512 + dab39f92d0ed468c80699f28c05c45fa + + + PERSON + Schockaert, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 2 + 513 + 3076f330d121489aa50964ce54a3b1ac + + + PERSON + Feng, Z. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 2 + 514 + c8e5d3afdcb54c8589e280f0c4a87417 + + + PERSON + Feng, X. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 2 + 515 + f3d30627e19245649e497ab49bf0fa30 + + + PERSON + Yang, M. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 2 + 516 + e3f1098c3d984bc7b5f30b9c0101f7a6 + + + PERSON + Qin, B. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 2 + 517 + 24b4a5f4db67418cbfa08c5316f0ab51 + + + PERSON + Fortunato, S. is an author of the paper "Community detection in graphs" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 518 + e4b707e3e6964197855b82fc66ef59e7 + + + PUBLICATION + The journal where the paper "Community detection in graphs" was published + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 519 + 109b8be5a8ee4180a1465cd23f019d7b + + + PERSON + Gao, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 520 + 49f771e31a0c4b35bc39e389f3623509 + + + PERSON + Xiong, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models. The paper provides a comprehensive survey of the methodologies and applications of retrieval-augmented generation, highlighting its significance in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 521 + aa946d4379694a74ba0da37e69d2810a + + + PERSON + Gao, X. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 522 + 268446fc52a54fd2837f73aeb3e0b74f + + + PERSON + Jia, K. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant area of research within the domains of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 523 + f6ddfa8491ff40d2839bb5b2e105df22 + + + PERSON + Pan, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 524 + db1295504da645b69d9786d54f233fed + + + PERSON + Bi, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 525 + 6ff4ed0dda4f4158af37be99f505565f + + + PERSON + Dai, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance and capabilities of large language models, a significant area of research within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 526 + 5d398b88ee4242a59c32feb188683ec3 + + + PERSON + Sun, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 527 + 0a784e00c9464bd3aeb830b908f73170 + + + PERSON + Wang, H. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 528 + b0966a0f455e44229e6c9705d57bfca9 + + + PUBLICATION + The arXiv identifier for the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 3 + 529 + 99761e9b89cc4060be3ed6b34532e7ff + + + PUBLICATION + The arXiv identifier for the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 4 + 530 + 8130a1a82bde46048952cf147690e630 + + + PUBLICATION + The arXiv identifier for the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 0 + 5 + 531 + 79c99026b7ef4946b9b8e0be841fd4c5 + + + PERSON + Goodwin, T. R. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 2 + 532 + fdcb1673254842f1935f53d0c38c467e + + + PERSON + Savery, M. E. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 1 + 533 + dcb3f4cc8abc46faabc193d9885e91d0 + + + PERSON + Demner-Fushman, D. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 1 + 534 + 3295be59128d451bb720c6688adc1e0b + + + CONFERENCE + COLING (International Conference on Computational Linguistics) is the conference where the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" was presented + 00e8e4e881bd0862022f4dfc913b900b + 0 + 535 + aca3eb8924ac494486fe0bfe892f7f2e + + + PERSON + He, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 536 + 66689accdd974295b7eb779e43578748 + + + PERSON + Tian, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 537 + 6b49c78aa1524609ab7aa74aeaa3e01d + + + PERSON + Sun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 538 + 7ff31ce54f424f0bbb297b0b3ba7c757 + + + PERSON + Chawla, N. V. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 539 + bac51e00d486420c8e91e824d8e17411 + + + PERSON + Laurent, T. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 540 + 4adee3aad6524a4aa4c4711c1ee05e64 + + + PERSON + LeCun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 541 + d034e4fd8ac849278e658daad1a1f033 + + + PERSON + Bresson, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 542 + 091e998370dd42d1b05ab0fcf6595a7e + + + PERSON + Hooi, B. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 543 + 1e6cabc18fab4c048281fd29d3044438 + + + PERSON + Jacomy, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 544 + dc08f6d7398b4b798a3bdccf508a2ad4 + + + PERSON + Venturini, T. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 545 + 1c7fd5af8d8041e186eae2431fc627cd + + + PERSON + Heymann, S. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 546 + b16eda56dcec40f2b3e109fb9246bee3 + + + PERSON + Bastian, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 547 + 43c68f9a86654a32a2215e23957ed184 + + + PUBLICATION + PLOS ONE is the journal where the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" was published + 00e8e4e881bd0862022f4dfc913b900b + 0 + 548 + 1ba06fe2e86140a59bbc4f4e969d0f71 + + + PERSON + Jin, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 549 + 36caa0a230c8422c8acb4dc62e35bb32 + + + PERSON + Yu, Z. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 550 + 09940fed9d154504948bba2df1789a50 + + + PERSON + Jiao, P. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 551 + 4d6608557eed49368a6d09c7c5c664c5 + + + PERSON + Pan, S. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 552 + eb7c93eeb9dc41aab57d29e97ebb4951 + + + PERSON + He, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 553 + 3b6e2ac584b64847b53828c9d779fed3 + + + PERSON + Wu, J. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 554 + e9b68002e035447baae848208cea5503 + + + PERSON + Philip, S. Y. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 555 + fe18353546824ca98294ce4be7b96e02 + + + PERSON + Zhang, W. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 556 + 0e9740e25f5a460c81318336e00ac880 + + + PUBLICATION + IEEE Transactions on Knowledge and Data Engineering is the journal where the paper "A survey of community detection approaches: From statistical modeling to deep learning" was published + 00e8e4e881bd0862022f4dfc913b900b + 0 + 557 + b7cd9a62710849778fdadced0d754687 + + + PERSON + Kang, M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 558 + 432a6b4962544200949421a96a405142 + + + PERSON + Kwak, J. M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 559 + d6700b360ac141d282cdb567414bf4ce + + + PERSON + Baek, J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 560 + c1b40a4039b44061a358e098867f7412 + + + PERSON + Hwang, S. J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 561 + 4643a7a319674adfb732b6f6122c7c64 + + + PERSON + Khattab, O. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 562 + 46e8056fb2ec4811ab33cb34a0dc9fb3 + + + PERSON + Santhanam, K. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 563 + 8b57a9f43a1942a49b58cf881835f974 + + + PERSON + Li, X. L. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 564 + f78b01b0d93948c283644ec58f7be74a + + + PERSON + Hall, D. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text, indicating its relevance within the domain of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 565 + 8dbe8f9867e4448f998416c18923eac4 + + + PERSON + Liang, P. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Liang, P. contributed to the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP," which explores the integration of retrieval and language models to enhance knowledge-intensive tasks in NLP. Additionally, Liang, P. authored the paper "Lost in the middle: How language models use long contexts," which investigates the utilization of extended contexts by language models. These contributions highlight Liang, P.'s significant role in advancing the understanding and application of language models in complex NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 7 + 566 + fe8ea8bf1395434393e04e8f7a33025f + + + PERSON + Potts, C. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 567 + 7d58b089bfc549e8951e91ad62541119 + + + PERSON + Zaharia, M. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 568 + 1fa6d3118bd846c8837b5fa9fb78f262 + + + PERSON + Kim, G. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 569 + 62c65bbae33c4ee9a21b61f6f454c4b4 + + + PERSON + Kim, S. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 570 + 30b7034c4468473f98ee18d00ee73b33 + + + PERSON + Jeon, B. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 571 + 00f78b85e5b84999a810e311e540037b + + + PERSON + Park, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 572 + 3e460d9f011d4b0b9ccaae7b6a5202de + + + PERSON + Kang, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 573 + 9d98dece22eb401aa1a5ce9c88c603f0 + + + PERSON + Klein, G. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 574 + 81446ea789b24eaf9eab02dc07c3d984 + + + PERSON + Moon, B. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 575 + 79f4b1c1b2be4cf7aa828846e20a4eb6 + + + PERSON + Hoffman, R. R. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 576 + de04830d6e414fd5b39a9e90769d9452 + + + PUBLICATION + The journal where the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" were published + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 577 + 69db426b97714835bf4937180774787a + + + PERSON + Koesten, L. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 578 + 9c7bc862339d4a5bb21ee5154d9b33bb + + + PERSON + Gregory, K. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 579 + 17bad53a0ebe4569839e5e151ff78593 + + + PERSON + Groth, P. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 580 + 53d98f08e7c74158b7318357b6c660b3 + + + PERSON + Simperl, E. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 581 + cd601f77419c403889aadeee591915b5 + + + PUBLICATION + The journal where the paper "Talking datasets–understanding data sensemaking behaviours" was published + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 582 + 0f564ebd53e940fba9d16674ac7bc038 + + + PERSON + Kuratov, Y. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 583 + 7deb75816e4f473480e0c79ae99b5bf4 + + + PERSON + Bulatov, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 584 + 7f85b181f1184f77aeb3ea2155cf4027 + + + PERSON + Anokhin, P. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 585 + d148b2b2033048618f1a090a492a40a5 + + + PERSON + Sorokin, D. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 586 + 4d839a10353e4144a26563b0966721d5 + + + PERSON + Sorokin, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 587 + 521a862bb196488389f17c0b0f4b6f4d + + + PERSON + Burtsev, M. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 588 + 22ea3328fb6343f4ad2862495ea27640 + + + TECHNOLOGY + Langchain graphs is a technology developed by LangChain + 71f6daf11e64e5273a3847d46bf228e1 + 18 + 0 + 1 + 589 + 3f9a2a2c1c0a424e8b4980ea9d48bdbe + + + PERSON + Laskar, M. T. R. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" and also contributed to the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models." Both works highlight Laskar's expertise in leveraging transformer models and transfer learning techniques to enhance the performance of query-focused abstractive text summarization, demonstrating a significant contribution to the field of Natural Language Processing and Information Retrieval. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 5 + 590 + aa2ec452728a4703ae1bdabe85b6c079 + + + PERSON + Hoque, E. is an author of two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning to improve the effectiveness of transformer models in query-focused abstractive summarization tasks. Both works contribute to advancing the understanding and application of transformer models in specialized summarization contexts. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 5 + 591 + c5ddb31e0a9c4b2683e4631283dd505b + + + PERSON + Huang, J. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 592 + 07d8eeb549044ac88d2e788c146a0ef1 + + + PUBLICATION + The conference where the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" was presented + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 593 + 47df2815030c4f1c99facd5cf2482526 + + + PUBLICATION + arXiv preprint refers to a preprint of a paper that is available on the arXiv repository + 71f6daf11e64e5273a3847d46bf228e1 + 18 + 594 + ae521508bdc244f99c4fce4ab5214c79 + + + EVENT + The 33rd Canadian Conference on Artificial Intelligence, held in Ottawa, ON, Canada, from May 13–15, 2020 + 6cd82819982879bd164547d2773ba5c7 + 2 + 595 + 6315b4bf135c40358823ed7e4e4060e2 + + + EVENT + The 2020 edition of the Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + 1 + 596 + 33905debec1a45ecae1c65daac1d854c + + + PUBLISHER + Springer is the publisher of the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + 1 + 597 + bfbe904780fe47daad1a04126b12923c + + + PERSON + Huang, J. X. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + 3 + 598 + 0614f00e932c4cd0b53928053811ebc1 + + + PUBLICATION + The journal where the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" was published + 6cd82819982879bd164547d2773ba5c7 + 3 + 599 + 9ef487dd0b574b108c60a56d6a2f146c + + + PERSON + Lewis, P. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9 + 600 + 4067269e7f6943cdbc299ce02b7eadbd + + + PERSON + Perez, E. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9 + 601 + 094a736ba43c4da48c556437f47f88d1 + + + PERSON + Piktus, A. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9 + 602 + 563c2af32bb3476299e9b24a646097ab + + + PERSON + Petroni, F. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks"Petroni, F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + PERSON + 9 + 603 + d59b49eb94ce442d89907e90c5d3a44e + + + PERSON + Karpukhin, V. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 5 + 604 + 8ea7cef407df48098046551e303e1c64 + + + PERSON + Goyal, N. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 5 + 605 + 186e60d2176547bf84e5bf87bd16bb40 + + + PERSON + Küttler, H. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 606 + e65017091c8d4c7daa45b6c8414e0465 + + + PERSON + Lewis, M. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 607 + a0f326b9597b49dda6563e9208316117 + + + PERSON + Yih, W.-T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 608 + bff3db70f9af4f2c87a93df48ecbb6bc + + + PERSON + Rocktäschel, T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 609 + bf91f36307cb43e1ab1e967cb3ba8274 + + + PERSON + Liu, N. F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 610 + cd58a8740ba54d86a77db9bb9544ef0d + + + PERSON + Lin, K. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 611 + e96d3475d43b42a781b297ae7e650afe + + + PERSON + Hewitt, J. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 612 + 1ce76a5547854d458878bd445f0ccbd6 + + + PERSON + Paranjape, A. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 613 + 11e4325f59394ff1bc89892f79288702 + + + PERSON + Bevilacqua, M. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 614 + 71743537a07c440ea1710a269da8b538 + + + PERSON + Liu, Y. is an author of the paper "Hierarchical transformers for multi-document summarization" + 6cd82819982879bd164547d2773ba5c7 + 0 + 615 + 1389192ce5464be6b3b5749bc9536709 + + + PERSON + Lapata, M. is an author known for significant contributions to the field of Natural Language Processing and Information Retrieval. Notably, Lapata, M. has authored the paper "Hierarchical transformers for multi-document summarization," which explores advanced techniques in summarizing information from multiple documents using hierarchical transformer models. Additionally, Lapata, M. has contributed to the paper "Text summarization with latent queries," which delves into innovative methods for summarizing text by leveraging latent query representations. These works highlight Lapata, M.'s expertise and active research in the domain of text summarization, showcasing a focus on developing sophisticated models and methodologies to enhance the efficiency and accuracy of summarization tasks. + 6cd82819982879bd164547d2773ba5c7,fc4b27d64f055b7fc30176ba110dd02e + 5 + 0 + 2 + 616 + b349041c0be64c62b964ab1234e055e6 + + + TECHNOLOGY + LlamaIndex Knowledge Graph Index is a technology developed by LlamaIndex + 6cd82819982879bd164547d2773ba5c7 + 0 + 617 + 969e1ea0b1e443a68e9a65dfef91d161 + + + PERSON + Manakul, P. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + 0 + 618 + 8e09e7cfea7d405db8b22ae2f836ccb1 + + + PERSON + Liusie, A. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + 0 + 619 + 490583524d394bf79289c5fe34f7dcf1 + + + PERSON + Gales, M. J. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + 0 + 620 + d7db38bb599c42cab7066f3fdd282282 + + + PERSON + Mao, Y. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 621 + efd87a59d01e47c8adc02f63ef2c5c3e + + + PERSON + He, P. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 622 + 80e3ce3de41e4601823a333e22b7bb3f + + + PERSON + Liu, X. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 623 + 50eabc166e8944a49197e79c32f27597 + + + PERSON + Shen, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Shen, Y.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 0 + 624 + 5197a3fb02ef4677abd1900aa87e4efa + + + PERSON + Gao, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 625 + 887f444240bb474da23cdfb6abf7a998 + + + PERSON + Han, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 626 + 5d29053f2ce74442aa1855b327ef3bb7 + + + PERSON + Chen, W. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Chen, W.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 627 + 7e40cd12839a4577a95e33d785147a31 + + + PERSON + Martin, S. is an author of the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing a comprehensive, open-source solution for the layout of large graphs, which is a critical task in the visualization and analysis of complex networks. The toolbox aims to facilitate the understanding and interpretation of large-scale graph data, making it a valuable resource for researchers and practitioners in fields such as computational linguistics, information retrieval, and data science. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 628 + 8fe58de8a04f4f8f807c77fb41829a3a + + + PERSON + Brown, W. M. is an author of the paper "Openord: An open-source toolbox for large graph layout." + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 629 + a9f50861273c4bb697d868a9d049d392 + + + PERSON + KLAVANS, R. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 630 + be4820f29fd942b282049fa49697b4ed + + + PERSON + Boyack, K. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on the development and application of Openord, a comprehensive open-source toolbox designed for the layout of large graphs. The paper likely discusses the methodologies, algorithms, and practical implementations of the toolbox, contributing to the fields of graph theory and data visualization. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 631 + 6deaefe707f84b3dbda979dea0d095ac + + + EVENT + The conference where the paper "Openord: An open-source toolbox for large graph layout" was presented + 833e7d67dcd30790b26b71c9b5306f6b + EVENT + 0 + 632 + d053ea9432a24fb192e8d6aa993b0caa + + + TECHNOLOGY + GPT-4 is a large language model used in Microsoft's study on scientific discovery + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + 3 + 0 + 1 + 633 + a3e683d294ed42a28d60d09a36cbeb54 + + + TECHNOLOGY + Project NaLLM is a project developed by Neo4J + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + 18 + 0 + 1 + 634 + 39887ca8567141d5b857b87a2bca4086 + + + PERSON + Newman, M. E. is the author of the paper "Modularity and community structure in networks" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 1 + 635 + 8df8563ab0394ee9a91b89dea7d59404 + + + PUBLICATION + The journal where the paper "Modularity and community structure in networks" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + 1 + 636 + 12398f70065143839d812fd42ac4b2e7 + + + PERSON + Ram, O. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 637 + 74d43d20f251441baf8e3db64fedca43 + + + PERSON + Levine, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 638 + 1b7a22f76f7741e8b140bdc3d8856d76 + + + PERSON + Dalmedigos, I. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 639 + b823ba1bfe944fa9887edd8faf8a5f17 + + + PERSON + Muhlgay, D. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 640 + d0bfb473fdc64643954cdb4675e2f389 + + + PERSON + Shashua, A. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 641 + a4db1b2a9c3e4d2d838725f8166c36b4 + + + PERSON + Leyton-Brown, K. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 642 + 8dae140578c841ae9373cbc607c4a6e6 + + + PERSON + Shoham, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 643 + b215cc33cf40434f87f284ff8f3506a4 + + + PUBLICATION + The journal where the paper "In-context retrieval-augmented language models" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + 0 + 644 + c1ff9d8e1b8745d6860c34ce26122d79 + + + PERSON + Ranade, P. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 1 + 645 + 9d1e6ca9ae8e4e068fb74631a633b20b + + + PERSON + Joshi, A. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 1 + 646 + 1d7b0deca7674777bf76c163ac065845 + + + PERSON + Sarthi, P. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 5 + 647 + 03afe9988f864c9fa501bfbf043f74c0 + + + PERSON + Abdullah, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 5 + 648 + 4084f614af494fa8ab73095fb5b6b07b + + + PERSON + Tuli, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 5 + 649 + 3ce25564af6e47f390a0b16b6f9433a1 + + + PERSON + Khanna, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 3 + 650 + 78213664d0eb45d1a9239ba4b85b10f7 + + + PERSON + Goldie, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 3 + 651 + 1226e4a4077b4b3a970db4d2509b590c + + + PERSON + Manning, C. D. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" and the paper "Raptor: Recursive abstractive processing for tree-organized retrieval". These contributions highlight Manning's involvement in advancing the fields of Natural Language Processing and Information Retrieval, particularly in the areas of multi-hop question answering and recursive abstractive processing. + 833e7d67dcd30790b26b71c9b5306f6b,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 4 + 652 + b4c7de7a824a4a71b9f52193d2f1a10d + + + PERSON + Scott, K. is associated with "Behind the Tech" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 0 + 653 + b609f1939dae4c7383c7d199bb3c7dc3 + + + PERSON + Shao, Z. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 0 + 654 + aeee2f443dfb4e3ea80af6ae1d9197ce + + + PERSON + Gong, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 0 + 655 + 8c46d37bc26e4d4dbd37d6ee26867bc6 + + + PERSON + Huang, M. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + 2 + 656 + 58a8fa7f29e347bdb9689b70b065a779 + + + PERSON + Duan, N. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + 2 + 657 + fae3fe31deb141ab93143ac411f1eaaa + + + PERSON + Su, D. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 0 + 5 + 658 + a2cb46c226b94831853a5d28c5d94b0a + + + PERSON + Xu, Y. is an author of multiple academic papers in the field of Natural Language Processing and Information Retrieval. Notably, Xu, Y. contributed to the paper titled "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," which addresses the management of scholarly information related to COVID-19 through advanced question answering and summarization techniques. Additionally, Xu, Y. co-authored the paper "Text summarization with latent queries," which explores innovative methods for text summarization by leveraging latent queries. These contributions highlight Xu, Y.'s expertise and active involvement in developing sophisticated systems for information retrieval and summarization. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 5 + 0 + 7 + 659 + d3511ecd27cd4166bdb39e757e275300 + + + PERSON + Yu, T. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 0 + 5 + 660 + de3b561f5cce4c83bccb39180e362c97 + + + PERSON + Siddique, F. B. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 0 + 5 + 661 + 5bfefaa0fce04002851733337bed714c + + + PERSON + Barezi, E. J. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 0 + 5 + 662 + b5fed5609f154df58c6a9f74e55fc0ba + + + PERSON + Fung, P. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 0 + 5 + 663 + 91ae5251eaab4c08afe6cd4cbefcaa6b + + + PERSON + Tang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 664 + bbdd53a15e99452a9deff05d1de2d965 + + + PERSON + Yang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 665 + 532bf54d5a924ff48aee254970efb914 + + + PERSON + Touvron, H. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9 + 666 + 2489232bd2bb492babe00617e7290282 + + + PERSON + Martin, L. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 7 + 667 + d2ed972353af4d1db74702638bfdbb58 + + + PERSON + Stone, K. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 668 + 575befc8d64c47eb95af8b1096e02963 + + + PERSON + Albert, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 669 + d6e6366617e04b0ba6732fd1d2d76429 + + + PERSON + Almahairi, A. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 670 + b4c4354c8edb40db984942799fe0c8b1 + + + PERSON + Babaei, Y. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 671 + 170507a64973429f818067b80506d428 + + + PERSON + Bashlykov, N. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 672 + fd9b298e6aea4685bbb2064b05fcda79 + + + PERSON + Batra, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 673 + eeecb159cc8a4c8989f8da0f3df09f2a + + + PERSON + Bhargava, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 674 + 70f22b1d7336492dbade94b8edefe457 + + + PERSON + Bhosale, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 675 + 66e098dc431146e19fc4bc2ea37efbd9 + + + PERSON + Traag, V. A. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 676 + 932e213c57134098a07073febd51dcc2 + + + PERSON + Waltman, L. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 677 + 9593428ad36746ae8af6d8ce639834ef + + + PERSON + Van Eck, N. J. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 678 + 1bcaeb58479d42a6963a073c09f3f397 + + + PUBLICATION + Scientific Reports is the journal where the paper "From Louvain to Leiden: guaranteeing well-connected communities" was published + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 679 + 1ef0c1c59ce946668ccf1a6a4f5ab7cc + + + PERSON + Trajanoska, M. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 680 + d734746e3d6146f780af91827e578dfd + + + PERSON + Stojanov, R. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 681 + 21ed913271614cbeb1b754cdbbef13af + + + PERSON + Trajanov, D. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 682 + 1505dfebbfb04652b0ba57de1a251d67 + + + PERSON + Trivedi, H. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 683 + 907ec65076e5494a8631efffb81b3178 + + + PERSON + Balasubramanian, N. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 684 + 2dc7f6b230db452190a09643ca3d5ec0 + + + PERSON + Khot, T. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 685 + c20ecfc93b3a4875ade5c92cfe4b94a1 + + + PERSON + Sabharwal, A. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 686 + 4bc7dc91ede345dfb63d7d4f7ac3554f + + + PERSON + Wang, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 6 + 0 + 8 + 687 + 0b2b815c9f834aaaac0c341097def9ba + + + PERSON + Liang, Y. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 6 + 0 + 8 + 688 + 424ae71c56024094a02e6fd9bfcfbb04 + + + PERSON + Meng, F. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 6 + 0 + 8 + 689 + 400d10f2ee1d49be9a66efa34dada0e6 + + + PERSON + Sun, Z. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 6 + 0 + 8 + 690 + 91deb9f152264e958d106d481ff2e1ee + + + PERSON + Shi, H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 6 + 0 + 8 + 691 + 586cf02da9494088aed9b3419725638f + + + PERSON + Li, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through their work on evaluating language models. Specifically, Li, Z. has co-authored the paper titled "Is ChatGPT a Good NLG Evaluator? A Preliminary Study," which explores the effectiveness of ChatGPT as a natural language generation evaluator. Additionally, Li, Z. has co-authored another paper, "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which examines the performance of large language models in evaluative roles using specific benchmarking tools. These contributions highlight Li, Z.'s active involvement in advancing the understanding and assessment of language models within the academic community. + 8d87efac8c50cf20cdf26bf61e5e2035,b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + 6 + 0 + 20 + 692 + 229d85a2783e4a2991f17d2ab5750af7 + + + PERSON + Xu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 6 + 0 + 9 + 693 + b7f97d1909a3433abef8ca8e9334fafa + + + PERSON + Qu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 6 + 0 + 9 + 694 + b7fdfffc38b94bf7872eabe9b022c8fd + + + PERSON + Zhou, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 6 + 0 + 9 + 695 + 6242e0c237a348908d0256ea790a0211 + + + PERSON + Wang, S. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" and also contributed to the paper "Is chatgpt a good nlg evaluator? a preliminary study." These works indicate Wang, S.'s involvement in cutting-edge research within the fields of federated search, retrieval augmented generation, and natural language generation evaluation, showcasing a focus on both the technical and evaluative aspects of Natural Language Processing and Information Retrieval. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 6 + 0 + 4 + 696 + 7cc9f26737e1442595e53253e98015ef + + + PERSON + Khramtsova is an author mentioned in the text + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 697 + 1868fec1493643208dbdcad7bc97dfa0 + + + PERSON + H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 0 + 4 + 698 + a87aa935dccf49cd98b40fb5afe7ad5c + + + PERSON + Khramtsova, E. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 0 + 4 + 699 + 36870a3393f6413e9bf647168eb6977a + + + PERSON + Zhuang, S. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through multiple academic papers. Notably, Zhuang, S. co-authored the paper titled "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," which explores the evaluation of federated search systems within the framework of retrieval-augmented generation. Additionally, Zhuang, S. co-authored another significant paper, "Judging llm-as-a-judge with mt-bench and chatbot arena," which delves into the assessment of large language models (LLMs) using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Zhuang, S.'s active involvement in advancing research in federated search and the evaluation of LLMs. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + 6 + 0 + 15 + 700 + 4fe3ff52700c491f8cc650aadb4d7cb0 + + + PERSON + Zuccon, G. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 0 + 4 + 701 + f1f6f6435a444e388d67e16e847afca6 + + + PERSON + Wang, Y. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4 + 0 + 6 + 702 + 0af2ca1c090843ea92679fd14c1fbc9a + + + PERSON + Lipka, N. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4 + 0 + 6 + 703 + 1b06d3e53ffd4771952fbef04d1e666c + + + PERSON + Rossi, R. A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4 + 0 + 6 + 704 + b8e966b34cba4b11b9995106767212ba + + + PERSON + Siu, A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4 + 0 + 6 + 705 + f6de923de6474d2cab6a9c2f0d81fa59 + + + PERSON + Zhang, R. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4 + 0 + 6 + 706 + 6915637e8d124fdc8473111d501e3703 + + + PERSON + Derr, T. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4 + 0 + 6 + 707 + 2233f31929194eac89333ce8731a5584 + + + PERSON + Yang, Z. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 708 + 61f1dc4267314470ac820b6a46c61f7b + + + PERSON + Qi, P. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 709 + f0c578614b224345974c3e4c110878af + + + PERSON + Zhang, S. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 710 + 7ffb88ebc729492c897ccfb569d7f6d0 + + + PERSON + Bengio, Y. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 711 + 60dce7d8bc1b4729a038178a400b9a59 + + + PERSON + Cohen, W. W. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 712 + 4cbb4e238c5b4656803fb9b4b6c3512e + + + PERSON + Salakhutdinov, R. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 713 + 652873bcd6d5432187e5deafc4fc5211 + + + CONFERENCE + The conference where the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" was presented + fc4b27d64f055b7fc30176ba110dd02e + 0 + 714 + 78f9b30c08134ac5abb4f4e0bff0f7f2 + + + PERSON + Yao, J.-g. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + 0 + 715 + f33e4e897b1e422bb516e8a2c941d9dc + + + PERSON + Wan, X. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + 0 + 716 + fac4e1553a9840e990bbfff46e64ff27 + + + PERSON + Xiao, J. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + 0 + 717 + 029a55d327ee4fb3a8314b36d52bdf34 + + + PUBLICATION + The journal where the paper "Recent advances in document summarization" was published + fc4b27d64f055b7fc30176ba110dd02e + 0 + 718 + 5a636c894c384532bff66212cf9d5824 + + + PERSON + Yao, L. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models"Yao, L. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 4 + 0 + 1 + 719 + a9c468ef78704e9aabfc0317a5b1b42d + + + PERSON + Peng, J. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 0 + 720 + 5df80c25d33a4d148a14aa614343cc6b + + + PERSON + Mao, C. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 0 + 721 + 6a87f06ed55a46f29b24f77e548a3f1d + + + PERSON + Luo, Y. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 0 + 722 + 0daf88ac4ec94cbb868e27e956c6d7f1 + + + PERSON + Zhang, J. is an author of the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 4 + 0 + 1 + 723 + 9ed120043e6247be9965e4904920991b + + + PERSON + Zhang, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 4 + 0 + 1 + 724 + 94d81d7de9254ae4b3b16fcc69aa22ea + + + PERSON + Gan, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 4 + 0 + 1 + 725 + 60c9212246f84ae5b6ab254127a39262 + + + PERSON + Wang, C. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 4 + 0 + 1 + 726 + 0f8d0c36a4274526a9eddedae5e63881 + + + PERSON + Zheng, L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zheng, L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Zheng, L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools like MT-Bench and Chatbot Arena. These contributions highlight Zheng, L.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR domains. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 6 + 0 + 12 + 727 + 6aedd377efbe4f07ae42e546996e7bfa + + + PERSON + Chiang, W.-L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Chiang, W.-L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Chiang, W.-L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Chiang, W.-L.'s active involvement in advancing the understanding and capabilities of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 6 + 0 + 12 + 728 + 1aa8484562784f378851c33843c89687 + + + PERSON + Sheng, Y. is an author known for contributing to the field of Natural Language Processing and Information Retrieval. Notably, Sheng, Y. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Sheng, Y. has contributed to the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Sheng, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic and technical community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 6 + 0 + 12 + 729 + f1a65d05dd5d456b889217020475ef80 + + + PERSON + Wu, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Wu, Z. co-authored the paper titled "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Wu, Z. is also credited with co-authoring the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Wu, Z.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 6 + 0 + 12 + 730 + c077d92b48b6477db91e1a0460600f52 + + + PERSON + Zhuang, Y. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zhuang, Y. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness of knowledge graphs. Additionally, Zhuang, Y. has also authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Zhuang, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the domain. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 6 + 0 + 12 + 731 + 5ca888df9b884e54accdd2ff29d125c1 + + + PERSON + Lin, Z. is an author of the paper "Exploring large language models for knowledge graph completion" and also contributed to the paper "Judging LLM-as-a-judge with MT-Bench and Chatbot Arena." These works indicate Lin, Z.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the application of large language models for tasks such as knowledge graph completion and the evaluation of language models in judgment tasks. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 6 + 0 + 12 + 732 + 8290a6212d6c4430ae0056c7e8eccd5f + + + PERSON + Li, D. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant research. Notably, Li, D. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Li, D. has also co-authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Li, D.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 6 + 0 + 12 + 733 + 14f8ac195fdb4e06a0b9ebc6ef391180 + + + PERSON + Xing, E. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Xing, E. contributed to the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Xing, E.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 6 + 0 + 12 + 734 + 667ee58a79194316ae2b82eadd3fc575 + + + TECHNOLOGY + Chatbot Arena is a platform or tool used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 6 + 0 + 11 + 735 + b0e3ee2324054c88adacdf80db13278f + + + 1.0 + Darren Edge and Ha Trinh co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 28b7457ca5dc4a38a488946a3f8e207e + 0 + 0 + + + 1.0 + Darren Edge and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8029a14d15404e6db95ddf5e2bf9fc15 + 1 + 0 + + + 1.0 + Darren Edge and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 389314ca89d445888c8d4985864dd733 + 2 + 0 + + + 1.0 + Darren Edge and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 87fe1462b9064d5692641ab48e826301 + 3 + 0 + + + 1.0 + Darren Edge and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + a55175ac57014df696ca09d0def9604b + 4 + 0 + + + 1.0 + Darren Edge and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 1766e8858d7b45ed97f71cb5a39e96ea + 5 + 0 + + + 1.0 + Darren Edge and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 6191e014f3f64e46a0777063ed4ac19a + 6 + 0 + + + 1.0 + Darren Edge is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + 21b0499cf14342269c46170c291d0535 + 7 + 0 + + + 1.0 + Ha Trinh and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + c1ef05b38b3f4d59888150fc0dd26826 + 8 + 0 + + + 1.0 + Ha Trinh and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 74cb9b3510e84498b9aee0d904316e8b + 9 + 0 + + + 1.0 + Ha Trinh and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 043d764b2e1b4d1294651ff938df5391 + 10 + 0 + + + 1.0 + Ha Trinh and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 31f2170fef004f3281c533a4a60dc3f3 + 11 + 0 + + + 1.0 + Ha Trinh and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 57f186c5c2754483ba66750e98222f95 + 12 + 0 + + + 1.0 + Ha Trinh and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 4b3fc569d91f4a7aa6501ad4fcf67b7a + 13 + 0 + + + 1.0 + Ha Trinh is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + cbc1667556f84a5eadf867a823e6986c + 14 + 0 + + + 1.0 + Newman Cheng and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + a876d1ab79864396bc47a039225fd5c7 + 15 + 0 + + + 1.0 + Newman Cheng and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + c09f67d4f25448c99f7c0552c30b7706 + 16 + 0 + + + 1.0 + Newman Cheng and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + c0866306dc8c4da2a8a81c0c3a78b657 + 17 + 0 + + + 1.0 + Newman Cheng and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 3884c37eb13a4c9097ee2c5be4eeefaf + 18 + 0 + + + 1.0 + Newman Cheng and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 90764eb2cab74cffb1c7d72d28b965cc + 19 + 0 + + + 1.0 + Newman Cheng is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + 01abe16e67c241a887aa62abe22d155c + 20 + 0 + + + 1.0 + Joshua Bradley and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 37049be0a2c240c6a06acf9339237b8b + 21 + 0 + + + 1.0 + Joshua Bradley and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + e785c52881704d95bf4ec03d2720f8ae + 22 + 0 + + + 1.0 + Joshua Bradley and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 654689c65613476b9905d7afb3809cd2 + 23 + 0 + + + 1.0 + Joshua Bradley and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 15dfb45a6ffa4d34ad72cfe4b3c5cc0d + 24 + 0 + + + 1.0 + Joshua Bradley is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + 427c3b7458f148d8bace1b768e2b5b7c + 25 + 0 + + + 1.0 + Alex Chao and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 95d506750fd94e72bbd9cf2d3fe18e28 + 26 + 0 + + + 1.0 + Alex Chao and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + bf0138ccbcc740089a55fd0c24897360 + 27 + 0 + + + 1.0 + Alex Chao and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 83cd5df42643494396b00d6cb6376def + 28 + 0 + + + 1.0 + Alex Chao is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + 909d28e443fd4e0bac189373125c8309 + 29 + 0 + + + 1.0 + Apurva Mody and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + efbc2439e5034801af83ac1a0b440535 + 30 + 0 + + + 1.0 + Apurva Mody and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + b9a2ef791a064f038cac2059ebea1138 + 31 + 0 + + + 1.0 + Apurva Mody is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + 1ce2b24bc93442148dc2240d3c6223b1 + 32 + 0 + + + 1.0 + Steven Truitt and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 804c1e94e7974332a817931363ddb643 + 33 + 0 + + + 1.0 + Steven Truitt is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + 32dc0b572ad84c75a64a2007788eb981 + 34 + 0 + + + 1.0 + Jonathan Larson is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + f5c11a5ac94e40068bca8be178a6bcd6 + 35 + 0 + + + 4.0 + RAG (Retrieval-Augmented Generation) and LLM (Large Language Models) are closely intertwined in the domain of Natural Language Processing and Information Retrieval. RAG is employed to enhance the capabilities of LLMs by enabling them to retrieve pertinent information from external knowledge sources. This symbiotic relationship allows LLMs to generate and assess text more effectively. Specifically, RAG leverages the power of LLMs to access and utilize relevant data, thereby augmenting the overall performance and accuracy of text generation tasks. + e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + ea28ff7f127e4677a913952595dce2f5 + 36 + 0 + + + 7.0 + Graph RAG is a specific implementation of RAG that combines the strengths of RAG with graph-based text indexing. This method leverages the natural modularity of graphs to partition data, facilitating global summarization. As a specialized approach within the RAG framework, Graph RAG enhances the capabilities of RAG by integrating graph structures to improve the efficiency and effectiveness of text data processing and summarization. + 21e52bc06a82796b1f4bcd73edda1f2a,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 192a6d23595045f38b0d46a3d8e52fd6 + 37 + 0 + + + 1.0 + Query-Focused Summarization is a task that RAG fails to address effectively + e8d83e6e7a7c0f57b218cef24976b745 + ef67c9fc60284b50aa15ac655b06a155 + 38 + 0 + + + 1.0 + RAG retrieves relevant information from an external knowledge source + e8d83e6e7a7c0f57b218cef24976b745 + cc8201cce1024b5192056fe8e98fda22 + 39 + 0 + + + 1.0 + Naive RAG is a specific implementation of RAG + e8c8f911135faf3ff35f24107eb3f99c + 97e097f9022540b88ab7c13d2805c25f + 40 + 0 + + + 1.0 + Ram et al., 2023 discusses RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + 829a6299a5fa4e7b8ff4020020a0be05 + 41 + 0 + + + 2.0 + Naïve RAG is a basic form of RAG + f35de4d9fb65f1d5a392064b20545c19 + dde2742459c24fb4a91172aa5c1a7620 + 42 + 0 + + + 2.0 + Modular RAG is an advanced form of RAG + f35de4d9fb65f1d5a392064b20545c19 + 323979a67d79498fa271acdf8cd1a0c2 + 43 + 0 + + + 2.0 + LLMs are used in various RAG tasks such as knowledge graph creation and completion + 92e93fc6449756c0a60200636b297f65 + c7e8b188b45841a0a1bcb22f3445ea6e + 44 + 0 + + + 2.0 + The paper by Trajanoska et al. discusses using LLMs for knowledge graph creation, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 5a4ad077106a4a3f951f43d2e01499b0 + 45 + 0 + + + 2.0 + The paper by Yao et al. discusses using LLMs for knowledge graph completion, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + a7ec8df038d7461689d28f1bdea84d9b + 46 + 0 + + + 2.0 + The paper by Ban et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 8ddefa32e2ed4eaf8f76d17a676f74f3 + 47 + 0 + + + 2.0 + The paper by Zhang et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 95ec30ce8dbe4ca28714e3e3735da8f3 + 48 + 0 + + + 2.0 + The paper by Gao et al. discusses advanced RAG where the index is a knowledge graph + 92e93fc6449756c0a60200636b297f65 + 259e7f5e2ec04418937513413b6d51d1 + 49 + 0 + + + 2.0 + KAPING is a method where the index is a knowledge graph, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 2289f06dd3804a3c84371dda0bab091e + 50 + 0 + + + 2.0 + G-Retriever is a method where subsets of the graph structure are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 34ff8ef897804691842071f9ff78708e + 51 + 0 + + + 2.0 + Graph-ToolFormer is a method where derived graph metrics are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + a7401447d994439993da7cc57f127649 + 52 + 0 + + + 2.0 + SURGE is a method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 754b0f2616064b18abb90f409ef0539a + 53 + 0 + + + 2.0 + FABULA is a method where retrieved event-plot subgraphs are serialized using narrative templates, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + acd35bb6b3cb4979a3f3fb68a86b3b05 + 54 + 0 + + + 2.0 + The paper by Wang et al. discusses a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 9e1e7f67ba044c7fbf64723af1ade58e + 55 + 0 + + + 2.0 + Sensemaking questions are used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + 57e16ff087a84b8ebd70de1e7e534225 + 56 + 0 + + + 2.0 + The evaluation of RAG systems focuses on corpora in the region of 1 million tokens + 92e93fc6449756c0a60200636b297f65 + bbf4007dc9c0486b8ea76d616045467a + 57 + 0 + + + 2.0 + Trade-offs are considerations involved in building a graph index for RAG systems + 92e93fc6449756c0a60200636b297f65 + 9535f4d754044e128cd3951a9d2e3702 + 58 + 0 + + + 2.0 + A graph index is a data structure used in RAG systems to organize and retrieve information + 92e93fc6449756c0a60200636b297f65 + e1ed13e29ee946d4aaafac50aaa3b68f + 59 + 0 + + + 2.0 + Performance of RAG systems varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + eb961d47a30c4870a1134b4a4672a8b2 + 60 + 0 + + + 2.0 + Different data types are used in RAG systems + 92e93fc6449756c0a60200636b297f65 + 5b019e8652264136b95306bac70a2e25 + 61 + 0 + + + 2.0 + Dataset sizes affect the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + 325fc9e2b37043b7af9f6ad338b09469 + 62 + 0 + + + 2.0 + Evaluation is the process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + 6bb11aa08b414232b5b45f10f5766f62 + 63 + 0 + + + 2.0 + Corpora are collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + 179737fd23c943babdfae01ac5c6bfc3 + 64 + 0 + + + 2.0 + Different question types are used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + 65a31e4da283411fb7c971f63d606723 + 65 + 0 + + + 2.0 + Target metrics are specific measures used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + de31810d43174a52aa2f31b72f4542f5 + 66 + 0 + + + 4.0 + Graph RAG utilizes Large Language Models (LLMs) to construct a graph-based text index, enabling the generation of summaries and the answering of queries. In this approach, LLMs play a crucial role in analyzing and generating text based on the information retrieved through the graph structure. Additionally, LLMs leverage the Graph RAG framework to provide comprehensive overviews of public figures in the entertainment industry. This integration of LLMs within Graph RAG enhances the system's ability to process and synthesize complex text data, making it a powerful tool for information retrieval and natural language processing tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + a6ae1d99330443fcacb06ace15a0d937 + 67 + 0 + + + 1.0 + Text chunks are processed using LLM to extract elements of a graph index + bc9e2c9e369c4108cf4f6dd5f60960f4 + 5174cdabb6024de0975762d3a80b059f + 68 + 0 + + + 1.0 + LLM is used to extract elements of a graph index from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + e379fba901174b529250169e62d98c09 + 69 + 0 + + + 2.0 + LLM (Large Language Model) and Few-Shot Examples are closely related in the context of Natural Language Processing and Information Retrieval. Few-shot examples are provided to the LLM for in-context learning, which helps tailor the extraction prompt. This technique is particularly useful for improving the performance of the LLM in specialized domains. By leveraging a small number of examples, the LLM can better understand and adapt to specific tasks, thereby enhancing its overall effectiveness in extracting and processing information within those specialized areas. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4 + 81ee8bb20bbb4d37bc0db642f1c75b8e + 70 + 0 + + + 1.0 + LLM extracts named entities from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 93b4aa6ce6e44123a861d4c3b3d509a2 + 71 + 0 + + + 1.0 + Kuratov et al. (2024) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + ee8414e314f547eeb369849cdb51bac2 + 72 + 0 + + + 1.0 + Liu et al. (2023) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + 9f77aa8888bd4f94abba8a77c4b0565c + 73 + 0 + + + 1.0 + LLM prompts are instructions given to the LLM for extracting elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + dcf33412678340319e7ec8f7be267ef9 + 74 + 0 + + + 1.0 + Recall degradation occurs with longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + 4aa1e0fa00c048939a5d006bfd305fb4 + 75 + 0 + + + 1.0 + The extraction process involves using LLM to identify and extract elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 03053ab4a9054384a5f5e88d28841621 + 76 + 0 + + + 1.0 + Default prompt is the standard set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + 9fd0f20997d541bca46c4ec9843a5d0f + 77 + 0 + + + 1.0 + Secondary extraction prompt is an additional set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + 27168beee1ff456696c330c9c3b3259f + 78 + 0 + + + 1.0 + The LLM uses covariate prompts to extract additional attributes associated with detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + e1c20e06aeac436788a9c6e918bcb844 + 79 + 0 + + + 1.0 + The LLM uses multiple rounds of gleanings to ensure no entities are missed + 2c6ed90897310eea2f28e33fff1c32b0 + 344417f626ef4da4be4539ef4037bf3f + 80 + 0 + + + 1.0 + Logit bias is used to force a yes/no decision from the LLM during entity extraction + 2c6ed90897310eea2f28e33fff1c32b0 + 8b1fff87c350475fb1d411a26c3c5b0c + 81 + 0 + + + 1.0 + The LLM extracts element instances from source texts + 2c6ed90897310eea2f28e33fff1c32b0 + 898a9458adfb4c13a1eafacf6a1068f6 + 82 + 0 + + + 1.0 + The LLM detects and summarizes communities of entities + 2c6ed90897310eea2f28e33fff1c32b0 + 5448f05781de44ea96e3dea40b285842 + 83 + 0 + + + 1.0 + LLM generates intermediate answers and scores for each chunk + 1d07b4248c2655081c7af0e373bd70c9 + 76b1e69904b84d09ba05c4b7efc48f32 + 84 + 0 + + + 1.0 + LLM generates a helpfulness score for each answer + 1d07b4248c2655081c7af0e373bd70c9 + 3f5590a604894d268603b4b27c3348b5 + 85 + 0 + + + 2.0 + LLM is used to generate questions for evaluating the Podcast Transcripts dataset + 922778ce1cb2fdd6dbab1746c8795620 + 68f998c9c8c34bb7a994de5a998bb9a0 + 86 + 0 + + + 2.0 + LLM is used to generate questions for evaluating the News Articles dataset + 922778ce1cb2fdd6dbab1746c8795620 + aafc13d02ade40adae13d3bee241817a + 87 + 0 + + + 1.0 + LLM uses Naive RAG to list public figures mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 81a4818e5cf84ea085abf09de385c86e + 88 + 0 + + + 1.0 + LLM-generated responses are evaluated using assessment metrics + e8c8f911135faf3ff35f24107eb3f99c + b69851bf63e34ced83827b0021628543 + 89 + 0 + + + 1.0 + LLM-generated responses are evaluated using specific questions + e8c8f911135faf3ff35f24107eb3f99c + b83a4e11bfa64559954327714b73293f + 90 + 0 + + + 1.0 + Ad-hoc LLM use involves the use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + de23b974cc90497eb4363e26d931a57c + 91 + 0 + + + 2.0 + LLMs are used for knowledge graph creation + f35de4d9fb65f1d5a392064b20545c19 + a9de65176e234a9f9073b8df9d675e90 + 92 + 0 + + + 2.0 + LLMs are used for knowledge graph completion + f35de4d9fb65f1d5a392064b20545c19 + 09a1bd11eb9347a9b466edad1a562cc5 + 93 + 0 + + + 2.0 + LLMs are used for the extraction of causal graphs + f35de4d9fb65f1d5a392064b20545c19 + 11d74eab1dcb4fcba7c45def5f0ee22d + 94 + 0 + + + 2.0 + LLMs are used for knowledge graph creation as per Trajanoska et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 4f6a6fd018a948f4bd0e630266b8bf61 + 95 + 0 + + + 2.0 + LLMs are used for knowledge graph completion as per Yao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 17dbfbecfaf0436bb11ed8f867c0caa1 + 96 + 0 + + + 2.0 + LLMs are used for the extraction of causal graphs as per Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + 2b1ec99684574c2ab26bb050d5b57a4d + 97 + 0 + + + 1.0 + Graph RAG is proposed as a method to combine the strengths of RAG and QFS + e8d83e6e7a7c0f57b218cef24976b745 + 1ccce5d1892a4b6995bbaec22882d34d + 98 + 0 + + + 7.0 + Graph RAG is an approach that utilizes community summaries in various capacities to enhance its functionality. Specifically, community summaries are generated within the Graph RAG framework and are employed to generate partial responses. These summaries are also used to compare against source texts, serving as a form of self-memory for the system. By leveraging community summaries, Graph RAG aims to improve the comprehensiveness and diversity of answers. Additionally, Graph RAG incorporates summaries of root-level communities within an entity-based graph index, further enriching its analytical capabilities. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 51cd93f89fbe4bcf883cdb2ca6774cd6 + 99 + 0 + + + 1.0 + Graph RAG is designed to handle global sensemaking questions over large datasets + e8d83e6e7a7c0f57b218cef24976b745 + 5f353b18fadb438f95ba0ea8feae137c + 100 + 0 + + + 2.0 + Graph RAG is implemented in Python. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + 947d70dd14b34cf398a1ab6dbdc51161 + 101 + 0 + + + 1.0 + The open-source implementation of Graph RAG will be available at this URL + e8d83e6e7a7c0f57b218cef24976b745 + 90f5597a558a4652bded9001a4ec2e56 + 102 + 0 + + + 1.0 + Graph RAG uses an entity knowledge graph to index text + e8d83e6e7a7c0f57b218cef24976b745 + 9532cf83e9324ea0a46e5ac89bac407d + 103 + 0 + + + 3.0 + Graph RAG is an approach that is evaluated for its comprehensiveness, a target quality used to assess its effectiveness. The method aims to improve the comprehensiveness of generated answers, ensuring that the information provided is thorough and complete. This evaluation metric is crucial in determining the success of Graph RAG in producing detailed and accurate responses. + 21e52bc06a82796b1f4bcd73edda1f2a,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + 8919fa72a9e74d1daff801e8f4c15b2b + 104 + 0 + + + 3.0 + Graph RAG is an approach in the domain of Natural Language Processing and Information Retrieval that focuses on improving the diversity of generated answers. Diversity, in this context, is a target quality used to evaluate the performance of the Graph RAG approach. By enhancing the diversity of responses, Graph RAG aims to provide a broader range of answers, thereby improving the overall effectiveness and robustness of the system. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745 + bef38889bb86413895d7dd25b4c3137c + 105 + 0 + + + 3.0 + Graph RAG uses a knowledge graph for global summarization + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + f770bc07cecf4aba8fe2d2c33fdc5542 + 106 + 0 + + + 1.0 + Community detection algorithms are used in the Graph RAG approach to partition graphs + 21e52bc06a82796b1f4bcd73edda1f2a + 13cd49512d5642989c2c72bb5e674807 + 107 + 0 + + + 1.0 + Podcast transcripts are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + e5c5c87a281b43868c344ff60f44c100 + 108 + 0 + + + 1.0 + News articles are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + a562ffbe986247b7943990e7151f4d69 + 109 + 0 + + + 2.0 + Graph RAG is evaluated using the target quality of Empowerment. Empowerment is specifically utilized to assess Graph RAG's capability in aiding users to achieve an informed understanding. This evaluation metric underscores the importance of user comprehension and the effectiveness of the Graph RAG approach in facilitating informed decision-making processes. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4 + 7ea0bc1467e84184842de2d5e5bdd78e + 110 + 0 + + + 2.0 + Graph RAG is compared to naive RAG in the evaluation. In this comparison, Graph RAG outperformed naive RAG on comprehensiveness and diversity. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + 056f23eb710f471393ae5dc417d83fd9 + 111 + 0 + + + 1.0 + Graph RAG is compared to global map-reduce summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + e1ae27016d63447a8dfa021370cba0fa + 112 + 0 + + + 1.0 + Query-focused summarization is a method used in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + f8c10f61a8f344cea7bdafa2d8af14b8 + 113 + 0 + + + 1.0 + Activity-centered sensemaking questions are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + aa7d003f25624e19bc88d3951d4dc943 + 114 + 0 + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 1c97184ce5ea4049be417a3fd125357b + 115 + 0 + + + 2.0 + The "Graph RAG" approach is evaluated in terms of its performance by considering "Token Costs." Token costs are measured to assess the efficiency of the Graph RAG method, indicating that the computational expense associated with processing tokens is a critical factor in determining the overall effectiveness of this approach. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 13a044c404394c34af1e9b07c48aa985 + 116 + 0 + + + 1.0 + Data flow describes the high-level process of the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 69ef1ac7b1f44372979149e82ecbc860 + 117 + 0 + + + 3.0 + Design parameters are key settings in the Graph RAG approach and significantly influence the Graph RAG approach and pipeline. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 6e26ce67bacc4fa089296843463f69ad + 118 + 0 + + + 1.0 + Graph RAG uses global summarization to summarize information from a large dataset + 21e52bc06a82796b1f4bcd73edda1f2a + ae0d3104647f4e6ab3ec2cf8e60be5ca + 119 + 0 + + + 1.0 + Graph RAG aims to answer specific queries + 21e52bc06a82796b1f4bcd73edda1f2a + 49e24b5f2c1d40d7857afe327db4f554 + 120 + 0 + + + 1.0 + Graph RAG uses a corpus for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + 587f39a32e93412395d9c22ad0ac2f94 + 121 + 0 + + + 1.0 + Activity-centered sensemaking is used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 8d9ded5fc9cf4c4faba8c6c8cd50e2f4 + 122 + 0 + + + 1.0 + Real-world datasets are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 595a841aa6034c93bd3dc55681e17710 + 123 + 0 + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + d0e58b78e8e84a0c8796e707b1f95f65 + 124 + 0 + + + 1.0 + Graph RAG is compared to source text summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + 215fcc6a3b5e452da123aa7f9ef0cbc9 + 125 + 0 + + + 1.0 + Low-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 0d0fc5d4ecb548079b28979186f19bf6 + 126 + 0 + + + 1.0 + Intermediate-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + e7d3fe0f87ff47f5a4c8d9572d27245a + 127 + 0 + + + 1.0 + High-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 6f7165b558ae427ca14b2b16d1e8e204 + 128 + 0 + + + 1.0 + The Graph RAG approach involves a specific pipeline for processing and summarizing text + bc9e2c9e369c4108cf4f6dd5f60960f4 + 2ec093d2a76d45f88ec508e45ba8c6a3 + 129 + 0 + + + 1.0 + Techniques are specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 16d5a528d6374612b87a5656e8d95193 + 130 + 0 + + + 1.0 + Implementation details are specific configurations used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 40293e74dbc643e8ab6546dff759ac7c + 131 + 0 + + + 2.0 + Graph RAG is a specific implementation of RAG systems + 922778ce1cb2fdd6dbab1746c8795620 + 1834b753dc7f4a8b98c2317a551b56ee + 132 + 0 + + + 2.0 + Graph RAG is a system that utilizes root-level community summaries, denoted as C0, to answer user queries. C0 represents these root-level community summaries within the Graph RAG analysis, serving as a foundational element in the system's ability to map out relationships and understand the structural dynamics within specialized communities. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + d9b127eab2f64e338d7adcd186786a45 + 133 + 0 + + + 1.0 + Graph RAG uses high-level community summaries (C1) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + a18f7c9f58ca49d6acf18e1ca69d3033 + 134 + 0 + + + 1.0 + Graph RAG uses intermediate-level community summaries (C2) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + f3c3dd44cf50495c81e362174991242e + 135 + 0 + + + 2.0 + Graph RAG utilizes low-level community summaries, represented by C3, to answer user queries. C3 plays a crucial role in the Graph RAG analysis by providing detailed summaries of community structures, which are essential for effectively addressing user inquiries. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + 86c2b3749a3c4342bbb3a8c70c3a76a0 + 136 + 0 + + + 2.0 + Graph RAG is a key entity in the analysis, serving both as a condition being compared and as a tool for comparing multiple conditions. This dual role highlights its significance in the study, where it functions not only as a subject of comparison but also as a methodological framework for evaluating other conditions. The analysis likely involves a detailed examination of various conditions, with Graph RAG playing a central role in facilitating these comparisons. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 571f65acb3134490932feeb91b01cca3 + 137 + 0 + + + 1.0 + Graph RAG uses different levels of graph communities to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + d3faf86c153f440eaa410305b3dc6617 + 138 + 0 + + + 1.0 + The Graph RAG mechanism uses an LLM evaluator for head-to-head comparison + 322e02986c8724eedbcf3ebfa20b989c + f85786004b0540349192d2ca05b15264 + 139 + 0 + + + 1.0 + Graph RAG is a multi-stage mechanism + 322e02986c8724eedbcf3ebfa20b989c + cf56bfc9fa7d47fe9cb553dd09f2b412 + 140 + 0 + + + 1.0 + Graph RAG mentions Taylor Swift as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + a077dbcd38b644f6929cf05272c2fb9d + 141 + 0 + + + 1.0 + Graph RAG mentions Travis Kelce as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + d2659a32b9de406eb750a35d078c9774 + 142 + 0 + + + 1.0 + Graph RAG mentions Britney Spears as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + 0b26876307ad4cc48839b61a21a1d03a + 143 + 0 + + + 1.0 + Graph RAG mentions Justin Timberlake as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + c68e6c694a554256846d12178ddb12dc + 144 + 0 + + + 1.0 + Graph RAG is determined to be the winner based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + ff25ce2e8ace4bdcb765c863b483852b + 145 + 0 + + + 1.0 + Answer 1 was generated using the Graph RAG method, which provides a comprehensive and structured overview of public figures across various sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + 757e402cb7ee4601ac1bc8c4fafb5207 + 146 + 0 + + + 1.0 + Graph RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + 62e8f5f04cd04384b246291cef3a9e4d + 147 + 0 + + + 1.0 + Graph RAG is compared with source texts for answer comprehensiveness and diversity + 6f33a085ff3304e5994f7fbb86c881a4 + c04abbd5e59b4c64b023908f6db05498 + 148 + 0 + + + 1.0 + TS represents source text summarization in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 6bb9bed2e39c4e31a81f12479af3d16c + 149 + 0 + + + 1.0 + Root-level summaries are used in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 26c926c6016d4639b05427f01ba629f5 + 150 + 0 + + + 1.0 + Answer comprehensiveness is used to evaluate the performance of Graph RAG + 6f33a085ff3304e5994f7fbb86c881a4 + 8f6872eeb81b432b91405d327636113c + 151 + 0 + + + 1.0 + Win rate is used to measure the success rate of Graph RAG in providing comprehensive and diverse answers + 6f33a085ff3304e5994f7fbb86c881a4 + ac80a99fda2b488285d29596dd4d1471 + 152 + 0 + + + 1.0 + Element extraction prompts are used in Graph RAG to retain specific details in the index + 6f33a085ff3304e5994f7fbb86c881a4 + 67d6a3481e4b419292247cef5cd5b737 + 153 + 0 + + + 2.0 + Graph RAG incorporates the concept of self-memory + f35de4d9fb65f1d5a392064b20545c19 + 904cd052ec194654bb72f4027e43daa3 + 154 + 0 + + + 2.0 + Graph RAG incorporates the concept of iterative retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + 7e88fd2e835147fbb71866612735e8d4 + 155 + 0 + + + 2.0 + Graph RAG incorporates the concept of federated retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + 029d1a8c3b184aa5bb21228f40cd12fd + 156 + 0 + + + 2.0 + Graph RAG incorporates concepts used in multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + a1ebc53a0bc74a0eb6dbdd18cf3c88cd + 157 + 0 + + + 2.0 + Graph RAG incorporates concepts used in multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + a51d063ad4c744049edb359eb88407b7 + 158 + 0 + + + 2.0 + Graph RAG uses a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + b67268f90338474e8e53b9a6715b6833 + 159 + 0 + + + 2.0 + Graph RAG incorporates the concept of a tree of clarifications + f35de4d9fb65f1d5a392064b20545c19 + acb53370e72b4430a752d9ea18c17352 + 160 + 0 + + + 3.0 + Graph RAG utilizes a self-generated graph index. This self-generated graph index is a crucial component of Graph RAG, enabling it to efficiently manage and retrieve information within its graph-based framework. The use of a self-generated graph index suggests that Graph RAG has an inherent capability to construct and maintain its indexing structure, which likely enhances its performance and adaptability in handling complex data relationships. + e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + a8738c7de11543df930169741381c252 + 161 + 0 + + + 2.0 + Graph RAG incorporates concepts from Gao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 3f8b5b2727924ba0b62e6286063b6861 + 162 + 0 + + + 2.0 + Graph RAG incorporates concepts from Cheng et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + bb5010633113442eaf814852995cfa22 + 163 + 0 + + + 2.0 + Graph RAG incorporates concepts from Mao et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + 9eb8c635538243a690366f8bc1de34e0 + 164 + 0 + + + 2.0 + Graph RAG incorporates concepts from Shao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 050c5b770d51409cb40f9c52f02d1329 + 165 + 0 + + + 2.0 + Graph RAG incorporates concepts from Wang et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + 9e12f514d26d48dfab65807568a6cff9 + 166 + 0 + + + 2.0 + Graph RAG incorporates concepts from Su et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + 97f98b1623104f48aa93196a1f7dede2 + 167 + 0 + + + 2.0 + Graph RAG incorporates concepts from Feng et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 87718ef799a34104b6ef9c2df6621cbc + 168 + 0 + + + 2.0 + Graph RAG incorporates concepts from Trivedi et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + f64e87431d674f298c533f6878458b95 + 169 + 0 + + + 2.0 + Graph RAG incorporates concepts from Khattab et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + e6d44d0db58f42799a02eacbd6b14543 + 170 + 0 + + + 2.0 + Graph RAG incorporates concepts from Sarthi et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + 64961fbc3a1641378be10bcb3b0955e1 + 171 + 0 + + + 2.0 + Graph RAG incorporates concepts from Kim et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 59bcc2ec512c4c1ba44272446b419230 + 172 + 0 + + + 2.0 + Graph RAG generates community answers in parallel + f35de4d9fb65f1d5a392064b20545c19 + 8f39ae56f8b54b1b94faf04dbd0b9d11 + 173 + 0 + + + 1.0 + Graph RAG is compared to a graph-free approach for global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + f3018b934ac241639a33c925c24bc507 + 174 + 0 + + + 1.0 + Graph RAG is compared to map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + ddedfd5179e64700adced4803c75cdba + 175 + 0 + + + 1.0 + Graph RAG uses rich text annotations + e4d9b12cf2b4c691c74019eefff4fb39 + 07d501edd4614e1d9d08d01b702688a3 + 176 + 0 + + + 1.0 + Graph RAG uses a hierarchical community structure + e4d9b12cf2b4c691c74019eefff4fb39 + f745075dedcf444daa9370cf32403d31 + 177 + 0 + + + 1.0 + Graph RAG can operate using embedding-based matching + e4d9b12cf2b4c691c74019eefff4fb39 + 1ef48284d238405f94190125092a3e28 + 178 + 0 + + + 1.0 + Graph RAG can be part of hybrid RAG schemes + e4d9b12cf2b4c691c74019eefff4fb39 + 8806b817446447e3b50f5bc85ff497e1 + 179 + 0 + + + 1.0 + Graph RAG can employ map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + 7303ee20690449db8c168df3fe008bc5 + 180 + 0 + + + 1.0 + Graph RAG can extend operations across the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + 2f1c535a14b14758bf1cacca81c74878 + 181 + 0 + + + 1.0 + Alonso contributed to the work on Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + 3b78cc7ce8224afcab3e4bbe550cde10 + 182 + 0 + + + 1.0 + Graph RAG includes local graph RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + 29ec9dd9f5864170a7e75c46c11c0090 + 183 + 0 + + + 1.0 + Graph RAG uses an entity-based graph index + e4d9b12cf2b4c691c74019eefff4fb39 + 7893ee15f0e941cbacad8cc1feaacbaf + 184 + 0 + + + 2.0 + NebulaGraph launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs + 833e7d67dcd30790b26b71c9b5306f6b + f53397f743ca4d7397c0a694fe787da0 + 185 + 0 + + + 1.0 + Community summaries are used to generate partial responses + e8d83e6e7a7c0f57b218cef24976b745 + 0041db9da3694ad397f37c76f8477770 + 186 + 0 + + + 1.0 + Community summaries are created from graph communities + f0306814bf64f5c9e79603fc6a52f4ea + a7c2a64e06374091adce74adb36801ab + 187 + 0 + + + 2.0 + Community answers are created from community summaries. These answers are generated by synthesizing information from the summaries provided by the community, ensuring that the responses are comprehensive and reflective of the collective knowledge and insights within the community. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 107568a67cac472c89dfce4bbe11157c + 188 + 0 + + + 1.0 + Domain-tailored summarization is used to create community summaries + f0306814bf64f5c9e79603fc6a52f4ea + 3d78aa9d14714ac189e4020f78b15d24 + 189 + 0 + + + 1.0 + Community descriptions are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + ce0366abadef410d9b65e2bfbbf0b0f9 + 190 + 0 + + + 1.0 + Partial answers are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + f4370806deb84d0eb7e85e742e7d4bbf + 191 + 0 + + + 1.0 + Community summaries are created for each level in the hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56 + c92392d168c2443e8ed7b04992d0c92b + 192 + 0 + + + 1.0 + Community summaries are useful for understanding the global structure and semantics of the dataset + 7fb7d9ce2da9c940a32afdd87d1d9e56 + b5800c807edd4087a2420007272d15d0 + 193 + 0 + + + 1.0 + Community summaries are used to answer global queries + 843fc5421e086120ffa1c75856ecf6cd + aa247540e90d4a7abc5bca6fafaaffa1 + 194 + 0 + + + 1.0 + Community summaries are generated from root communities + 843fc5421e086120ffa1c75856ecf6cd + 34537afa1e954e08bdb52ead3a49e2f3 + 195 + 0 + + + 1.0 + Community summaries are generated from sub-communities + 843fc5421e086120ffa1c75856ecf6cd + ae043af0299f4b32a98cf187efd2a5db + 196 + 0 + + + 1.0 + Community summaries are added to the LLM context window until the token limit is reached + 843fc5421e086120ffa1c75856ecf6cd + 6016863be3414d5a92397f2d45fdfd78 + 197 + 0 + + + 1.0 + Global answers are generated from community summaries + 843fc5421e086120ffa1c75856ecf6cd + a9b900821b8444d69f432da08a77539f + 198 + 0 + + + 1.0 + The level of summary detail affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + 1fee51d6f4614127a3e1cc80d018506e + 199 + 0 + + + 1.0 + The scope of information affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + 00dc2c0748214e52bc799ca3e25204e9 + 200 + 0 + + + 1.0 + Community summaries are used for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + 42d1a9e749ad40daa34c7b0b695f8751 + 201 + 0 + + + 2.0 + Community summaries are divided into chunks of pre-specified token size + 843fc5421e086120ffa1c75856ecf6cd + 20de9a1af6ab4e88acf003cb7be0217c + 202 + 0 + + + 1.0 + Summary detail and scope affect the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + f3229f10a5a54cb1b91a26ffa6ee77a3 + 203 + 0 + + + 1.0 + Community summaries are divided into chunks + 1d07b4248c2655081c7af0e373bd70c9 + 5154b4a4f3ac43729703c69fccb54633 + 204 + 0 + + + 1.0 + Community summaries are prepared to answer user queries + 1d07b4248c2655081c7af0e373bd70c9 + 2091070e709e45f5ae56d40a9da45520 + 205 + 0 + + + 1.0 + Intermediate answers are generated from community summaries + 1d07b4248c2655081c7af0e373bd70c9 + 09045ef5c4314dde9a631a206274563f + 206 + 0 + + + 1.0 + Community summaries are part of the graph community hierarchy + 36db32c37e1987e2c5863898ad882190 + 1b9baa98ede84164883e8cdcbc7000c1 + 207 + 0 + + + 1.0 + Community summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + e4f3fcc475a74756925b730caffcb70d + 208 + 0 + + + 1.0 + Community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 7f3d5282303f4fc3a009e04f7de0ad84 + 209 + 0 + + + 1.0 + Summaries of root-level communities are used in Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + 1219a14eaf5f49ab84c9287ebf58db7a + 210 + 0 + + + 1.0 + Global sensemaking questions are evaluated over datasets in the 1 million token range + e8d83e6e7a7c0f57b218cef24976b745 + efaa386bd5e9454b87e1851cd8b28ac3 + 211 + 0 + + + 1.0 + Global sensemaking questions are directed at an entire text corpus + e8d83e6e7a7c0f57b218cef24976b745 + 073241be9b6a4952ad01dd14b94fb89c + 212 + 0 + + + 1.0 + The Python-based implementation of Graph RAG approaches will be available at this URL + e4d9b12cf2b4c691c74019eefff4fb39 + f7ac6bc4a9ca4250ad29a3adb5d08657 + 213 + 0 + + + 1.0 + Query-focused summarization is used to produce the global answer + f0306814bf64f5c9e79603fc6a52f4ea + ac2ee54e75a2492c8db372dadfccd083 + 214 + 0 + + + 1.0 + Map-reduce is used for query-focused summarization of an entire corpus + 21e52bc06a82796b1f4bcd73edda1f2a + ee895ad0b8cd40c29465e8527748d847 + 215 + 0 + + + 1.0 + Query-focused summarization is used for answering global queries + 7fb7d9ce2da9c940a32afdd87d1d9e56 + fe38c996c2d64bc899eabd6389034075 + 216 + 0 + + + 1.0 + An entity knowledge graph is derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + c0e28ae832c94405b8ddd4d2ad978be5 + 217 + 0 + + + 2.0 + In the domain of Natural Language Processing and Information Retrieval, "SOURCE DOCUMENTS" and "TEXT CHUNKS" are closely related entities. Text chunks are extracted from source documents, specifically for processing in the Graph RAG (Retrieval-Augmented Generation) approach. This method involves breaking down the source documents into smaller, manageable pieces, or text chunks, which can then be effectively utilized in various computational processes, enhancing the efficiency and accuracy of information retrieval and generation tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 7a4573a19ef94e25b4480cb4d953ae7a + 218 + 0 + + + 1.0 + Intermediate-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + 05f6639803524537b67a7f2b0c66ad23 + 219 + 0 + + + 1.0 + Low-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + 21bfd14cbc1f4cbc8ac59f7fd8c75b31 + 220 + 0 + + + 1.0 + Document corpus consists of source documents being processed + bc9e2c9e369c4108cf4f6dd5f60960f4 + c19cf2d7b067421990ab9f3acec9e736 + 221 + 0 + + + 1.0 + Partial responses are summarized to generate a final response + e8d83e6e7a7c0f57b218cef24976b745 + 3e1981b9301c4d339a9228ae7a089a04 + 222 + 0 + + + 1.0 + The LLM evaluator assesses answers based on the comprehensiveness metric + 322e02986c8724eedbcf3ebfa20b989c + 0948efa844814529b4c023aacbc23d64 + 223 + 0 + + + 1.0 + Naive RAG is evaluated for comprehensiveness + e8c8f911135faf3ff35f24107eb3f99c + fcdc0cc5ff93453eb0b94b9254760999 + 224 + 0 + + + 1.0 + Comprehensiveness is a metric used to determine the decision + e8c8f911135faf3ff35f24107eb3f99c + 0ec4ad4398a8457ab3d71bd2561858dc + 225 + 0 + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + 3c06988555334a389eab093f98679e85 + 226 + 0 + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + 81ceb8db419b4697ad24e9d7f46422ff + 227 + 0 + + + 1.0 + The smallest context window size (8k) was universally better for comprehensiveness + 4c855404ee3d3c94aa2136f1513c666f + fd05d8198d0947b39b8fa1b16f3ecf5f + 228 + 0 + + + 1.0 + The final evaluation prioritized comprehensiveness in answers + 4c855404ee3d3c94aa2136f1513c666f + d984f08ad62f47ab9aabb9aeec1b245e + 229 + 0 + + + 1.0 + Global approaches achieved higher comprehensiveness win rates + 36db32c37e1987e2c5863898ad882190 + 43603c7868164ac38c659bce7a77f45a + 230 + 0 + + + 1.0 + The LLM evaluator assesses answers based on the diversity metric + 322e02986c8724eedbcf3ebfa20b989c + 54a20cc6062d4b7193d023b6ff20461f + 231 + 0 + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + 6bb190069a704ccca3d8e1648a384185 + 232 + 0 + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + 47d2036509bf408095ab440bd052ac24 + 233 + 0 + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on diversity + 4c855404ee3d3c94aa2136f1513c666f + c20e6b1418a140389c31c7b71a6eba0c + 234 + 0 + + + 1.0 + The final evaluation prioritized diversity in answers + 4c855404ee3d3c94aa2136f1513c666f + ad96e5294247465a9c7d5ea8161dc305 + 235 + 0 + + + 1.0 + Global approaches achieved higher diversity win rates + 36db32c37e1987e2c5863898ad882190 + 25c968bf5a4f48369fded6c260f71540 + 236 + 0 + + + 1.0 + Human endeavors rely on sensemaking to understand and reason about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + 269b441634a144219f539202309bc9fb + 237 + 0 + + + 1.0 + Human endeavors rely on analyzing document collections for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + d0baf2392635468db7f5657f89eb2024 + 238 + 0 + + + 1.0 + LLMs are used to automate sensemaking in complex domains + f0306814bf64f5c9e79603fc6a52f4ea + 4f29bcf5377d4c9f94ff3f8ca2f8d941 + 239 + 0 + + + 1.0 + Microsoft uses LLMs for automating sensemaking in scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + e7072a582d9b4c1ea8b171ee940d4d6e + 240 + 0 + + + 1.0 + Ranade uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + cb6fcf84e3d04ef59b01f97ac94823a1 + 241 + 0 + + + 1.0 + Joshi uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + 97a21db5f5954e2c8868b298a3f0090e + 242 + 0 + + + 1.0 + LLM prompts are used to tailor the responses of large language models + f0306814bf64f5c9e79603fc6a52f4ea + c8f3e6cadcf34c8fafe8987e4a9b66f8 + 243 + 0 + + + 1.0 + Ranade and Joshi discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + 26c9c44e5059429bb8abc3308bc6c814 + 244 + 0 + + + 2.0 + GPT is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + 7cea9903153f43b895c0b23d25bc90a3 + 245 + 0 + + + 2.0 + Llama is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + b54436ccc23745c88d24edcc3fdd8ed1 + 246 + 0 + + + 2.0 + Gemini is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + 977c895bb98d4136a76e8749533154b6 + 247 + 0 + + + 2.0 + Kuratov et al., 2024, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + 8d75cfea884248aba1f372de5e1b82a9 + 248 + 0 + + + 2.0 + Liu et al., 2023, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + 90f4ee186bcd4996ad8002888569fffc + 249 + 0 + + + 1.0 + Sensemaking is applied in the domain of scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + 4bb78401581b4240b0967309e96af00b + 250 + 0 + + + 1.0 + Sensemaking is applied in the domain of intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + a18dd9ea4143411cb32e261db056cf0c + 251 + 0 + + + 1.0 + Klein defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + cd8d9795f540413390927ea2a9e77c26 + 252 + 0 + + + 1.0 + Klein et al. defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + 2917f3b478b04ffcacd4b47602f4d0f5 + 253 + 0 + + + 2.0 + Element instances are extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 3984bd063b384901862e68506c77cc68 + 254 + 0 + + + 1.0 + Entity references are extracted from text chunks during processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + 4137a2c7dd884bc2a8469b7fa937346c + 255 + 0 + + + 1.0 + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + 60b6bf585ccc477d830d4b69b8c7b62a + 256 + 0 + + + 2.0 + Element instances are converted into element summaries by the LLM (Large Language Model). Element summaries are created from element instances, indicating a transformation process facilitated by the LLM. This process involves the LLM taking detailed element instances and generating concise element summaries, which encapsulate the essential information from the instances. + 2c6ed90897310eea2f28e33fff1c32b0,f0306814bf64f5c9e79603fc6a52f4ea + 4330f73cb78a4bb39a384eb29112201b + 257 + 0 + + + 1.0 + Covariates are additional attributes associated with extracted element instances + bc9e2c9e369c4108cf4f6dd5f60960f4 + 45c4ed77967746e485ec9e52c0dcc0d2 + 258 + 0 + + + 1.0 + Domain-tailored summarization is used to create element summaries + f0306814bf64f5c9e79603fc6a52f4ea + 17c2cc25d00347c3bf2422d4f7a4ad7e + 259 + 0 + + + 1.0 + Element summaries include descriptions of entity nodes + 2c6ed90897310eea2f28e33fff1c32b0 + 0057fb2ddc0e4088ae5099b7ffa137da + 260 + 0 + + + 1.0 + Element summaries include descriptions of relationship edges + 2c6ed90897310eea2f28e33fff1c32b0 + d67d67cc3698438db76eb4a7f75e1ea0 + 261 + 0 + + + 1.0 + Element summaries include descriptions of claim covariates + 2c6ed90897310eea2f28e33fff1c32b0 + c23761290af24cf29adc1ee8644bdad0 + 262 + 0 + + + 1.0 + Element summaries are used to understand the structure and semantics of graph communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + de51b828ce1f442bbb19a7b20bce9dda + 263 + 0 + + + 1.0 + Element summaries include descriptions of nodes + 843fc5421e086120ffa1c75856ecf6cd + 4a3ff6a3471945fd8c7fd5c171c56d56 + 264 + 0 + + + 1.0 + Element summaries include descriptions of edges + 843fc5421e086120ffa1c75856ecf6cd + 31bb84eb2a834dabacc0ed51af4fcefd + 265 + 0 + + + 1.0 + Element summaries include descriptions of covariates + 843fc5421e086120ffa1c75856ecf6cd + 5070012e83e7442381bcba1cdacdb7d8 + 266 + 0 + + + 1.0 + Sub-community summaries are used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + 5eda9074df124f5497f17b61badd52ac + 267 + 0 + + + 2.0 + Community detection is a technique used to identify graph communities. Graph communities are groups of nodes within a graph that are more densely connected to each other than to the rest of the graph. This process of identifying such communities is crucial for understanding the structural dynamics and relationships within complex networks, particularly in the domain of Natural Language Processing and Information Retrieval. By leveraging community detection algorithms, researchers can uncover hidden patterns and insights within large datasets, facilitating more effective data analysis and interpretation. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 4cf4107b0e2842778aaa658a1a85f3b3 + 268 + 0 + + + 1.0 + Global answer is created from community answers + f0306814bf64f5c9e79603fc6a52f4ea + 7f4857f94b4e4e49be7236a42071e167 + 269 + 0 + + + 2.0 + Global answers are generated in response to user queries + 843fc5421e086120ffa1c75856ecf6cd + d21a1fef903f4a399bd3cd366aad3c9e + 270 + 0 + + + 1.0 + Global answer is generated by sorting intermediate answers based on helpfulness scores + 1d07b4248c2655081c7af0e373bd70c9 + fc596a598ff74a4c843e405b597551b5 + 271 + 0 + + + 1.0 + Intermediate answers are combined to form the global answer + 1d07b4248c2655081c7af0e373bd70c9 + e2aacff6b4404574b818e7a3ece57b5b + 272 + 0 + + + 1.0 + The final context window is used to generate the global answer + 1d07b4248c2655081c7af0e373bd70c9 + 2ec5cae98c7a485881f0680fbca6d67f + 273 + 0 + + + 1.0 + Graph RAG pipeline operates at indexing time + f0306814bf64f5c9e79603fc6a52f4ea + c87b815d61af448596d3194a804b57b3 + 274 + 0 + + + 1.0 + Graph RAG pipeline operates at query time + f0306814bf64f5c9e79603fc6a52f4ea + 2f92fc82c3b74417896bad3bd8e61f5e + 275 + 0 + + + 1.0 + Nodes are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + fb61c68efe5b4d69a9623e531e7c639c + 276 + 0 + + + 1.0 + Edges are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + dc61e34c1ca8419e923aeeff7d83d949 + 277 + 0 + + + 1.0 + Covariates are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + 697fb824eef34759852f1d5588921aec + 278 + 0 + + + 1.0 + Leiden method is used in the graph RAG pipeline for community detection + f0306814bf64f5c9e79603fc6a52f4ea + b872fcc5b18a4f32b976f4693f22e88e + 279 + 0 + + + 1.0 + Graph RAG pipeline uses the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + 64be9b98299f4d349e0f4358685ca235 + 280 + 0 + + + 1.0 + The Podcast dataset graph consists of 8564 nodes + 36db32c37e1987e2c5863898ad882190 + 8302a03f6ede471bb955c0bbf44a4b3c + 281 + 0 + + + 1.0 + The News dataset graph consists of 15754 nodes + 36db32c37e1987e2c5863898ad882190 + a02263dd89964a1c8ab2d0e9aba0f4eb + 282 + 0 + + + 1.0 + The Podcast dataset graph consists of 20691 edges + 36db32c37e1987e2c5863898ad882190 + 6b7aa6ce4cac4edbaaab831286e67e5e + 283 + 0 + + + 1.0 + The News dataset graph consists of 19520 edges + 36db32c37e1987e2c5863898ad882190 + 655d40ea08e348ad94ae49785797da90 + 284 + 0 + + + 1.0 + Traag contributed to the development of the Leiden method + f0306814bf64f5c9e79603fc6a52f4ea + 254cea99330f4f2aa062c771146da7ea + 285 + 0 + + + 2.0 + Traag et al. are the authors of the Leiden algorithm and developed the Leiden method. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + a2836232227c4e3383d166db860cb2a3 + 286 + 0 + + + 2.0 + Leiden is a specific type of community detection algorithm used in various analytical pipelines. It is designed to identify and map out the structural dynamics within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. The algorithm is known for its efficiency and accuracy in detecting community structures, making it a valuable tool for researchers and practitioners in the field. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + 8a9247ee9bac45bdbf69c9d0bb8419b5 + 287 + 0 + + + 1.0 + Leiden is known for its ability to recover hierarchical community structures efficiently + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 757a0f78fcdd4bf6b8326a75fcee9e15 + 288 + 0 + + + 1.0 + The Leiden algorithm is used to detect graph communities in the MultiHop-RAG + 7fb7d9ce2da9c940a32afdd87d1d9e56 + b5235cb24b8f440389f250ebd5b6e2f8 + 289 + 0 + + + 1.0 + Figure 3 shows graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + bdee1849252749efa2e671ed87641f61 + 290 + 0 + + + 1.0 + Lewis contributed to the development of the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + 057641c1476247958d8c357e17095d8e + 291 + 0 + + + 1.0 + Lewis et al. developed the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + b61dfd0b24664f37af4046bdf0cb7b19 + 292 + 0 + + + 2.0 + Lewis et al., 2020, are the authors who established the RAG approach + fb3c48579608fa28be585ceb6cd2f0fe + 0bc00f14e6194df7b0fe9ef9ba28d34f + 293 + 0 + + + 1.0 + Kevin Scott is the CTO of Microsoft + 1d07b4248c2655081c7af0e373bd70c9 + b823c5d22037423da919eee6c35c4c8b + 294 + 0 + + + 2.0 + Microsoft conducted a study on the impact of large language models on scientific discovery using GPT-4 + 833e7d67dcd30790b26b71c9b5306f6b + cd7f555e4ab948ba94bade14e262ff84 + 295 + 0 + + + 1.0 + Preprint is available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 86cd53087b2542f898d6cecca31e6145 + 296 + 0 + + + 1.0 + Baumel, T. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 5dc3480806b04fdd8089a3be46e22540 + 297 + 0 + + + 1.0 + Eyal, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 50c91820a91f488d8606198540aba894 + 298 + 0 + + + 1.0 + Elhadad, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + a38eace89e7e40de8f007fde24597e9e + 299 + 0 + + + 1.0 + Es, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 5d75097d065e4b049a1678deab40949b + 300 + 0 + + + 1.0 + James, J. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + c277134d380a42cd886a14a953554792 + 301 + 0 + + + 1.0 + Espinosa-Anke, L. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + b680be879404440885b1d3af5b9af583 + 302 + 0 + + + 1.0 + Schockaert, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 4cc609b1a64a442aac6b72078a315ac6 + 303 + 0 + + + 1.0 + Feng, Z. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + a3ee323c9c9a4f81b5907030122b80d2 + 304 + 0 + + + 1.0 + Feng, X. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 19aa5f0b738c4f4a96668c80c3e93331 + 305 + 0 + + + 1.0 + Zhao, D. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + f8402b10349f4db888ac4fb6fd81723a + 306 + 0 + + + 1.0 + Yang, M. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 5927f9089289429da4adf2bbd5641e44 + 307 + 0 + + + 1.0 + Qin, B. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 60724b8b268044b69a4b3d939f1757e2 + 308 + 0 + + + 1.0 + LangChain is an organization that has published on arXiv + 71f6daf11e64e5273a3847d46bf228e1 + d931685d35e149909472f736114ca62f + 309 + 0 + + + 1.0 + Wang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 3f5e9927a4114a958d75f5ed313526a8 + 310 + 0 + + + 1.0 + Khramtsova, E. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 4728bf0cb7564bbd85c90ceaa846f290 + 311 + 0 + + + 1.0 + Zhuang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + cdac6338c3234797a0d3a32cd68d1b2e + 312 + 0 + + + 1.0 + Zuccon, G. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 372f78df13f9452b84d898c703a1ba95 + 313 + 0 + + + 1.0 + Wang, Y. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 7af06d2b32a941a4b044579a7c423371 + 314 + 0 + + + 1.0 + Lipka, N. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + a10b8fad74744ae981747dadf7234b78 + 315 + 0 + + + 1.0 + Rossi, R. A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0cb2118ecc87439a91409deef7ef9830 + 316 + 0 + + + 1.0 + Siu, A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + ea27218042d640fd81c23eb64aff6b46 + 317 + 0 + + + 1.0 + Zhang, R. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 9e5d626681094933abf87cf797f2fa46 + 318 + 0 + + + 1.0 + Derr, T. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 545358ff14f84601a22e9f39f5ef1534 + 319 + 0 + + + 1.0 + Xu, Y. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 1b0e7dbc7c5944a7833f6540bde1fa4f + 320 + 0 + + + 1.0 + Lapata, M. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0c0f2d8c623949f1ae89c67d0753aeab + 321 + 0 + + + 1.0 + Zhang, J. published the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 20c3844c80a140ac97b62dc444feee41 + 322 + 0 + + + 1.0 + Zhang, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + c5fac1bea509464d9dc934275d938039 + 323 + 0 + + + 1.0 + Gan, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 45b64fbddd8f4abdb86a9c3c6f53f802 + 324 + 0 + + + 1.0 + Yao, L. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0e504b58cbda4d9188050bc43004c01f + 325 + 0 + + + 1.0 + Wang, C. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + c06bd37e120e4af49ec8bd6ce399473b + 326 + 0 + + + 1.0 + Zheng, L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 5d507985f2f540d8a1fa2d1191eae2a8 + 327 + 0 + + + 1.0 + Chiang, W.-L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 8e0b5b4011d74bbb8dc09fa05d88369c + 328 + 0 + + + 1.0 + Sheng, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 5d8184f5d52040d8bb67d1a6b889e9fe + 329 + 0 + + + 1.0 + Wu, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + b3bf669489ae4913bb60ddfe50e41697 + 330 + 0 + + + 1.0 + Zhuang, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0eba9d55a3ff46298665a0c292e2237f + 331 + 0 + + + 1.0 + Lin, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 55e3f4a200eb4619ae2b6efb645464d1 + 332 + 0 + + + 1.0 + Li, D. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + c44324c171674d00a743413042e9b944 + 333 + 0 + + + 1.0 + Xing, E. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 4bdaba79a3274241ab98e27aeaf98f57 + 334 + 0 + + + 1.0 + Preprint is classified under cs.CL on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 7c8c464ed7044a7896adfeb35f58a04d + 335 + 0 + + + 1.0 + Preprint was submitted on 24 Apr 2024 + f0306814bf64f5c9e79603fc6a52f4ea + 5fa2eec73bec481b85eba22ea7a2a927 + 336 + 0 + + + 1.0 + Preprint has the identifier 2404.16130v1 on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + e6aa5eedca984c56b5fa5e179127951d + 337 + 0 + + + 1.0 + Community detection results in the partition of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1c4bd4ba4ef64a93acd55faa8fd97ca9 + 338 + 0 + + + 1.0 + The pipeline includes a step for community detection + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 5b85c70d578c4d67b5cb4743552bd559 + 339 + 0 + + + 2.0 + Dang, 2006, is the author who established the QFS approach + fb3c48579608fa28be585ceb6cd2f0fe + 956113fb770840c38bce65bb5832f988 + 340 + 0 + + + 2.0 + Baumel et al., 2018, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + 785bb55e79954b0c84a4a53cd7f0b454 + 341 + 0 + + + 2.0 + Laskar et al., 2020, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + 1239281fd3774b91a99358c9c1e6ee1c + 342 + 0 + + + 2.0 + Yao et al., 2017, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + 32b29a842b224f4c99fa1d5c764efc9a + 343 + 0 + + + 2.0 + Goodwin et al., 2020, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + f5ae7dc11fd64822a3a15e7d3839031a + 344 + 0 + + + 2.0 + Laskar et al., 2022, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + e1e254e67719488894eaa3553112a8cf + 345 + 0 + + + 2.0 + Liu and Lapata, 2019, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + ebdd79169d7d41b99faf09b039a66204 + 346 + 0 + + + 2.0 + Achiam et al., 2023, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + e036534e17b24dd2895167a20873230f + 347 + 0 + + + 2.0 + Brown et al., 2020, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + a00bc5e4be634b08b1f084b6a07abafd + 348 + 0 + + + 2.0 + Touvron et al., 2023, are the authors who worked on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + ce8241c964724429bb361b7b53867007 + 349 + 0 + + + 2.0 + Anil et al., 2023, are the authors who worked on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + 61cd7f168f7f44d6a23415e9497f1e65 + 350 + 0 + + + 1.0 + Modularity is an inherent quality of knowledge graphs + 21e52bc06a82796b1f4bcd73edda1f2a + 3be77a7b57e34c55acc1f1dfbc64ee10 + 351 + 0 + + + 1.0 + Brown et al. (2020) discuss in-context learning with few-shot examples + bc9e2c9e369c4108cf4f6dd5f60960f4 + 751c564f8ff6444d9d4c8de4a677e655 + 352 + 0 + + + 1.0 + Kuratov et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + 96963c158fb64680bded290f442ff9aa + 353 + 0 + + + 1.0 + Liu et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + bdbfbde5dd244447a2a0674b30ae3e8f + 354 + 0 + + + 1.0 + Louvain is a type of community detection algorithm + 21e52bc06a82796b1f4bcd73edda1f2a + f970bfe31db74929abff6ea38e5d18e6 + 355 + 0 + + + 1.0 + Community detection algorithms are used to partition the graph index into communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 6f0c2a8b79e6406a8ab7a20864ae2ce2 + 356 + 0 + + + 1.0 + Fortunato has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 7b09e60e33f44ffdab9c656c5b9c1d50 + 357 + 0 + + + 1.0 + Jin et al. have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 8b7beab7c0a143aea7bffc31df7528d5 + 358 + 0 + + + 1.0 + HotPotQA dataset is used to evaluate the entity extraction prompt with gpt-4-turbo + 21e52bc06a82796b1f4bcd73edda1f2a + d03eb34a0612420680555ab9f10d03d5 + 359 + 0 + + + 1.0 + Yang et al. (2018) introduced the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + b066746cdff7440c8a3591f0c098201d + 360 + 0 + + + 2.0 + Yang et al. are the authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + 1e2eded8ef7b4b458c33fbc2d36c4380 + 361 + 0 + + + 1.0 + GPT-4-Turbo was tested with varying context window sizes + 4c855404ee3d3c94aa2136f1513c666f + c59e3e931b0f4cf888c2eb70857ee753 + 362 + 0 + + + 1.0 + Tech journalist uses podcast transcripts to look for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + 305b80bb4df5488b8a34129daeeae0c7 + 363 + 0 + + + 3.0 + Kevin Scott is a participant in the podcast conversations compiled in the Podcast Transcripts dataset. His conversations are included as part of the podcast transcripts, contributing to the overall content and discussions captured within this dataset. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + 66fa0de756da440bad8da583306410c4 + 364 + 0 + + + 1.0 + Technology leaders participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + d1e9c550a0e74c48ae81c319f26ccafc + 365 + 0 + + + 2.0 + RAG systems are used to evaluate the Podcast Transcripts dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + 3730b5d759ba4fd28a54af0a02151f09 + 366 + 0 + + + 2.0 + C0 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 82b7f7c27e2348f880c94ffb80942de7 + 367 + 0 + + + 2.0 + C1 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 0980c4f558654466b4d691d0cb7ce16d + 368 + 0 + + + 2.0 + C2 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + f1e47cf5daa441649c3474c3339bb704 + 369 + 0 + + + 2.0 + C3 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 0964dcfbff934c92af8961155673ac7f + 370 + 0 + + + 2.0 + TS is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 51b82bcdffe04056bad1c082c3830047 + 371 + 0 + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + c62bb148852b49a98e2779ca23a0919d + 372 + 0 + + + 1.0 + SS is a category used in the analysis of podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + 72b5a0c357c24b739084d501b9354bc1 + 373 + 0 + + + 1.0 + Units are used to measure the context in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + c827b62ebf134e55a3ccf0b63f976870 + 374 + 0 + + + 1.0 + Tokens are used to measure the word count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + b51ef388758845e880e736309ae791e3 + 375 + 0 + + + 1.0 + % Max is used to measure the percentage of maximum token count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + 0a841cd4b6664423b033f22e3a80f33c + 376 + 0 + + + 1.0 + Both are datasets used in the analysis + 36db32c37e1987e2c5863898ad882190 + 16911c51c65b42f8a2d04c05f45b2c58 + 377 + 0 + + + 1.0 + Educator uses news articles to incorporate current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + fc3f77f29574410d991a2aa333950bf6 + 378 + 0 + + + 2.0 + RAG systems are used to evaluate the News Articles dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + 4f847eb72cbe48678d5634dcf93fc0e2 + 379 + 0 + + + 1.0 + C0 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + 829e64159ae04301982e88e93a2f0e49 + 380 + 0 + + + 1.0 + C1 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + cf37d3d4bc154f65b3d79c831c587763 + 381 + 0 + + + 1.0 + C2 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + 4b4fce341d554012bc73d7886860749e + 382 + 0 + + + 1.0 + C3 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + 9f6e7a08bd814d19b45fac58928027f8 + 383 + 0 + + + 1.0 + TS is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + ff9410fed5e64c04a875e040e3d182b2 + 384 + 0 + + + 1.0 + Units are used to measure the context in news articles + 36db32c37e1987e2c5863898ad882190 + 1161272728914953b568f384d7a9f2f1 + 385 + 0 + + + 1.0 + Tokens are used to measure the word count in news articles + 36db32c37e1987e2c5863898ad882190 + f09c82eb89944ae9846df82135123b90 + 386 + 0 + + + 1.0 + % Max is used to measure the percentage of maximum token count in news articles + 36db32c37e1987e2c5863898ad882190 + d221b743a51d464b87de3b72b85f6b59 + 387 + 0 + + + 1.0 + Map-reduce is the method used in the text summarization condition + 973164fa90bf2b4ee267f4fd795916bf + 9fd31a28e1384b40a9d1658a765871cd + 388 + 0 + + + 1.0 + The LLM evaluator assesses answers based on the empowerment metric + 322e02986c8724eedbcf3ebfa20b989c + 0119f233c8394b9584e55fadcce173f0 + 389 + 0 + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for news articles + ebf5249c888e07fedce6572a4c03f88c + 5c20b469b92446dabb1b68976807be7c + 390 + 0 + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on empowerment + 4c855404ee3d3c94aa2136f1513c666f + 2c2392247a35456da663adfcffd12e73 + 391 + 0 + + + 1.0 + Empowerment has an average win rate of 51.3% + 36db32c37e1987e2c5863898ad882190 + 167a32ff67ce4471baa8cf019ee7c17b + 392 + 0 + + + 1.0 + Naive RAG mentions Taylor Swift as a public figure + e8c8f911135faf3ff35f24107eb3f99c + 3280fc12ef414827838e6ac7089f0618 + 393 + 0 + + + 1.0 + Naive RAG mentions Travis Kelce as a public figure + e8c8f911135faf3ff35f24107eb3f99c + 556fba72a0854ce4831f6cfea6fd035e + 394 + 0 + + + 1.0 + Naive RAG mentions Britney Spears as a public figure + e8c8f911135faf3ff35f24107eb3f99c + 8e2e6eeed5a04c9f80efbcfc624ced95 + 395 + 0 + + + 1.0 + Naive RAG mentions Justin Timberlake as a public figure + e8c8f911135faf3ff35f24107eb3f99c + ea6d546f1caa4b4aaacdad8b8af195ec + 396 + 0 + + + 1.0 + Naive RAG is determined to be the loser based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + 267ce44e6dae43ee94d0d375ec08ef17 + 397 + 0 + + + 1.0 + Global approaches consistently outperformed the naive RAG + 36db32c37e1987e2c5863898ad882190 + b37e5d15f3154ee39df016b8eac8de66 + 398 + 0 + + + 1.0 + Naive RAG produces the most direct responses + 36db32c37e1987e2c5863898ad882190 + e13eb574e885414b80f0b66992767ef2 + 399 + 0 + + + 1.0 + SS represents naive RAG in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 93f4140f654e41ccba908c6f6dc65f17 + 400 + 0 + + + 1.0 + Gao et al., 2023 discusses naive RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + a102d091986749ef90b45d411e707bef + 401 + 0 + + + 1.0 + Community partitions enable divide-and-conquer global summarization + 7fb7d9ce2da9c940a32afdd87d1d9e56 + cd6ae38a5a6742899d14f4a064f42c19 + 402 + 0 + + + 1.0 + Global summarization can be performed using a graph-free approach + e4d9b12cf2b4c691c74019eefff4fb39 + fe18688bd4ef44d1a184ec6d1451a5cf + 403 + 0 + + + 1.0 + Source texts are used in global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + 0f1282bdfedb4f6e8765007a90dd2959 + 404 + 0 + + + 1.0 + Final global answer is generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + 540af5c5d4cd41ceb29c40c5fb02e2fe + 405 + 0 + + + 1.0 + Short descriptions are used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + bbf83708095f47019eaee93d6879bc77 + 406 + 0 + + + 1.0 + Low-level community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 245a56f01d1b48a7b4d88ed0e354155a + 407 + 0 + + + 1.0 + The use of rich descriptive text for homogeneous nodes in a graph index aligns with the capabilities of LLMs + 7fb7d9ce2da9c940a32afdd87d1d9e56 + d3aa564fb4eb430a8ca6813ca76bfff6 + 408 + 0 + + + 1.0 + Graph indexes differ from typical knowledge graphs in their use of rich descriptive text instead of concise knowledge triples + 7fb7d9ce2da9c940a32afdd87d1d9e56 + d9b948357d96419ca135065ce1c360ef + 409 + 0 + + + 1.0 + The graph index supports condition C0 + 973164fa90bf2b4ee267f4fd795916bf + 20a79ddd91ba48e4bb7bc194c79baaf6 + 410 + 0 + + + 1.0 + The graph index supports condition C1 + 973164fa90bf2b4ee267f4fd795916bf + b95728a0b96b405cbccafa6c12fd8722 + 411 + 0 + + + 1.0 + The graph index supports condition C2 + 973164fa90bf2b4ee267f4fd795916bf + 5d6dc034d2014e8c930fde69c31b99cf + 412 + 0 + + + 1.0 + The graph index supports condition C3 + 973164fa90bf2b4ee267f4fd795916bf + 127cbb53940f4efa8e1807b4452375ba + 413 + 0 + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + f1ea6ef9539043ab887bcce22ccf9625 + 414 + 0 + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + b50c4f053f0546029c4095b7b93aa05e + 415 + 0 + + + 1.0 + The graph index was created using generic prompts for entity and relationship extraction + 973164fa90bf2b4ee267f4fd795916bf + 0cea7f7a7fab49339cdd6fb02d0d183e + 416 + 0 + + + 1.0 + Few-shot examples tailored to the domain of the data were used in the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + 5b89f0d8101c419b86e1959cca2db848 + 417 + 0 + + + 1.0 + The graph indexing process used a context window size of 600 tokens + 973164fa90bf2b4ee267f4fd795916bf + cdb407fc600b45caa6f94f82e89d2e4f + 418 + 0 + + + 1.0 + The decision to build a graph index depends on the expected number of lifetime queries per dataset + e4d9b12cf2b4c691c74019eefff4fb39 + 7f4905fcb43e4d6ca23e6d2b40f6958e + 419 + 0 + + + 1.0 + The decision to build a graph index depends on the value obtained from it + e4d9b12cf2b4c691c74019eefff4fb39 + f5ad4fe84df544c69db25f0e30c6eace + 420 + 0 + + + 1.0 + The decision to build a graph index depends on the value obtained from other graph-related RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + 237a46cc973b41dc9af4190c71c5c9e1 + 421 + 0 + + + 1.0 + Recall measures the completeness of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + aaa27aa0b1024e3aa3c87a6ec821a348 + 422 + 0 + + + 1.0 + Precision measures the accuracy of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + f0a28fe3f68546dba7850815f7933275 + 423 + 0 + + + 1.0 + Few-shot examples are used to tailor the default prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + 45b59feba7134bc18632cb42530c189a + 424 + 0 + + + 1.0 + Few-shot examples are used to tailor the secondary extraction prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + 7747cd2048f94d378e83265b9561d921 + 425 + 0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of science + 2c6ed90897310eea2f28e33fff1c32b0 + c4e9532dbc734264a0e3e827bc8014c6 + 426 + 0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of medicine + 2c6ed90897310eea2f28e33fff1c32b0 + 003e5d505a01434596c6d65ff20b0bdf + 427 + 0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of law + 2c6ed90897310eea2f28e33fff1c32b0 + f79358f3535045d9aad3b828df59293b + 428 + 0 + + + 1.0 + A single extraction round is part of the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + 7d375c18c1e2415faecd9f7397068a32 + 429 + 0 + + + 1.0 + Domain refers to the specific area of knowledge of the document corpus + bc9e2c9e369c4108cf4f6dd5f60960f4 + dfa0e847a6704c93a0fe014b01858ff7 + 430 + 0 + + + 1.0 + Covariate prompts are used to extract claims linked to detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + 9e91823feb174cd1b6a3bf8d0a5cb86b + 431 + 0 + + + 1.0 + Source text span is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + ad76c8dc8dd94412a5e79005cf8e0f2f + 432 + 0 + + + 1.0 + Start date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 26a03482961e41918ea049018080af7a + 433 + 0 + + + 1.0 + End date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 1cfd220ff4d2493ca4b92d725d171d32 + 434 + 0 + + + 1.0 + Description is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 97738fe0830d405ba53598b5cb1e5e38 + 435 + 0 + + + 1.0 + Subject is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 354cea4f6e164a48ad12122c28a5b30d + 436 + 0 + + + 1.0 + Object is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 1ee2380c1eda4ebb8c9304820750ac88 + 437 + 0 + + + 1.0 + Communities of entities help manage variations in a noisy graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 57e00d4d4e0e4679a150f048deb80af3 + 438 + 0 + + + 1.0 + Common entities are described using rich descriptive text for homogeneous nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + c1e4a9dbe55c4fb89f0d927c9fb067a4 + 439 + 0 + + + 1.0 + LLMs are used to generate metrics for evaluating natural language generation + 973164fa90bf2b4ee267f4fd795916bf + 1474a72a5cff4b72ae6f99e804ceaa95 + 440 + 0 + + + 1.0 + Wang et al. (2023) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + 738fda68df7a49a0bae96673a8711afc + 441 + 0 + + + 1.0 + Zheng et al. (2024) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + 89dd5a0943c64247adae624abbc95afb + 442 + 0 + + + 1.0 + Relationship edges connect homogeneous nodes in a graph + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 405e9907440d4deab71f3960ae36f47b + 443 + 0 + + + 1.0 + Edge weights represent the normalized counts of detected relationship instances on relationship edges + 7fb7d9ce2da9c940a32afdd87d1d9e56 + f91e7c9600ca4623a8cc4a56d2dccd07 + 444 + 0 + + + 1.0 + Each level of the hierarchical community structure provides a community partition + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 3af2a8619c394be6adf06e4bc742b7ec + 445 + 0 + + + 2.0 + The hierarchical community structure is a framework used to organize and understand the relationships and dynamics within specialized communities. Root communities are an integral part of this structure, serving as the top-level communities. These root communities form the foundational layer in the hierarchical community structure, providing a basis for further subdivision and organization of more specific sub-communities. This hierarchical approach allows for a systematic analysis of complex networks, facilitating a deeper understanding of the interconnections and dependencies within the domain of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + c10ffc51dcb54708a1dc757693010bfe + 446 + 0 + + + 2.0 + The hierarchical community structure is a framework that organizes communities into different levels, with sub-communities representing the lower-level communities within this structure. Sub-communities are integral components of the hierarchical community structure, indicating that they are nested within larger communities and contribute to the overall organization and dynamics of the community. This hierarchical arrangement allows for a more detailed and nuanced understanding of the relationships and interactions within the community, facilitating more effective analysis and mapping of complex text data, particularly in specialized domains such as Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + e67ce34d48364422973ccf3a6b57af83 + 447 + 0 + + + 1.0 + Community levels are part of the hierarchical community structure + 843fc5421e086120ffa1c75856ecf6cd + 98773a34c9bb474d8a789ea08f57250e + 448 + 0 + + + 1.0 + The Leiden algorithm is used to detect communities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + ae260498423e4d55aa413423cd0eb20b + 449 + 0 + + + 1.0 + OpenORD is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 4aeecb9d885743ca9373337a43957dd8 + 450 + 0 + + + 1.0 + Force Atlas 2 is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 1121b50f7858427fa679d81861238825 + 451 + 0 + + + 1.0 + Nodes represent entities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 6e3c8aa3abab475bb0148faa9112f0bf + 452 + 0 + + + 1.0 + Edges represent connections between nodes in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 948a00e8ee1246cc90c47b292d03ddff + 453 + 0 + + + 1.0 + Covariates are variables linked to nodes and edges in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 45c42e619f5e488f914608780dcf0579 + 454 + 0 + + + 2.0 + Tang and Yang are the authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 2b3bea0d9ede41f193828526bcb8e02c + 455 + 0 + + + 1.0 + Questions are generated based on the target datasets + 1d07b4248c2655081c7af0e373bd70c9 + 6b2586cc1f8e4dc8af64913af63d9837 + 456 + 0 + + + 1.0 + N represents the number of test questions per dataset + 973164fa90bf2b4ee267f4fd795916bf + 7983bfa8d173414685272b3844d6612e + 457 + 0 + + + 1.0 + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + 09294e8220a445e288ea8841f234a440 + 458 + 0 + + + 1.0 + Root communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + d4e043cf972c4d129b6b855f1731caae + 459 + 0 + + + 1.0 + Level 0 represents the root-level communities in the hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + e0d63137270c426dbbfe7fcf78c474de + 460 + 0 + + + 1.0 + Reports provide detailed information about specific subtopics within sub-communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + c50bca18bc454a98b935df012b7fd6f9 + 461 + 0 + + + 1.0 + Sub-communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + 434b133c64bd46219e67c6eb296ad0ff + 462 + 0 + + + 1.0 + Level 1 represents sub-communities within the root-level communities + 843fc5421e086120ffa1c75856ecf6cd + cb895bf7e7c147e6b5d923b6c8f67d63 + 463 + 0 + + + 1.0 + Partitions can be organized into a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 87776e869a01402499a317cb9cf09453 + 464 + 0 + + + 1.0 + Level 0 is the root level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + d1e5359d2e344260bf1b83823df839b7 + 465 + 0 + + + 1.0 + Level 1 is a sub-level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 0522f6580b824bc39792b695fc8be66b + 466 + 0 + + + 1.0 + The token limit defines the maximum number of tokens in the LLM context window + 843fc5421e086120ffa1c75856ecf6cd + 580fd6d19460460fa40613f66b3ee200 + 467 + 0 + + + 1.0 + Prominence is used to prioritize community edges + 843fc5421e086120ffa1c75856ecf6cd + 84f4684a7a5241c18bb087ccb00550d3 + 468 + 0 + + + 1.0 + Combined source and target node degree is used to measure prominence + 843fc5421e086120ffa1c75856ecf6cd + 9607ba4a796f46be8d4f79bc7065d60b + 469 + 0 + + + 1.0 + Chunks are divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + 236dd7dce9ee4cf5918fddd44b4863e5 + 470 + 0 + + + 1.0 + Helpfulness scores are assigned to intermediate answers + 1d07b4248c2655081c7af0e373bd70c9 + 9e92fed814a64d9d88bfab9a227859d3 + 471 + 0 + + + 1.0 + Tech journalist is interested in episodes dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + 7dccecb29d3a419093b279b22e207539 + 472 + 0 + + + 1.0 + Tech journalist is interested in how guests perceive the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + 89857eb61e63461cbad7c5014f5098f9 + 473 + 0 + + + 1.0 + Tech journalist is interested in discussions about the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + 7b2e7a0d910c4988a7b64489f4159a65 + 474 + 0 + + + 1.0 + Tech journalist is interested in suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + 38630cf0996f4cff8d32b2dbdaa5ba85 + 475 + 0 + + + 1.0 + Tech journalist is interested in discussions about collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + bd0fb68ac7014b91a314c93ec55897f5 + 476 + 0 + + + 1.0 + Educator is interested in current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + f24dcb3cd6d644f8af2b6c47983e280b + 477 + 0 + + + 1.0 + Educator is interested in how news articles address the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + b1cad695afbc4ec3bbcd46ea34bd26ca + 478 + 0 + + + 1.0 + Educator is interested in examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + 72f7974758d74e5d89ddb64ad739abb8 + 479 + 0 + + + 1.0 + Educator is interested in insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + e6ee83249adf4e14b98d1676b1c6b05f + 480 + 0 + + + 1.0 + Educator is interested in highlighting the importance of health literacy through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + f805fd9fe42947a38b92a3db6e8cc986 + 481 + 0 + + + 1.0 + The size of the context window and the prompts used for answer generation are the same across all conditions + 973164fa90bf2b4ee267f4fd795916bf + e8b956218d5c4e5d9d390abcf527a514 + 482 + 0 + + + 1.0 + The task is an activity or goal that the user aims to achieve + 1d07b4248c2655081c7af0e373bd70c9 + 9525aa223d774e62ad856c2201cfab1b + 483 + 0 + + + 1.0 + Questions are generated based on the user's task + 1d07b4248c2655081c7af0e373bd70c9 + 1087596b06d1400a8f863d0ac1af64a4 + 484 + 0 + + + 1.0 + Datasets were used in combination with questions for the analysis + 4c855404ee3d3c94aa2136f1513c666f + 39058965295643c8a7738350cc18ceac + 485 + 0 + + + 1.0 + Questions were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + 9a8a2e5e3f2645619a0403532d935afe + 486 + 0 + + + 2.0 + Zheng et al. are the authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + f0c21c67baac47f097f74f5055b89877 + 487 + 0 + + + 1.0 + Zheng, L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 323a4c7407ac401db79a6023c3a5a17d + 488 + 0 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 686bc2bd59644e398dde88ffd37bf49b + 489 + 0 + + + 1.0 + Sheng, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + e368f8e9c9864acc880fdb5113631f3f + 490 + 0 + + + 1.0 + Zhuang, S. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 05063c19ddb847a89ae1746588464288 + 491 + 0 + + + 1.0 + Wu, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 019b34e800414f7b87f38a14adf2eb67 + 492 + 0 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 1064a663ca4742a78e743128546f6d87 + 493 + 0 + + + 1.0 + Lin, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 9a5e0a4ae34f46b39a5a028cbc135264 + 494 + 0 + + + 1.0 + Li, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 5a224002ecbc4725abeb5a424aaca6a6 + 495 + 0 + + + 1.0 + Li, D. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 8826a17bbda34012b3ea84d58ae531eb + 496 + 0 + + + 1.0 + Xing, E. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + bab69d76defb402da2a2a358739f1497 + 497 + 0 + + + 1.0 + MT-Bench and Chatbot Arena are both tools used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + ea465e5cd92247829f52ff0c8591d1bb + 498 + 0 + + + 2.0 + Koesten et al. authored a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + 2dbac25b512c4f21965169a95a910a94 + 499 + 0 + + + 2.0 + Xu and Lapata authored a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + 97958ed004f645b1b331fa0e66faa313 + 500 + 0 + + + 1.0 + Text summarization method applies a map-reduce approach directly to source texts (TS) + 973164fa90bf2b4ee267f4fd795916bf + 48129b4ee99f4e30843fd4395d4815c0 + 501 + 0 + + + 1.0 + Text summarization is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + 6de4c00e48b3480883e696e24df9fda4 + 502 + 0 + + + 1.0 + Semantic search RAG is a na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached (SS) + 973164fa90bf2b4ee267f4fd795916bf + 4b3d236101de4904ab348e3e3b11b4be + 503 + 0 + + + 1.0 + Semantic search RAG is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + 5be2ce9957ba404f939b6c8175015619 + 504 + 0 + + + 1.0 + C0 uses root-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + fe77344850214c1cac923094de81098c + 505 + 0 + + + 1.0 + C0 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 798f739abfc14a13bf3911d0a9cfb63b + 506 + 0 + + + 1.0 + C0 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 3105de8188fd41d88d0dbf0a5d48e443 + 507 + 0 + + + 1.0 + C0 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + 8108dde0e62a48008a270138a690a0b9 + 508 + 0 + + + 1.0 + C1 uses high-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + d07207b853c14504a44eea1d4778f902 + 509 + 0 + + + 1.0 + C1 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 406822a1a01a4140baf9bbf1d479f07e + 510 + 0 + + + 1.0 + C1 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + fe47ba3762ae4feda39904d59cbb4160 + 511 + 0 + + + 1.0 + C1 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 53032c2afcb5474a88446ad7c5506980 + 512 + 0 + + + 1.0 + C1 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + 5c66a88612a245cb91fbba9c094f12fc + 513 + 0 + + + 1.0 + C2 uses intermediate-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + b4c54fb7ce0b4b77afd5fbe5a8a2527f + 514 + 0 + + + 1.0 + C2 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + c8b60cdb74104667b5d2b4b70d74d039 + 515 + 0 + + + 1.0 + C2 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + 333e294d7cc34df4abc47ad9ced3d186 + 516 + 0 + + + 1.0 + C2 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 15e66e10d12f4520abca20985d2cb39c + 517 + 0 + + + 1.0 + C2 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + 2a271d9b5d7b46fea4046d5590eed1d7 + 518 + 0 + + + 1.0 + C3 uses low-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + 99e372089bed4a0394af57175679f8e4 + 519 + 0 + + + 1.0 + C3 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 62afe93767684ea38f861d20fb05ff71 + 520 + 0 + + + 1.0 + C3 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + 8fc1fbff7e6c459c93ce2c2f5a62226e + 521 + 0 + + + 1.0 + C3 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 04b3ae04020349a9bc568f26d17eab14 + 522 + 0 + + + 1.0 + C3 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + bbc4d367c60f41ad8a279c12e5cc7da6 + 523 + 0 + + + 1.0 + TS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 9a1aff251eda416ea6270e6158e663fc + 524 + 0 + + + 1.0 + TS is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 7a9e50846c274338ab09e7313b540edb + 525 + 0 + + + 1.0 + TS is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + b268cc3ef860434ba663dd46af633cc5 + 526 + 0 + + + 1.0 + SS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 1c9f67904a4c4fcc8cdac6a605900248 + 527 + 0 + + + 1.0 + The graph indexing process used 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + e00c403d1dc84ba6a37ee193596e320f + 528 + 0 + + + 1.0 + A graph was created for the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 3f2e726c3b624fe7bf11de9be2c0457e + 529 + 0 + + + 1.0 + Units are used to measure the context in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + f71dc0c394f04771af7e2ed37f85647e + 530 + 0 + + + 1.0 + Tokens are used to measure the word count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 2fea9c1856e54a91b79a9ce85755fbf5 + 531 + 0 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 7823b4c5b3364c5f890d05f33a46bdde + 532 + 0 + + + 1.0 + Intermediate-level summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 183f3a0b73ff41c5bb4a19fd7adf0c1d + 533 + 0 + + + 1.0 + The graph indexing process used 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + 392e06f17d724484a9cfb85fe69aac50 + 534 + 0 + + + 1.0 + A graph was created for the News dataset + 36db32c37e1987e2c5863898ad882190 + 6f49e00cdac04a358173ecd40351ee00 + 535 + 0 + + + 1.0 + Units are used to measure the context in the News dataset + 36db32c37e1987e2c5863898ad882190 + 3fef96af4ec343da8c34f8b09518de8a + 536 + 0 + + + 1.0 + Tokens are used to measure the word count in the News dataset + 36db32c37e1987e2c5863898ad882190 + bd403eff654e42c997e5656a2b1c1a20 + 537 + 0 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the News dataset + 36db32c37e1987e2c5863898ad882190 + 5763d829837144f199fac2b490b38110 + 538 + 0 + + + 1.0 + Datasets were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + 234c6f1859f0405ab607f0be53e7b06c + 539 + 0 + + + 1.0 + Wang et al., 2023a discusses the state-of-the-art results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + 21800eab85b94d4880bcada7a60763e5 + 540 + 0 + + + 1.0 + Zheng et al., 2024 discusses the competitive results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + b8bb28a7a9624b6d805be89adfe29eb5 + 541 + 0 + + + 1.0 + Zheng et al., 2024 discusses the LLM-as-a-judge method + 322e02986c8724eedbcf3ebfa20b989c + 61f26f8850504d56a6b7cd764c33299d + 542 + 0 + + + 1.0 + Embedding-based matching is used to match user queries + e4d9b12cf2b4c691c74019eefff4fb39 + d4456fac0ada4b6fbe3cfee873403d00 + 543 + 0 + + + 1.0 + Query-time LLM use was evaluated with different context window sizes + 4c855404ee3d3c94aa2136f1513c666f + f8fd3fcf650b47b2b1692506ebe77762 + 544 + 0 + + + 2.0 + The **CONTEXT WINDOW SIZE** and **FINAL EVALUATION** are closely related in the given data. A fixed context window size of 8k tokens was used for the final evaluation. This indicates that during the final evaluation phase, the system or model was configured to process and analyze text data within a predefined window of 8,000 tokens, ensuring consistency and standardization in the evaluation process. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + d95acc24180c47caa34114627d501592 + 545 + 0 + + + 1.0 + Natural Language Generation achieves state-of-the-art results + 322e02986c8724eedbcf3ebfa20b989c + f4753ab09adc42a9a52754e95440d4b9 + 546 + 0 + + + 1.0 + Natural Language Generation achieves competitive results + 322e02986c8724eedbcf3ebfa20b989c + 12f5a7c56b454a3d8aae97f65908f96b + 547 + 0 + + + 1.0 + Natural Language Generation is compared against human judgements + 322e02986c8724eedbcf3ebfa20b989c + 95f79ff0b8a34080ae2ac8448ce561f1 + 548 + 0 + + + 1.0 + Natural Language Generation can generate reference-based metrics + 322e02986c8724eedbcf3ebfa20b989c + 8733d4602c084e1cab1384dde0306abf + 549 + 0 + + + 1.0 + Natural Language Generation can measure qualities in a reference-free style + 322e02986c8724eedbcf3ebfa20b989c + ded3a49efdf6479a991cad53d0758cf4 + 550 + 0 + + + 1.0 + Es et al., 2023 discusses the RAGAS method + 322e02986c8724eedbcf3ebfa20b989c + 816fceb7e1ca4b5d9277368f78e6ed80 + 551 + 0 + + + 1.0 + RAGAS evaluates context relevance + 322e02986c8724eedbcf3ebfa20b989c + 50539d4503a4495097f49a8ed83e2462 + 552 + 0 + + + 1.0 + RAGAS evaluates faithfulness + 322e02986c8724eedbcf3ebfa20b989c + d6f67aa7ef0e4a19bf5830e777aafea5 + 553 + 0 + + + 1.0 + RAGAS evaluates answer relevance + 322e02986c8724eedbcf3ebfa20b989c + bbf61f9cd3e14f46a010d704e86be008 + 554 + 0 + + + 1.0 + The LLM evaluator assesses answers based on the directness metric + 322e02986c8724eedbcf3ebfa20b989c + 5d34e587bd2f41dba285e9178f179577 + 555 + 0 + + + 1.0 + Table 2 shows an example of LLM-generated assessment + 322e02986c8724eedbcf3ebfa20b989c + 901b491be7344401b4544ff05e591a0e + 556 + 0 + + + 1.0 + The LLM evaluator uses a head-to-head comparison approach + 322e02986c8724eedbcf3ebfa20b989c + ecacbf62b81d485396a56e1730e75a04 + 557 + 0 + + + 1.0 + The LLM evaluator assesses answers based on target metrics + 322e02986c8724eedbcf3ebfa20b989c + ba0ad1bcf02b4928a1b7ff7b23acdd6f + 558 + 0 + + + 1.0 + The LLM evaluator uses a control metric for validity + 322e02986c8724eedbcf3ebfa20b989c + 0e3c66c25d7e43a7960c37d28315e5d8 + 559 + 0 + + + 1.0 + The LLM evaluator accounts for stochasticity + 322e02986c8724eedbcf3ebfa20b989c + a0e0d5b7db9f4efcb5277856db799775 + 560 + 0 + + + 1.0 + The LLM evaluator uses mean scores from multiple comparisons + 322e02986c8724eedbcf3ebfa20b989c + 3f85dab93736440f9776020b6410aa9b + 561 + 0 + + + 1.0 + Directness is used to evaluate the straightforwardness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + 710ed70c346342ff81ccf205e30271bb + 562 + 0 + + + 1.0 + The question asks about public figures mentioned in entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + b3d3e8ba2ede4574a0498f082f0c15ae + 563 + 0 + + + 1.0 + Public figures are repeatedly mentioned across various entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + 8686013390614eca9116ccbab27431d7 + 564 + 0 + + + 1.0 + Answer 1 covers a wide range of public figures from different sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + fd8c8b7e3b9248abb1d8cb8958ab86d3 + 565 + 0 + + + 1.0 + Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports. + 718017a4871c909420f84b85b8ba969d + 039594428123415f95deb246f5097169 + 566 + 0 + + + 1.0 + Controversies involve public figures and impact public discourse. + 718017a4871c909420f84b85b8ba969d + d78ce7696ff14234a544de945ffe40d6 + 567 + 0 + + + 1.0 + Entertainment articles cover topics related to the entertainment industry + 322e02986c8724eedbcf3ebfa20b989c + 59b21508be904875af22b5c1cfdcd211 + 568 + 0 + + + 1.0 + Taylor Swift is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + e9c7a1d505b14229afbbef7c0d04751e + 569 + 0 + + + 1.0 + Travis Kelce is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 4b0efcd54efc40e8a884ac6c31deada2 + 570 + 0 + + + 1.0 + Britney Spears is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 0970f08f3d1a4d638d44e2ccb9237382 + 571 + 0 + + + 1.0 + Justin Timberlake is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 8f10c11ecb5142029869025521c73431 + 572 + 0 + + + 1.0 + Taylor Swift is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + e36a0e3901864a7eaa5f5ad4280a6471 + 573 + 0 + + + 1.0 + Travis Kelce is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6fce354faa104fe58ba8a565eb3c43f2 + 574 + 0 + + + 1.0 + Britney Spears is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 20585e9a43c04375aa334e946e2dd144 + 575 + 0 + + + 1.0 + Justin Timberlake is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 32e343c0ae454660bdfcd1d3133baf0a + 576 + 0 + + + 1.0 + Actors and Directors are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 505ab840f6cc4fa6a839ebfe82d255ed + 577 + 0 + + + 1.0 + Musicians and Executives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + e38eb1698900424bb7392a74ff0f3351 + 578 + 0 + + + 1.0 + Athletes and Coaches are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 855c57eecf2a45c7aab02ff1ac36938d + 579 + 0 + + + 1.0 + Influencers and Entrepreneurs are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6ee77949c94d4906bd98c24341fdfa03 + 580 + 0 + + + 1.0 + Public Figures in Controversy are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + d06f506604b249feb423915db282ed75 + 581 + 0 + + + 1.0 + Film is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 0f642f63d4af4fc38298822bfc952719 + 582 + 0 + + + 1.0 + Television is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + e017ad1f09b049a7ad41d5a11dc1e3d9 + 583 + 0 + + + 1.0 + Music is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 5cbced0ba7044b7490f520a436261c57 + 584 + 0 + + + 1.0 + Sports is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + d45dea925f8d4e7e93d0e17317001eec + 585 + 0 + + + 1.0 + Digital Media is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 8123eee04a3a4c779f03bdb85de99f9f + 586 + 0 + + + 1.0 + Cultural Narratives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6129d90c83194bcfaede9ff00a011297 + 587 + 0 + + + 1.0 + Trends are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6ef76e963a564dbe9c9feff4f8ce1683 + 588 + 0 + + + 1.0 + Social Discussions are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 1c8bad73fda646f8b3f413e432f0e351 + 589 + 0 + + + 1.0 + Public Discourse is a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 7e75749d13d24321b8b10c5be0138805 + 590 + 0 + + + 1.0 + Reference-based metrics require gold standard answers + 322e02986c8724eedbcf3ebfa20b989c + 05bfaf60aa304a288e6789443bd6fd6c + 591 + 0 + + + 1.0 + Gold standard answers are lacking for sensemaking questions + 322e02986c8724eedbcf3ebfa20b989c + 6097e047a74d41ca996a0b7949ef6f0e + 592 + 0 + + + 3.0 + End users play a crucial role in the validation process of sensemaking questions and target metrics. Sensemaking questions are specifically validated with end users to ensure their relevance and accuracy. This collaborative approach ensures that the questions and metrics are aligned with the needs and expectations of the end users, thereby enhancing the overall effectiveness and applicability of the sensemaking process. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + e257439ce5be47a88faaeb0fe01bc4a1 + 593 + 0 + + + 1.0 + Target metrics are validated with end users + e4d9b12cf2b4c691c74019eefff4fb39 + 067b9486d59f45d2963235220f723a41 + 594 + 0 + + + 1.0 + The control metric is used as an indicator of validity + 322e02986c8724eedbcf3ebfa20b989c + 87c46c7ead5447bc8309ab116a316959 + 595 + 0 + + + 1.0 + Taylor Swift is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + f607d795f00347109cab3b2370c414f7 + 596 + 0 + + + 1.0 + Taylor Swift is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + 8f0610c89e9f42e9b8c3d8a947fa2852 + 597 + 0 + + + 1.0 + Travis Kelce is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + 75ef3591790a49748154ddbba20e9cdf + 598 + 0 + + + 1.0 + Travis Kelce is a public figure in the sports sector. + 718017a4871c909420f84b85b8ba969d + 58b7f26cb17b4b2283d3cacbaed15cfc + 599 + 0 + + + 1.0 + Britney Spears is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + 277cdf13617e47ca883b949f495bc243 + 600 + 0 + + + 1.0 + Britney Spears is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + 26080c121c9645b2bb258e4d61d47672 + 601 + 0 + + + 1.0 + Justin Timberlake is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + ee91a06f13b4495f95c800a0c7329ef7 + 602 + 0 + + + 1.0 + Justin Timberlake is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + 6ed8b67be79242e98aa1b9283431d5df + 603 + 0 + + + 1.0 + Answer 1 includes public figures from the film sector. + 718017a4871c909420f84b85b8ba969d + 40c2425cb1c34c1591f7cb89f9f5e0bf + 604 + 0 + + + 1.0 + Answer 1 includes public figures from the television sector. + 718017a4871c909420f84b85b8ba969d + 7cf59650687a435ba26a7c5ffc6c4f4c + 605 + 0 + + + 1.0 + Answer 1 includes public figures from the music sector. + 718017a4871c909420f84b85b8ba969d + 53c2882604b74192a649a4eaa0536c5e + 606 + 0 + + + 1.0 + Answer 2 focuses on public figures primarily from the music sector. + 718017a4871c909420f84b85b8ba969d + 3fbb8aeacea54ca9a957118fba613ccf + 607 + 0 + + + 1.0 + Answer 1 includes public figures from the sports sector. + 718017a4871c909420f84b85b8ba969d + 496ae6a894584a6cb12e50b516341788 + 608 + 0 + + + 1.0 + Answer 2 focuses on public figures primarily from the sports sector. + 718017a4871c909420f84b85b8ba969d + dd1a82c597794ba3a490cb70d488d9dd + 609 + 0 + + + 1.0 + Answer 1 includes public figures from the digital media sector. + 718017a4871c909420f84b85b8ba969d + bbd206ae4c1a4794813fd239fcfef313 + 610 + 0 + + + 1.0 + Answer 1 cites specific data sources from the News article dataset for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + 708ac105e8bb4beeade0472c899f214d + 611 + 0 + + + 1.0 + Answer 1 provides insights into controversies involving public figures and their impact on public discourse. + 718017a4871c909420f84b85b8ba969d + b4fe3c6aea95472db73a5e8bf575895a + 612 + 0 + + + 1.0 + Answer 1 includes public figures from the gaming sector. + 718017a4871c909420f84b85b8ba969d + a861f44aa7dd414790ee82b3f651c609 + 613 + 0 + + + 1.0 + Answer 1 cites specific data sources for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + add9948a221a4aabafbaaed650b1db26 + 614 + 0 + + + 1.0 + Answer 2 was generated using the Naïve RAG method, which directly lists specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d + 6c04f2ab7c9843ea900c3444b014bed8 + 615 + 0 + + + 2.0 + ANSWER 2 is a generated answer for a question in the NEWS ARTICLE DATASET. It relies heavily on a single source from the NEWS ARTICLE DATASET for data. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + 88356435ca9d43ebaf93134b3af8a53e + 616 + 0 + + + 1.0 + Answer 2 relies heavily on a single data source. + 718017a4871c909420f84b85b8ba969d + 233edf428a04436a8d32849af584f9d8 + 617 + 0 + + + 1.0 + Naïve RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + 28a6317db3d141db82a4a22525265fef + 618 + 0 + + + 1.0 + The global approach to Graph RAG shows improvements over naïve RAG + e4d9b12cf2b4c691c74019eefff4fb39 + 90051a1b69cd40f696e440d54085887e + 619 + 0 + + + 1.0 + LLM-generated assessments are used to evaluate the answers produced for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + 973fe6e8a1314a269748f40a98786115 + 620 + 0 + + + 1.0 + Example question is part of the News article dataset used for analysis + ebf5249c888e07fedce6572a4c03f88c + 9a442a8c054d48339aff04923bafe47f + 621 + 0 + + + 1.0 + Head-to-head win rate percentages were used to compare different conditions + 4c855404ee3d3c94aa2136f1513c666f + ffdacb33c3a94b7f9d890d7cc03a1f40 + 622 + 0 + + + 1.0 + Win rate percentages were used to measure the performance of different conditions + 4c855404ee3d3c94aa2136f1513c666f + 8792fc245cc94235a7764481ebad4828 + 623 + 0 + + + 1.0 + The overall winner per dataset and metric was determined for each condition + 4c855404ee3d3c94aa2136f1513c666f + b5982d09c32e4e7387e88f9160b4dd78 + 624 + 0 + + + 1.0 + Self-win rates were shown as the expected 50% for each condition + 4c855404ee3d3c94aa2136f1513c666f + 04ed223f57e44cf18284ba42ba760423 + 625 + 0 + + + 1.0 + The indexing process resulted in the creation of graphs + 36db32c37e1987e2c5863898ad882190 + 0debfb49a28d480db1b7d5ef713cac8f + 626 + 0 + + + 1.0 + Map-reduce summarization requires the highest number of context tokens + 36db32c37e1987e2c5863898ad882190 + 1f9abc7d006f4afa86200385acc3d1ae + 627 + 0 + + + 1.0 + Root-level community summaries require dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + cac3f76fbc11413e92cdfd3064d56ece + 628 + 0 + + + 2.0 + Queries are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + f120d98b793a4276a6f1a0a8e51a589a + 629 + 0 + + + 2.0 + Self-memory is related to generation-augmented retrieval + f35de4d9fb65f1d5a392064b20545c19 + bfda4c94278b49ab98cd3f407980d4d8 + 630 + 0 + + + 2.0 + CAiRE-COVID is a system for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + 7e5e5b80b84749d98cb36f56dbfcb47b + 631 + 0 + + + 2.0 + ITRG is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + 6330604339ca4113b94624bc9bed5ede + 632 + 0 + + + 2.0 + IR-CoT is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + 9638492c4f034be6b3bf88f8abd82edc + 633 + 0 + + + 2.0 + DSP is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + a76322b81f644f3c8733d04fa046b4e4 + 634 + 0 + + + 2.0 + RAPTOR is a method for generating a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + 653ee6584dbc46d1b8e97a05a3eac81e + 635 + 0 + + + 2.0 + The paper by Baek et al. discusses the KAPING method + 92e93fc6449756c0a60200636b297f65 + 9f0d58a479ec404d8e8f493f9269b08d + 636 + 0 + + + 2.0 + The paper by He et al. discusses the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + c5ae09d00a3f417981fc4177ef333eff + 637 + 0 + + + 2.0 + The paper by Zhang discusses the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + 4dd086fcba704d26b976c08a81c1465c + 638 + 0 + + + 2.0 + The paper by Kang et al. discusses the SURGE method + 92e93fc6449756c0a60200636b297f65 + f1ef6375ea84496eaed13c03318d80c6 + 639 + 0 + + + 2.0 + The paper by Ranade and Joshi discusses the FABULA method + 92e93fc6449756c0a60200636b297f65 + ba6829116d114532b99530f101ff0c72 + 640 + 0 + + + 2.0 + Both LangChain and LlamaIndex support a variety of graph databases + 92e93fc6449756c0a60200636b297f65 + 1ab2048463174873883061373d480ac4 + 641 + 0 + + + 2.0 + LangChain supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + 97038fe907af4710859c3daeb13972e9 + 642 + 0 + + + 2.0 + LangChain supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + e0595082eb9f41a4ac2afd9e614b363c + 643 + 0 + + + 1.0 + LangChain developed Langchain graphs + 71f6daf11e64e5273a3847d46bf228e1 + 5bd2ef268d4f4ba18925c17242370e21 + 644 + 0 + + + 2.0 + LlamaIndex supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + 96a21e764d1143fc90de0b2cc7751983 + 645 + 0 + + + 2.0 + LlamaIndex supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + 801b7142ab5b4c5eac41dade999a7c1f + 646 + 0 + + + 2.0 + NaLLM is a method that can create and reason over knowledge graphs in Neo4J format + 92e93fc6449756c0a60200636b297f65 + aac39de4e7e74d1c83f0eb835e635c88 + 647 + 0 + + + 2.0 + Neo4J developed Project NaLLM + 833e7d67dcd30790b26b71c9b5306f6b + c2e801c8221c4806a4f59ba5b793c784 + 648 + 0 + + + 2.0 + GraphRAG is a method that can create and reason over knowledge graphs in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + 351fc21986564103b324540289e2e608 + 649 + 0 + + + 2.0 + The paper by Manakul et al. discusses the SelfCheckGPT method + 92e93fc6449756c0a60200636b297f65 + 1c8a90b0aed7439286bbf85903d423d4 + 650 + 0 + + + 1.0 + SelfCheckGPT is an approach mentioned in the work by Manakul et al., 2023 + e4d9b12cf2b4c691c74019eefff4fb39 + 6c98609312154f118c04d8781663b16a + 651 + 0 + + + 1.0 + SelfCheckGPT is used to compare fabrication rates + e4d9b12cf2b4c691c74019eefff4fb39 + b91a6bf16e334b3ab7ec57665e980ceb + 652 + 0 + + + 1.0 + Embedding-based matching is used to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + c067e41adf9840df89052b111e6c0a6a + 653 + 0 + + + 1.0 + Hybrid RAG schemes combine embedding-based matching against community reports + e4d9b12cf2b4c691c74019eefff4fb39 + 76d7feb8140b4064b5492d3055736ee0 + 654 + 0 + + + 1.0 + The roll-up operation can be extended using map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + 53273797509d45178c49045830ec9fc2 + 655 + 0 + + + 1.0 + The drill down mechanism follows the information scent in the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + 2383fd3c3b4a4249a5a96550c494edb2 + 656 + 0 + + + 1.0 + The global approach to Graph RAG achieves competitive performance at a fraction of the token cost + e4d9b12cf2b4c691c74019eefff4fb39 + 29f172df150042e0a6db5481d5d91cfc + 657 + 0 + + + 1.0 + The open-source implementation of Graph RAG approaches is Python-based + e4d9b12cf2b4c691c74019eefff4fb39 + a243935f440241a281fbabb20422c641 + 658 + 0 + + + 1.0 + The drill down mechanism follows the information scent + e4d9b12cf2b4c691c74019eefff4fb39 + 34b704124fe94c2f933a344c11165f2e + 659 + 0 + + + 1.0 + Alonso Guevara Fernández and Amber Hoak both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + e429a497b01c40f3aef7e2205eaf01d8 + 660 + 0 + + + 1.0 + Alonso Guevara Fernández and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + f521933b91564693b07bd838160083ac + 661 + 0 + + + 1.0 + Alonso Guevara Fernández and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + e0d361b6991b40debf5599e86f2638ca + 662 + 0 + + + 1.0 + Alonso Guevara Fernández and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + cf16005cfadf4e48832ffd0e43f57be1 + 663 + 0 + + + 1.0 + Alonso Guevara Fernández and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + db10b0e690754748b0d75639f3e8d2b8 + 664 + 0 + + + 1.0 + Alonso Guevara Fernández and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + ec13b80209e246588bb5486d516f85eb + 665 + 0 + + + 1.0 + Alonso Guevara Fernández and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 617a76d54ed546e29428a31dea955b96 + 666 + 0 + + + 1.0 + Alonso Guevara Fernández and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + c88ffd2aa7284ac38eb4351c5fad6f44 + 667 + 0 + + + 1.0 + Alonso Guevara Fernández and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0dace3b24df14aae909a2815653e9db1 + 668 + 0 + + + 1.0 + Alonso Guevara Fernández and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + b3cfde857302479aa59b91d6648a40df + 669 + 0 + + + 1.0 + Alonso Guevara Fernández and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 01a52a986b6a444badc83fb11aa7a160 + 670 + 0 + + + 1.0 + Alonso Guevara Fernández and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2516b3485b754bdabf6820863c918e3d + 671 + 0 + + + 1.0 + Alonso Guevara Fernández and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0523ed6330f4429f8468f5b49169c940 + 672 + 0 + + + 1.0 + Alonso Guevara Fernández and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0b7ac591dfd34971b24d38e344b40c37 + 673 + 0 + + + 1.0 + Alonso Guevara Fernández and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2864f68297e94d7e84213833e22da077 + 674 + 0 + + + 1.0 + Alonso Guevara Fernández and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 629cd969d05c4c329bbe24f5d86e0089 + 675 + 0 + + + 1.0 + Alonso Guevara Fernández and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + f7bc5d1fb1404acdb77d50a6b9129141 + 676 + 0 + + + 1.0 + Alonso Guevara Fernández and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + d47535a52c2b40a3bacb3d520b8f0f1c + 677 + 0 + + + 1.0 + Alonso Guevara Fernández and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 557ed8720c2845cabcce0287f7284b3e + 678 + 0 + + + 1.0 + Alonso Guevara Fernández and Sarah Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + cef8ff96a0e842fdae4751933bcb1a28 + 679 + 0 + + + 1.0 + Alonso Guevara Fernández and Shane Solomon both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 93c8356953da486e9630d7e7304a6ff3 + 680 + 0 + + + 1.0 + Amber Hoak and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + dc52f3641c1548bba5b3cf8c65a5c072 + 681 + 0 + + + 1.0 + Amber Hoak and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + d0fdcc6945d84b20aa1de4afe2786592 + 682 + 0 + + + 1.0 + Amber Hoak and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + af6e03260c5946be96737b148b5edd9d + 683 + 0 + + + 1.0 + Amber Hoak and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1949eb874c544c58a71bbd04d6241a22 + 684 + 0 + + + 1.0 + Amber Hoak and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 09384ed4453846cb8c4d0076ecbf928a + 685 + 0 + + + 1.0 + Amber Hoak and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 181833ae880a4d0ab24ba0ccb158138d + 686 + 0 + + + 1.0 + Amber Hoak and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 3426a7ea02f740aeabcb552feee11bcc + 687 + 0 + + + 1.0 + Amber Hoak and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + e47ae0eea85b4f6e86b77fe56396460e + 688 + 0 + + + 1.0 + Amber Hoak and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1ad7e0ad19334488b5d3b008f93a4ef4 + 689 + 0 + + + 1.0 + Amber Hoak and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 8c76a8cb5951422ba3b3cc6fcb66a391 + 690 + 0 + + + 1.0 + Amber Hoak and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 096dee591c1f4141a73fd628a59ffbe9 + 691 + 0 + + + 1.0 + Amber Hoak and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 9886a385806e46a69d92a726017b99b6 + 692 + 0 + + + 1.0 + Amber Hoak and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 4ecf891e9a5b4daf9e02d5b2ec963079 + 693 + 0 + + + 1.0 + Amber Hoak and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 59207227178147e39296a4059ac1055d + 694 + 0 + + + 1.0 + Amber Hoak and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0682e47ed49146c0bc5e2b77fb924b6c + 695 + 0 + + + 1.0 + Amber Hoak and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 6022e4aa784f42b88dbcb27a5d9d2614 + 696 + 0 + + + 1.0 + Amber Hoak and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 386047cff59549ea83158b69bbac1870 + 697 + 0 + + + 1.0 + Amber Hoak and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2ead31e49ce643ebae4d5f047bb7a37b + 698 + 0 + + + 1.0 + J. Achiam and S. Adler co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + a74ee5b02e1e41b0ac4cf5449f7cdf2c + 699 + 0 + + + 1.0 + J. Achiam and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0f73fcdab31348a880a468124099071c + 700 + 0 + + + 1.0 + J. Achiam and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 24d2dcb7f28144cbad714b0a8b6c9e70 + 701 + 0 + + + 1.0 + J. Achiam and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 093447e0342e490aa6a55bd70ce7c2f2 + 702 + 0 + + + 1.0 + J. Achiam and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0439db7ac7d2484596e02246bd340424 + 703 + 0 + + + 1.0 + J. Achiam and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 81ffef86ebb341bebf145c742fb33dbd + 704 + 0 + + + 1.0 + J. Achiam and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + a0da2872126f43769f75c8533fca5e26 + 705 + 0 + + + 1.0 + J. Achiam and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 7facdc1f91014f42a67e34bac31a95ce + 706 + 0 + + + 1.0 + J. Achiam and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + ca7a635373294067b5f3050c82d38983 + 707 + 0 + + + 1.0 + S. Adler and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 2833c46f05984f729c7ec15e071f0c8e + 708 + 0 + + + 1.0 + S. Adler and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + ada7cf1171b74ad793f7856febc9c6fe + 709 + 0 + + + 1.0 + S. Adler and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 33b355c0a8044ef2b2b8be81bea0d431 + 710 + 0 + + + 1.0 + S. Adler and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 85698526e09a47878e3255a251d95406 + 711 + 0 + + + 1.0 + S. Adler and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 1a625c51e7ad497b86041757d1cde642 + 712 + 0 + + + 1.0 + S. Adler and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + df91c0e5657a4bafa849c8a3079ca582 + 713 + 0 + + + 1.0 + S. Adler and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 091a9788b29443509feda24aa5f5c241 + 714 + 0 + + + 1.0 + S. Adler and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 6f750deaedcb4612b419c3d8dd7e5cb2 + 715 + 0 + + + 1.0 + S. Agarwal and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + d5ea3f061e9c419fb1c07b680bfb287a + 716 + 0 + + + 1.0 + S. Agarwal and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 1f4fd98283df43c69d5537c002b98f58 + 717 + 0 + + + 1.0 + S. Agarwal and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + f3bb299bf6454785a8a406dce9776789 + 718 + 0 + + + 1.0 + S. Agarwal and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 218d3d5a4a544df99caed612e48add5b + 719 + 0 + + + 1.0 + S. Agarwal and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 02393af06d3649549b3e9290b4e46c0a + 720 + 0 + + + 1.0 + S. Agarwal and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + ffa9208f15744978a4ea45c1cff18a86 + 721 + 0 + + + 1.0 + S. Agarwal and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 23ea2a1d78984eb38721adeadee662e1 + 722 + 0 + + + 1.0 + L. Ahmad and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0d0a729e30634e1fb198609ce10c69bf + 723 + 0 + + + 1.0 + L. Ahmad and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 896e7d03cad7450e8044fcb0fd9f6e92 + 724 + 0 + + + 1.0 + L. Ahmad and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + a777a0c3a34b4990899f2e1e1f1c2074 + 725 + 0 + + + 1.0 + L. Ahmad and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 6cd46299702049bcbd39407fa97f0dc0 + 726 + 0 + + + 1.0 + L. Ahmad and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 7af01185ebd648e49bf9a57481e0dc7c + 727 + 0 + + + 1.0 + L. Ahmad and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + d6aad3495b4d4c7ab2a03c44600584ba + 728 + 0 + + + 1.0 + I. Akkaya and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 482d5ecf4ce949e9a5d81f1b368769ee + 729 + 0 + + + 1.0 + I. Akkaya and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 64f7a02f89bd4a37844c482f00d00643 + 730 + 0 + + + 1.0 + I. Akkaya and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 26609be86d614d85ae97deeae4a4be1e + 731 + 0 + + + 1.0 + I. Akkaya and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 5dffe69693734eaeb360de4582d489b0 + 732 + 0 + + + 1.0 + I. Akkaya and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + e4fe96090a7641c68d0b1995d1f238b4 + 733 + 0 + + + 1.0 + F. L. Aleman and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 3fa1055cd26840678d546570e8b423d9 + 734 + 0 + + + 1.0 + F. L. Aleman and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9802dae4757b42269e93c66b5214a396 + 735 + 0 + + + 1.0 + F. L. Aleman and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + c39e66acec274a5980ce275709a847ba + 736 + 0 + + + 1.0 + F. L. Aleman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 667794d1397a40bb904d406205960864 + 737 + 0 + + + 1.0 + D. Almeida and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 891f50162c0140e4b9c0e4ba33f69a1b + 738 + 0 + + + 1.0 + D. Almeida and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 922773e5841a475d89d5904fe7a324f8 + 739 + 0 + + + 1.0 + D. Almeida and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0338bebae19c41c196ee6c09ccba36e3 + 740 + 0 + + + 1.0 + J. Altenschmidt and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0b407647077c4288b2324f06ac355985 + 741 + 0 + + + 1.0 + J. Altenschmidt and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 4e9254fd4b234106843cf8ff91fd3b6f + 742 + 0 + + + 1.0 + S. Altman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + e6a7f4ccd6f54136b784572db0d5cb88 + 743 + 0 + + + 1.0 + R. Anil and S. Borgeaud co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + ccf54360ef954353b71c1c8175cd7f4e + 744 + 0 + + + 1.0 + R. Anil and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + bd1c72f46b81427892b1f415fecce77e + 745 + 0 + + + 1.0 + R. Anil and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 148e7caecdf740e58ee09a9ff549d19c + 746 + 0 + + + 1.0 + R. Anil and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + df78f3e3415a4d47b6dffdd3890f3eee + 747 + 0 + + + 1.0 + R. Anil and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 003abb3c5743482aa63022cf20cf5ccc + 748 + 0 + + + 1.0 + R. Anil and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 4788433078b843079ccd9a64e5430169 + 749 + 0 + + + 1.0 + R. Anil and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + cbecfdbc04c9405aa139566d727d3a33 + 750 + 0 + + + 1.0 + R. Anil and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + f514a867efb948868009b435fecbe372 + 751 + 0 + + + 1.0 + S. Borgeaud and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 6e4c8a7f1da147f5b38103c51c999502 + 752 + 0 + + + 1.0 + S. Borgeaud and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + ac0e52b6b3ae4cc485f9eef2f2dea7e7 + 753 + 0 + + + 1.0 + S. Borgeaud and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + b150af2f3df24f17a7fd836ba663680a + 754 + 0 + + + 1.0 + S. Borgeaud and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + f9740be4adc946149b5941f355d45c74 + 755 + 0 + + + 1.0 + S. Borgeaud and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + c6afe51b28f94c3ba21640387edd2ee8 + 756 + 0 + + + 1.0 + S. Borgeaud and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 040b86f496dc4930a895f2c21cb0731c + 757 + 0 + + + 1.0 + S. Borgeaud and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + e13ed464574f483c9f1db5f569e91445 + 758 + 0 + + + 1.0 + Y. Wu and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + e1131985dc53451fa7543912b2e7db07 + 759 + 0 + + + 1.0 + Y. Wu and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + d6c00e0a975e4adc979afd25d4037d4d + 760 + 0 + + + 1.0 + Y. Wu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 9cac1113be2148ce8abaa957620f9d59 + 761 + 0 + + + 1.0 + Y. Wu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + c277297e3e7b417892e986c8767f58ad + 762 + 0 + + + 1.0 + Y. Wu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 744f492f46d841c0b0fee5f4a9b40b6c + 763 + 0 + + + 1.0 + Y. Wu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + d59386dbfa0349b49f7b904e288b21ad + 764 + 0 + + + 1.0 + J.-B. Alayrac and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 1005971b7e764bffa0a4610ad403976b + 765 + 0 + + + 1.0 + J.-B. Alayrac and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 1eaf3527e2804c75bbd9e3ccac9d760e + 766 + 0 + + + 1.0 + J.-B. Alayrac and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 301ab7b5f81d40788e46dacb09579b50 + 767 + 0 + + + 1.0 + J.-B. Alayrac and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + d6f25f03a08b41b4a2eaa9df3db9dceb + 768 + 0 + + + 1.0 + J.-B. Alayrac and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 6529cf63872440a98aeab73beee3762a + 769 + 0 + + + 1.0 + J. Yu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 65faae6c13f5444b8d71b4b2be38eba3 + 770 + 0 + + + 1.0 + J. Yu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 9187af05545a4c8d92e38c2b46254092 + 771 + 0 + + + 1.0 + J. Yu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + f12578d1ff7b46f5ae84c7672fac8deb + 772 + 0 + + + 1.0 + J. Yu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 6b5c42bba0ec48c1a5de177a7f1b9bfc + 773 + 0 + + + 1.0 + R. Soricut and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 444ab529d10c47f19ef33e931489b8b8 + 774 + 0 + + + 1.0 + R. Soricut and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 0ffea4c3c86849ab828036b67b58acdc + 775 + 0 + + + 1.0 + R. Soricut and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 65b488142164407a81b496b4820ef556 + 776 + 0 + + + 1.0 + J. Schalkwyk and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + b3b006f8629b44df81a266c1e4d81d3f + 777 + 0 + + + 1.0 + J. Schalkwyk and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 2a48f66b6a424b9ebf38562836fe1c82 + 778 + 0 + + + 1.0 + A. M. Dai and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 54e486668fa94feda285f377fb05d14d + 779 + 0 + + + 1.0 + J. Baek and A. F. Aji co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + d7c4899260084560905ac54dba81f0e6 + 780 + 0 + + + 1.0 + J. Baek and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 880db899ce864932843fe230e3d364ad + 781 + 0 + + + 1.0 + A. F. Aji and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 59463c48d2fb48308cd38ee8dd869f59 + 782 + 0 + + + 1.0 + T. Ban and L. Chen co-authored the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + ff69f1aae7404c38b8bde6abc5a79b57 + 783 + 0 + + + 1.0 + Baumel, T. and Eyal, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 8db22f709edb4ae98f0fef060ccd24b8 + 784 + 0 + + + 1.0 + Baumel, T. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 0e412e834e62475a9fe1920438f7b75b + 785 + 0 + + + 1.0 + Baumel, T. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + 2e3f6dbac98742ddb213037ae77f0a82 + 786 + 0 + + + 1.0 + Eyal, M. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 59ced15f5a1d485ebf0eac7fa85c1cdf + 787 + 0 + + + 1.0 + Eyal, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + 7cef69e2a71c4379b0816844799fc71e + 788 + 0 + + + 1.0 + Elhadad, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + 0f565c275f8148d885ae53c315ddc568 + 789 + 0 + + + 1.0 + Blondel, V. D. and Guillaume, J.-L. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 29c05af68cc541b79fdf499eac42b9c6 + 790 + 0 + + + 1.0 + Blondel, V. D. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 8d5d7b8fb7d14d0ba46ce7f0be6de661 + 791 + 0 + + + 1.0 + Blondel, V. D. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 607a66de21cf42e497c23013327b751f + 792 + 0 + + + 1.0 + Guillaume, J.-L. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + f4f85b6086384211a25248f614bfb786 + 793 + 0 + + + 1.0 + Guillaume, J.-L. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + c1a4bcd4e7874e699f06bc795e291150 + 794 + 0 + + + 1.0 + Lambiotte, R. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + ca585d891433495aa70a3a01b252e50c + 795 + 0 + + + 1.0 + Brown, T. and Mann, B. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + f26b5e1c52e445998b6a63738d203b38 + 796 + 0 + + + 1.0 + Brown, T. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 0945030309e14518a16df16fbb25c76f + 797 + 0 + + + 1.0 + Brown, T. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 35939cc34a734b5f867f8d75df419f37 + 798 + 0 + + + 1.0 + Brown, T. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 807561f61906451b880e04ac6a33687f + 799 + 0 + + + 1.0 + Brown, T. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 0538e2dc04174140a43bc0359fed2d23 + 800 + 0 + + + 1.0 + Brown, T. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 6e87d210775b45e4a09e518492329bce + 801 + 0 + + + 1.0 + Brown, T. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + bde6223f81884473a1acc3b75dd056aa + 802 + 0 + + + 1.0 + Brown, T. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 7beb3a2ecfd5419b950a20a155e06169 + 803 + 0 + + + 1.0 + Brown, T. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 961dda09b0df497a974c38c28eb90686 + 804 + 0 + + + 1.0 + Mann, B. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 1fb946075dd54b218b8dfad20647d33e + 805 + 0 + + + 1.0 + Mann, B. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + bf90efb1858e49b19987cbd280d0e911 + 806 + 0 + + + 1.0 + Mann, B. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + f9949e43ea004014abec1b59f2155b5a + 807 + 0 + + + 1.0 + Mann, B. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 3d89c7fe0b6448e0a0d27bceccc09f09 + 808 + 0 + + + 1.0 + Mann, B. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 185f98d04d9f484ab3d626fd459a23a2 + 809 + 0 + + + 1.0 + Mann, B. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 688e9b443bc44782855aea4afd8a9d16 + 810 + 0 + + + 1.0 + Mann, B. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + cd56c4963a0e49d7bab0e25f0e068779 + 811 + 0 + + + 1.0 + Mann, B. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + c4d4216677af42f5a29a0f4dcb442220 + 812 + 0 + + + 1.0 + Ryder, N. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + d99684f88f2d43eaacd62ba9082b64a5 + 813 + 0 + + + 1.0 + Ryder, N. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9ca07c62b7e146298882e33f3c6cb653 + 814 + 0 + + + 1.0 + Ryder, N. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + eef3aeb29aba43da93b433a816e77203 + 815 + 0 + + + 1.0 + Ryder, N. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 1c2a05515b9f49e1966a4ceb4bb0a3a5 + 816 + 0 + + + 1.0 + Ryder, N. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 86db00646d264b0a922c6b639dc9d16b + 817 + 0 + + + 1.0 + Ryder, N. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 7892ab98e1b0475c97a798aa8b2d7f6c + 818 + 0 + + + 1.0 + Ryder, N. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9732636cdd50433bb146a241cd72dbc5 + 819 + 0 + + + 1.0 + Subbiah, M. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + a5a8a63d5baf4946b7d7d1696f0e4e0e + 820 + 0 + + + 1.0 + Subbiah, M. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + a69a82dd8773426096c58ddc56832770 + 821 + 0 + + + 1.0 + Subbiah, M. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 72a7215e3e4a4b0db851351dfe5afd37 + 822 + 0 + + + 1.0 + Subbiah, M. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 5f8224530d424618acb32b74a3afe2c9 + 823 + 0 + + + 1.0 + Subbiah, M. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 99d4510732d843299514461aebd5f176 + 824 + 0 + + + 1.0 + Zhao, D. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 1730cbfab07747508d5b5ea421b97953 + 825 + 0 + + + 1.0 + Es, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + 7dd3aadc8f424988a72f8ba3ccf17155 + 826 + 0 + + + 1.0 + James, J. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + a2e7cbaf38c24564b2abe61680cacd72 + 827 + 0 + + + 1.0 + Espinosa-Anke, L. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + 162f1ccf8dfc46cea4d54a36ed9ec823 + 828 + 0 + + + 1.0 + Schockaert, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + 17127080df794121830177e93631aa3b + 829 + 0 + + + 1.0 + Feng, Z. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 374a8f81e6304b6d90b44cdceb90ecb4 + 830 + 0 + + + 1.0 + Feng, X. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 8c6aafd2a5da496385bea2c69be03a5a + 831 + 0 + + + 1.0 + Yang, M. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 06b9da0d4d9b4d6bb762bd2eeca7028a + 832 + 0 + + + 1.0 + Qin, B. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + a5f6de474fb24ec9af7403231c616831 + 833 + 0 + + + 1.0 + Gao, Y. and Xiong, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + a0ef87eb823b400594300f5c47e5c9c3 + 834 + 0 + + + 1.0 + Gao, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + ed076834490640acbb5d837aaac9fed5 + 835 + 0 + + + 1.0 + Gao, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + f935684600f34a27906def1902627ff2 + 836 + 0 + + + 1.0 + Gao, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 72fc2f604fc644e39f7d70e25094e347 + 837 + 0 + + + 1.0 + Gao, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 3ddb886b220c4bb2ab3d68f7f29ce5c5 + 838 + 0 + + + 1.0 + Gao, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 2f482bb08d564072a5ff4f2509dfdda6 + 839 + 0 + + + 1.0 + Gao, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + fc86507186da4c6c94fe3b788d77c471 + 840 + 0 + + + 1.0 + Gao, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + c235c2c649484c83967e2a42523028bb + 841 + 0 + + + 1.0 + Xiong, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 7e9d748907ea4b74925a32999a2b40d9 + 842 + 0 + + + 1.0 + Xiong, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + eefaef8a5c7149d18d304f39bf41f280 + 843 + 0 + + + 1.0 + Xiong, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 327114716cea49a79d33ba609158cd87 + 844 + 0 + + + 1.0 + Xiong, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + b45920e405af45f787ab167f54cfd2e9 + 845 + 0 + + + 1.0 + Xiong, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 00db8f6e99254c99be6c6f5c14a79500 + 846 + 0 + + + 1.0 + Xiong, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + aadd82f0e70c4fc49b1bdee3f60c1890 + 847 + 0 + + + 1.0 + Xiong, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 9528d92ccc10454793c4df59e24586db + 848 + 0 + + + 1.0 + Gao, X. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 80618f4e809e4af1bcdb59342c375377 + 849 + 0 + + + 1.0 + Gao, X. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 32d785e275be458fb7178ad2021ecdfc + 850 + 0 + + + 1.0 + Gao, X. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 0757f97d1fbf49748169ba696a364e4c + 851 + 0 + + + 1.0 + Gao, X. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + ca9a355bf38b452cbde62dba747ec65f + 852 + 0 + + + 1.0 + Gao, X. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + ba297c67512447e4b86f0cbc39fbc301 + 853 + 0 + + + 1.0 + Gao, X. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 00a9c8745b404b659c76a694dba9851c + 854 + 0 + + + 1.0 + Jia, K. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + eb338f2214414f0f9fa396f06ca12860 + 855 + 0 + + + 1.0 + Jia, K. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + d119680bac3744e58d2ed3273b1208b6 + 856 + 0 + + + 1.0 + Jia, K. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + d60eefaddf1e4b1db125d8f9ac49bacb + 857 + 0 + + + 1.0 + Jia, K. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 6058343c0824402e9843c92b2991f778 + 858 + 0 + + + 1.0 + Jia, K. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 383a003edc5a4f2387c7dd7865a984c9 + 859 + 0 + + + 1.0 + Pan, J. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 6bf9623c44824e48b7451bdfa1b47816 + 860 + 0 + + + 1.0 + Pan, J. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + cc62f333666e427eb1c66ec3f12a7a55 + 861 + 0 + + + 1.0 + Pan, J. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + c6d99481f5f545278ca8a73650b66e87 + 862 + 0 + + + 1.0 + Pan, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 5a0887b99d8b4bd89286962cd6f07037 + 863 + 0 + + + 1.0 + Bi, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 75174e7af26f434c9154b182087b58dc + 864 + 0 + + + 1.0 + Bi, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 0c43dd117fe6495991d4b4d8c2f5d70e + 865 + 0 + + + 1.0 + Bi, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 65e5d27c4f8a4dfa8ad92f227964b9cf + 866 + 0 + + + 1.0 + Dai, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 63c4595187884af29aa46d03319acded + 867 + 0 + + + 1.0 + Dai, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + b32482039edd4d50bc43514570500345 + 868 + 0 + + + 1.0 + Sun, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + f3e6e2c82bab4430a33987a19e3d1835 + 869 + 0 + + + 1.0 + Goodwin, T. R. and Savery, M. E. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 6768cc24da5d4a2492ff936dd4b35661 + 870 + 0 + + + 1.0 + Goodwin, T. R. and Demner-Fushman, D. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 223e1e3e7c4f4282b086e940f8c935c2 + 871 + 0 + + + 2.0 + Khattab, O. and Santhanam, K. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + f5bb365c9a814b909df0351498d79bb5 + 872 + 0 + + + 2.0 + Khattab, O. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and highlights their collaborative work in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + ab4ad26863b44497a1e48aa7c17a096c + 873 + 0 + + + 2.0 + Khattab, O. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 145bc384290c49228a231ac124ce88a8 + 874 + 0 + + + 2.0 + Khattab, O. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + f64c99537adf489ea58940e417cb5924 + 875 + 0 + + + 2.0 + Khattab, O. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + fe98ea566cf6486b85f8ed14aabb2618 + 876 + 0 + + + 2.0 + Khattab, O. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 6bfb83cb716745fcb591c8d2fb54f8f4 + 877 + 0 + + + 1.0 + Khattab, O. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 9866640f891944c7bb0a08748aa8b91f + 878 + 0 + + + 2.0 + Santhanam, K. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This collaboration is mentioned in the text, highlighting their joint contribution to the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + b0d513adad104e14a89a767a66f30848 + 879 + 0 + + + 2.0 + Santhanam, K. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + db37a25770a34437b472fa0038837868 + 880 + 0 + + + 2.0 + Santhanam, K. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 67746ba67d80491da102aab7704dfd30 + 881 + 0 + + + 2.0 + Santhanam, K. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 873a1ca522e6461090d5cdebc2c9ae98 + 882 + 0 + + + 2.0 + Santhanam, K. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + d919ccd28e2248b5ab1dcdd7af8b00cf + 883 + 0 + + + 1.0 + Santhanam, K. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 1883a3ca4d6a4bfd984e7053e2553e16 + 884 + 0 + + + 2.0 + Li, X. L. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 2a0c93cf781a4020aceef7230b286bbf + 885 + 0 + + + 2.0 + Li, X. L. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 06282cc8998f4b2ea43e0a9522383639 + 886 + 0 + + + 2.0 + Li, X. L. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 182efa2240c14212bb021746a18936bd + 887 + 0 + + + 2.0 + Li, X. L. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 727bc610ea9a4393bfa5de453b84340f + 888 + 0 + + + 1.0 + Li, X. L. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 960c3b508a294332ba7c05ffd897db31 + 889 + 0 + + + 2.0 + Hall, D. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + bd0c1d79ef1144a49f3ce09d4cdf099b + 890 + 0 + + + 2.0 + Hall, D. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + b8ae79ed2d6d43f98e0808b5bea884dd + 891 + 0 + + + 2.0 + Hall, D. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + a03b33a4ee97467c808946679e240ddf + 892 + 0 + + + 1.0 + Hall, D. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 00b685bcb7a54c4493cd78da1f4752ab + 893 + 0 + + + 2.0 + Liang, P. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 9a54e0361b684d24aefdc05fc340cf41 + 894 + 0 + + + 2.0 + LIANG, P. and ZAHARIA, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + d423b97f085947bd89529bc1ed2c41a7 + 895 + 0 + + + 1.0 + Liang, P. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 6fd2b0d5156c424a89cb1c068cf1e076 + 896 + 0 + + + 2.0 + Potts, C. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + a8c8dd8ddbc44363ac2102b9b8989c6d + 897 + 0 + + + 1.0 + Potts, C. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 00cb0db6e46749f7af97701ad26e23be + 898 + 0 + + + 1.0 + Zaharia, M. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 4baa53e4336d4807964fa8d186b32bc5 + 899 + 0 + + + 1.0 + Kim, G. and Kim, S. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + f325a83e0c854a7ba5d46663ddff1a29 + 900 + 0 + + + 1.0 + Kim, G. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + c2b7e5d9761e423a81149a94537f6def + 901 + 0 + + + 1.0 + Kim, G. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 3d1ccd312d3a4e7387e888aaa137c7c2 + 902 + 0 + + + 1.0 + Kim, G. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 840966e7ac4a4b14ac912e75102d50b7 + 903 + 0 + + + 1.0 + Kim, G. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + f0ede764bdb1437b8cfcc20ca9598712 + 904 + 0 + + + 1.0 + Kim, S. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + dd38d562c21f444190768c8a154280da + 905 + 0 + + + 1.0 + Kim, S. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 1e66c1cbb56b41269555d27e1505ec92 + 906 + 0 + + + 1.0 + Kim, S. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + a653bd47ad3d4009ab6a5b8e6ff18679 + 907 + 0 + + + 1.0 + Kim, S. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 37b7cf055e604ec6927a9f0b15b2698d + 908 + 0 + + + 1.0 + Jeon, B. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 182e0f3d0abd4181820acdd2bf8e5eaf + 909 + 0 + + + 1.0 + Jeon, B. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 6fd6398a9bfd496f9a0505d9f3190362 + 910 + 0 + + + 1.0 + Jeon, B. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 2192d6436ca840a1bce77dbf9fd354af + 911 + 0 + + + 1.0 + Park, J. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + ef5e0bbdb3774a22900cf45e9b8863ad + 912 + 0 + + + 1.0 + Park, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 307de250d35e43a8b122c4232fa8fb7c + 913 + 0 + + + 1.0 + Kang, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 0e2f68c8ff734b279b7aad333bcf2fda + 914 + 0 + + + 1.0 + Klein, G. and Moon, B. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 172b0d0fa0794494a3c50b135c1f2cd6 + 915 + 0 + + + 1.0 + Klein, G. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 96f016e343b34ac894b0b7153f474ab0 + 916 + 0 + + + 1.0 + Klein, G. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + d356499ae9a345b6bbfb33b5fa01f47b + 917 + 0 + + + 1.0 + Moon, B. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + d7ebead985b34576821f30d83a416cd2 + 918 + 0 + + + 1.0 + Moon, B. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + 02c1447586fc46dfa65b793e0105a878 + 919 + 0 + + + 1.0 + Hoffman, R. R. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + 1b1560bb4b0447e5860f8ba351af112e + 920 + 0 + + + 1.0 + Koesten, L. and Gregory, K. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 23b929895979486cba3bf6a13f4ce655 + 921 + 0 + + + 1.0 + Koesten, L. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + f1ebe367253a4a4088b363a6cc4601a1 + 922 + 0 + + + 1.0 + Koesten, L. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + cf575adc3eb140f9aec33757ec040eb8 + 923 + 0 + + + 1.0 + Koesten, L. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + 4e581bc7d4424c2fb0023e5b11687e02 + 924 + 0 + + + 1.0 + Gregory, K. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + ae9f8a02ac0f43d4ba67ccce412989d6 + 925 + 0 + + + 1.0 + Gregory, K. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + b972541545604529a30cabc262d83dae + 926 + 0 + + + 1.0 + Gregory, K. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + ec3f73ffbb9742e090b65893d040434b + 927 + 0 + + + 1.0 + Groth, P. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 20be7b3222174d31839fac6a278f8b61 + 928 + 0 + + + 1.0 + Groth, P. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + 35688e258b0e4cc78c8b92ef8a13d3e3 + 929 + 0 + + + 1.0 + Simperl, E. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + 3a9503f2d69343c396c9b1d842d1aa74 + 930 + 0 + + + 1.0 + Kuratov, Y. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 39516d28d39e49a2b80e6cfac32e2609 + 931 + 0 + + + 1.0 + Bulatov, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 43cf2e01543540789eb8781fdb5f287d + 932 + 0 + + + 1.0 + Anokhin, P. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 54beb6d012d844058715f8ef8a91c5da + 933 + 0 + + + 1.0 + Sorokin, D. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 5a291bf96ac141b98730ac27c96e829e + 934 + 0 + + + 1.0 + Sorokin, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 5e572d8b0a614ce1839ec9a568078cdc + 935 + 0 + + + 1.0 + Burtsev, M. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 3f7b36b371da40568ce15510a35b58e7 + 936 + 0 + + + 1.0 + Laskar, M. T. R. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + 6db7771bcc674e4ead899fbdd417930f + 937 + 0 + + + 2.0 + Laskar, M. T. R. and Hoque, E. co-authored two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning with transformer models to improve the effectiveness of query-focused abstractive summarization. Both works contribute to advancing the application of transformer models in specialized summarization tasks. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 176a96262ad64ccbacb1efdfb36bd88a + 938 + 0 + + + 1.0 + Laskar, M. T. R. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + bd120225f7b84bbdb0567048ca803e3c + 939 + 0 + + + 1.0 + Laskar, M. T. R. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + 8ecda003a3d044279b1f0bdc1c96c25e + 940 + 0 + + + 1.0 + Laskar, M. T. R. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + 8e02a61bda6a4470b693e7e234abfc94 + 941 + 0 + + + 1.0 + Hoque, E. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + b62e3858d801445facc3a501c5100723 + 942 + 0 + + + 1.0 + Hoque, E. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + 64e8971970e94ea79d10e46c55b3e761 + 943 + 0 + + + 1.0 + Hoque, E. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + b10fd3628e7a45d29a2814771f53ad60 + 944 + 0 + + + 1.0 + Hoque, E. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + 826bb16575a141d683fb871ec94517e0 + 945 + 0 + + + 1.0 + Huang, J. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + 52384316108d433397224cb36486407c + 946 + 0 + + + 1.0 + The 33rd Canadian Conference on Artificial Intelligence is also known as Canadian AI 2020 + 6cd82819982879bd164547d2773ba5c7 + 9a27717e1a1b499981031fd69c58aff1 + 947 + 0 + + + 1.0 + Springer published the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + 4efbe8fc23a64506b36d6cf29f968baa + 948 + 0 + + + 1.0 + Huang, J. X. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + d3b80330258d412f9ac6a7670fe79044 + 949 + 0 + + + 1.0 + Lewis, P. and Perez, E. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 52703e888bf4493b866186b889d85783 + 950 + 0 + + + 1.0 + Lewis, P. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 5e25a814a9a04bcda6017c9cc99880a7 + 951 + 0 + + + 1.0 + Lewis, P. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + bb2070d133f74049b88c96510fc807ba + 952 + 0 + + + 1.0 + Lewis, P. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + d0734be4aaab40eb9f2be6229f4a803c + 953 + 0 + + + 1.0 + Lewis, P. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + a5839bfcc6c0471c9337257ed05b361b + 954 + 0 + + + 1.0 + Lewis, P. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + bee441f5c41e41ff8220254bbf714eb4 + 955 + 0 + + + 1.0 + Lewis, P. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 8d93d45adbe547f78460a9ef3eb40ab2 + 956 + 0 + + + 1.0 + Lewis, P. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + c6f21de31a6f4fbda2eed1780ffed5b1 + 957 + 0 + + + 1.0 + Lewis, P. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 07c3f03764874b7680710ca030cdb60c + 958 + 0 + + + 1.0 + Perez, E. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 7a7990b6045c440ba606d142bd8ddc02 + 959 + 0 + + + 1.0 + Perez, E. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 803cb895664c40319ca40cc9abb6a03d + 960 + 0 + + + 1.0 + Perez, E. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4c9e4e36560946699b6cb1e67b1437ae + 961 + 0 + + + 1.0 + Perez, E. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 8a7a2d6266424b9f9006502e82fcd778 + 962 + 0 + + + 1.0 + Perez, E. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 6b38285eedc544b08b444ee781db9f0c + 963 + 0 + + + 1.0 + Perez, E. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9d59b69c7c984abb9d3e281c04e73510 + 964 + 0 + + + 1.0 + Perez, E. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 83ee1d8db753419f8b240f419a139815 + 965 + 0 + + + 1.0 + Perez, E. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9211b015bb074bcd89ae6c75ec10e6da + 966 + 0 + + + 1.0 + Piktus, A. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + cbc280dac03a4c19bb6737e3789c928f + 967 + 0 + + + 1.0 + Piktus, A. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 166cfa435aac4465943f59c2d04a0da1 + 968 + 0 + + + 1.0 + Piktus, A. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 44bced1d9b184aa29376cf3b0cdac625 + 969 + 0 + + + 1.0 + Piktus, A. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + ca58c4e2fae84899a780ff379e1927eb + 970 + 0 + + + 1.0 + Piktus, A. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 2060ce64f028490798a3ed69832e048d + 971 + 0 + + + 1.0 + Piktus, A. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + fa8fa48e2a7542fc8ff2c43c35e1b32b + 972 + 0 + + + 1.0 + Piktus, A. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + a37bd07ff1694b6c90572399f084e1ec + 973 + 0 + + + 1.0 + Petroni, F. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 826e32d5ea1d455f8ae2d3b77cd2b41e + 974 + 0 + + + 1.0 + Petroni, F. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 1a3bd511e04d4929a45a36fb80127353 + 975 + 0 + + + 1.0 + Petroni, F. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + d0cd49577d6a49f4a21fdc389aa84805 + 976 + 0 + + + 1.0 + Petroni, F. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + aa667f462aae45a8a700d83a68c1982f + 977 + 0 + + + 1.0 + Petroni, F. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 37f40795969a42b3b50e61b76a96fa07 + 978 + 0 + + + 1.0 + Petroni, F. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 6db48bd5e4ce4337aaac4648376ed07d + 979 + 0 + + + 1.0 + Karpukhin, V. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 422e6a449e7e4ce69182113a6493a4e5 + 980 + 0 + + + 1.0 + Xu, Y. and Lapata, M. co-authored the paper "Text summarization with latent queries" + fc4b27d64f055b7fc30176ba110dd02e + 68511afc6e204c0b996d76cb75de081c + 981 + 0 + + + 1.0 + Huang, M. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + 050f02fbf9e64d08b108c5b921581335 + 982 + 0 + + + 1.0 + Duan, N. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + 97367a34fd6b4451b6be397496d646ea + 983 + 0 + + + 3.0 + Martin, S. and Brown, W. M. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with large graph structures, making it a valuable resource for researchers and practitioners in the domain of graph theory and network analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 80da3caff5344d56b6ca12660594949a + 984 + 0 + + + 3.0 + Martin, S. and Klavans, R. co-authored the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + e7efb7b459ae4ed4aa412cd20d808970 + 985 + 0 + + + 3.0 + Martin, S. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the capabilities and applications of the Openord toolbox, emphasizing its utility in handling extensive graph data efficiently. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + f0cc756b32314a1aae3e3cbb507850a2 + 986 + 0 + + + 3.0 + Brown, W. M. and Klavans, R. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + f3082b58c8a54c538cf3a0110296955b + 987 + 0 + + + 3.0 + Brown, W. M. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + c5f2254b67c04ad4add88875e5623e5a + 988 + 0 + + + 3.0 + KLAVANS, R. and BOYACK, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 6e151c66c5574df39a7f243858e2ad3f + 989 + 0 + + + 2.0 + Newman, M. E. published the paper "Modularity and community structure in networks" in the Proceedings of the National Academy of Sciences + 833e7d67dcd30790b26b71c9b5306f6b + 166366ae9ec842ec9a1deeb13c94026e + 990 + 0 + + + 2.0 + Ram, O. and Levine, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + cdecc293edb847ae92c3bf8ff39d1e9a + 991 + 0 + + + 2.0 + Ram, O. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 3a615d980a124616a659136b4fd277b7 + 992 + 0 + + + 2.0 + Ram, O. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + a9b46d1f9ef747b69d6211386b5aaa20 + 993 + 0 + + + 2.0 + Ram, O. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 891a6dbec2ef4a039efaca78040b00c1 + 994 + 0 + + + 2.0 + Ram, O. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 981d45442a11448097acebc6080da414 + 995 + 0 + + + 2.0 + Ram, O. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + fad1c8144b504954bea46ede106d93ec + 996 + 0 + + + 2.0 + Levine, Y. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 53edb7e587204ed48e523c6f1f8f4056 + 997 + 0 + + + 2.0 + Levine, Y. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + f386b02e36884167a5db1a12ee6fcb1a + 998 + 0 + + + 2.0 + Levine, Y. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 2809d8a73b71495ca4220571dd54ba1e + 999 + 0 + + + 2.0 + Levine, Y. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 49f82fa775fb466bb9ae3db14db5b29a + 1000 + 0 + + + 2.0 + Levine, Y. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + f9312ee0bac24ea1b497e16e0958d621 + 1001 + 0 + + + 2.0 + Dalmedigos, I. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + c84bc84ffea84df9ad25ae9f972b4ec0 + 1002 + 0 + + + 2.0 + Dalmedigos, I. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + be1068c6efb24cde96e5a523eb04aee8 + 1003 + 0 + + + 2.0 + Dalmedigos, I. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + ecc4f28a7d574a5886f4c80a0b7cddd4 + 1004 + 0 + + + 2.0 + Dalmedigos, I. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + b506a4813da44600b136e949db4f2832 + 1005 + 0 + + + 2.0 + Muhlgay, D. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 77735209cabb415289c8ae4e102ff6df + 1006 + 0 + + + 2.0 + Muhlgay, D. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 1f1d88a6f6ce46bab94a4b50693c89ff + 1007 + 0 + + + 2.0 + Muhlgay, D. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 03454aaf00c54112a09ea4e52185b195 + 1008 + 0 + + + 2.0 + Shashua, A. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 3c5cd9cbad0d456cab4c76f1dfcde25b + 1009 + 0 + + + 2.0 + Shashua, A. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 8503eae9f6c746afae0caa58070f25e6 + 1010 + 0 + + + 2.0 + Leyton-Brown, K. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 6e3c655e5b544277a62832a0974aa0ed + 1011 + 0 + + + 2.0 + Ranade, P. and Joshi, A. co-authored the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + bd0363bace6b42b0b3879bed5a064274 + 1012 + 0 + + + 2.0 + Sarthi, P. and Abdullah, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 99969eec8bf8441eaf9cb004cb61a13e + 1013 + 0 + + + 2.0 + Sarthi, P. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 0bcceb946a94486faf935f58dabea978 + 1014 + 0 + + + 2.0 + Sarthi, P. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 4d8421b4a6c74627afaa45aefa08c43a + 1015 + 0 + + + 2.0 + Sarthi, P. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + f879674860784f9eb4289aeb91728351 + 1016 + 0 + + + 2.0 + Sarthi, P. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + dea50b1765e54936b3d0b1e499ab2053 + 1017 + 0 + + + 2.0 + Abdullah, S. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 26634913d18f4629b39dffa19c1df734 + 1018 + 0 + + + 2.0 + Abdullah, S. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + a69bde74fc9d41cfa669f148c7c43dd8 + 1019 + 0 + + + 2.0 + Abdullah, S. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + e65667ec99e145fea2055d6b583cb05b + 1020 + 0 + + + 2.0 + Abdullah, S. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + cdfcbba5664d42508cd34df9af42b0dc + 1021 + 0 + + + 2.0 + Tuli, A. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 570de818eba04615a6afb3a573e82ff1 + 1022 + 0 + + + 2.0 + Tuli, A. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 473533c454d34975a17a0193e39e0bac + 1023 + 0 + + + 2.0 + Tuli, A. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + e22af264c702440f93070465f45e630e + 1024 + 0 + + + 1.0 + Yang, Z. and Manning, C. D. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + ad95fbd957ae4c22859f58446dd8c9cc + 1025 + 0 + + + 1.0 + Huang, M. and Duan, N. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + 37d42129ca4d49dea240f66d1fdd4b78 + 1026 + 0 + + + 1.0 + Su, D. and Xu, Y. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1f40481f4ee342d4be51d33ffafc17d1 + 1027 + 0 + + + 1.0 + Su, D. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + e892d46e07e44bd5a2d1626875cc024f + 1028 + 0 + + + 1.0 + Su, D. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + fcb033a54d734ce5a87e0d8ad555867a + 1029 + 0 + + + 1.0 + Su, D. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 964fc01bfd9a400eb668761539dc9d9f + 1030 + 0 + + + 1.0 + Su, D. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 86b0d920fa504eba81c26cfc3f4d2b9f + 1031 + 0 + + + 1.0 + Xu, Y. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + e0221df7b7e44dd7956c8d0348d46b6d + 1032 + 0 + + + 1.0 + Xu, Y. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + a20d7f4fee104273b9628d648c05a5ac + 1033 + 0 + + + 1.0 + Xu, Y. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + b352398c5b1742d8a61acd8534ef0f53 + 1034 + 0 + + + 1.0 + Xu, Y. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 38b778af3c3f4be2a23e3932c94390c3 + 1035 + 0 + + + 1.0 + Yu, T. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 35d528e52a6441a58e58385d85bfae4b + 1036 + 0 + + + 1.0 + Yu, T. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 06d0d34ab3d043c689044a0fbfc65e10 + 1037 + 0 + + + 1.0 + Yu, T. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9e0ec036c91e44daa8e1a2af50df2081 + 1038 + 0 + + + 1.0 + Siddique, F. B. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 580b2395e68442539a606d37ddba691d + 1039 + 0 + + + 1.0 + Siddique, F. B. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 370fd1c6563045499e1d711fcd9ef9d5 + 1040 + 0 + + + 1.0 + Barezi, E. J. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 3b6c4319026844ecb645f650e30b7d1a + 1041 + 0 + + + 1.0 + Tang, Y. and Yang, Y. co-authored the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + e044334b7d2e426ca2cab7eb763d8bc9 + 1042 + 0 + + + 1.0 + Touvron, H. and Martin, L. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 4a2fa382e77946d2be8e95edc04c6a64 + 1043 + 0 + + + 1.0 + Touvron, H. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + cd015281a069460e844faeb327b7d65f + 1044 + 0 + + + 1.0 + Touvron, H. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + c7a9fbf22a054056bf4f4562eaecfc08 + 1045 + 0 + + + 1.0 + Touvron, H. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 44bf341f78d74c4bb15ae209649d0ca9 + 1046 + 0 + + + 1.0 + Touvron, H. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 045ff6885d424b4caeabc76c50468c7c + 1047 + 0 + + + 1.0 + Touvron, H. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 543a52396f0b4f7f99ea755fba11d290 + 1048 + 0 + + + 1.0 + Touvron, H. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 33e487870db646e5b3d9c1f2962a7c6a + 1049 + 0 + + + 1.0 + Touvron, H. and Bhargava, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + b5b628d809514bfe9bbb3bd362815e79 + 1050 + 0 + + + 1.0 + Touvron, H. and Bhosale, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9d9ae51d2af44ebe8324dd2dd1dcd83b + 1051 + 0 + + + 1.0 + Martin, L. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + b7c606aa6ad1416e9f934628acce5f24 + 1052 + 0 + + + 1.0 + Martin, L. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 4521342f29774fab85e6acb0490d46e5 + 1053 + 0 + + + 1.0 + Martin, L. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + e7ac741e4aa4433ca5f2379726f90b33 + 1054 + 0 + + + 1.0 + Martin, L. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 43645eb9258244a8bd334ce77216b1c0 + 1055 + 0 + + + 1.0 + Martin, L. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0c9c52488ad647abbaf2b4589c976957 + 1056 + 0 + + + 1.0 + Martin, L. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0e8fb49a531e4ea48fece73957bd8a54 + 1057 + 0 + + + 1.0 + Wang, J. and Liang, Y. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 6187217c38ca4225b97d04d9644dcdf0 + 1058 + 0 + + + 1.0 + Wang, J. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8c8330abe276487294eba3a341ee9e0c + 1059 + 0 + + + 1.0 + Wang, J. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 200df54d93964e81ae2dcf727bffb23c + 1060 + 0 + + + 1.0 + Wang, J. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 4d0478c05f614675b336a76a0c088b3e + 1061 + 0 + + + 1.0 + Wang, J. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + c0dc4d618b5e49f7a18efa34dbf450ac + 1062 + 0 + + + 1.0 + Wang, J. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + dec05f79120940b78cd921a0a67f1540 + 1063 + 0 + + + 1.0 + Wang, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8edbf3e4f0d94f6ab78127c61bf87b76 + 1064 + 0 + + + 1.0 + Wang, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1d722426930a42eeadfa624a6eb2408f + 1065 + 0 + + + 1.0 + Liang, Y. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 6d22f2009f6a4df9a242f03e2642981e + 1066 + 0 + + + 1.0 + Liang, Y. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + db5035c544214c72987eed4d4d9e327f + 1067 + 0 + + + 1.0 + Liang, Y. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + fd00e337b5c4465cbcbdf07bc294a3a8 + 1068 + 0 + + + 1.0 + Liang, Y. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + ea1e51b558c149649711a29157f4e604 + 1069 + 0 + + + 1.0 + Liang, Y. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + b888ad14e84347f8831a7dd2cea294fd + 1070 + 0 + + + 1.0 + Liang, Y. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0e2323c0256d40579e7526dbdd019a8d + 1071 + 0 + + + 1.0 + Liang, Y. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + deb3bc5537a14352b22a0a473a59d8c7 + 1072 + 0 + + + 1.0 + Meng, F. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + ba445c400c8e405bb646387eab98a62b + 1073 + 0 + + + 1.0 + Meng, F. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 95a90d05e82d44ada6f8577ca49dd491 + 1074 + 0 + + + 1.0 + Meng, F. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + fa15140701a54689835604665d187c54 + 1075 + 0 + + + 1.0 + Meng, F. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8346468f7f4c46bebe1eaafd9753d55f + 1076 + 0 + + + 1.0 + Meng, F. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 46ddbfe32d444912b423dd1769fbaa43 + 1077 + 0 + + + 1.0 + Meng, F. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + ffce5a64e9394d1399319588d7fd4e3e + 1078 + 0 + + + 1.0 + Sun, Z. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + e934202aa3b344ba9fef89ecb42530b4 + 1079 + 0 + + + 1.0 + Sun, Z. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 22d5ff62439047ccaeaa63fd8a30f3e5 + 1080 + 0 + + + 1.0 + Sun, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9b2f77b408ec4147bd5dd67a01d9f439 + 1081 + 0 + + + 1.0 + Sun, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 96b1264e89394adfaf026471e3b6ad47 + 1082 + 0 + + + 1.0 + Sun, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 00741dfb8f6d477f913d20406dfcd65d + 1083 + 0 + + + 1.0 + Shi, H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 3eb344aa8b05448984dacac93482ebc4 + 1084 + 0 + + + 1.0 + Shi, H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8db8a8680d534161b0772d7a771df6bd + 1085 + 0 + + + 1.0 + Shi, H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 7f2c628b4fa54c0b9254049602ed20d2 + 1086 + 0 + + + 1.0 + Shi, H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2f8f5d33916d4824bec6773bacd37d87 + 1087 + 0 + + + 2.0 + Li, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + b8e1f95f9d3e497393d86e6bd137fe10 + 1088 + 0 + + + 2.0 + Li, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 13a41e2ff8b847ee8073e1e23b0bffc6 + 1089 + 0 + + + 2.0 + Li, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 9c61fb5ee44744a48bc5638bd42f654b + 1090 + 0 + + + 1.0 + H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 946213d345b64cbaa6becb8723b01d87 + 1091 + 0 + + + 1.0 + Zheng, L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e75c65762f064dfc95787fa331c95392 + 1092 + 0 + + + 1.0 + Chiang, W.-L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 9a586c1629464133920fb19d8bd1e690 + 1093 + 0 + + + 1.0 + Sheng, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e7a82e12e4f84f3e82c1ec74d3088235 + 1094 + 0 + + + 1.0 + Zhuang, S. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 51828127e71d40829039e033add265c4 + 1095 + 0 + + + 1.0 + Wu, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 04918b80bc714753b00af559d439a4ec + 1096 + 0 + + + 1.0 + Zhuang, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 3b4dbcb1c7c24bf8b6d55485c0304f7e + 1097 + 0 + + + 1.0 + Lin, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + fdd2a43d9b9f450c899adfb60b05e711 + 1098 + 0 + + + 1.0 + Li, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + bced585ff9d54fb7acd03f54f5729391 + 1099 + 0 + + + 1.0 + Li, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 50aec048280a4cdb8572993faab794dd + 1100 + 0 + + + 1.0 + Li, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + e128a7179b6e476c98d6bbfecf2a3f9a + 1101 + 0 + + + 2.0 + Xu, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + a3fa60c3370e4d5e8147250e2a18104a + 1102 + 0 + + + 2.0 + Xu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 1a7ef91522514b9f8b1ddaf68424351d + 1103 + 0 + + + 1.0 + H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + e5f094bf02d84a0889cd041199156ad7 + 1104 + 0 + + + 2.0 + Qu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 5c64fc0a74044110906120ca1d5c7919 + 1105 + 0 + + + 1.0 + H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 524d1b6a01d34b0098a0da8af056bfc8 + 1106 + 0 + + + 1.0 + H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 73af37cadd3c4d3dbfb8bfd697aeef58 + 1107 + 0 + + + 1.0 + Wang, S. and Khramtsova, E. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + c6b26fcc94044c368b2fe0db4b9b72f2 + 1108 + 0 + + + 1.0 + Wang, S. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 36b9f94e17c6481fb83670b70b192eb7 + 1109 + 0 + + + 1.0 + Wang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 24c63641b4e241589336236d5f916e34 + 1110 + 0 + + + 1.0 + Khramtsova, E. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 718e03207dcd44a080806880d08268ea + 1111 + 0 + + + 1.0 + Khramtsova, E. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 6d0efadfba5046eb86869827544c2703 + 1112 + 0 + + + 1.0 + Zhuang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 2c33b12183ad4722ab1ab2cbd75f8312 + 1113 + 0 + + + 1.0 + Zheng, L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + f334bc9701204b1b943f9ece317ca68a + 1114 + 0 + + + 1.0 + Chiang, W.-L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 4ea3ecb74c91452da866f4c9163386e2 + 1115 + 0 + + + 1.0 + Sheng, Y. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + a12fe8fd9bc34db69d8de6944283d3c9 + 1116 + 0 + + + 1.0 + Zhuang, S. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 02b877f830cc4bb58dfad02f13a6d6ce + 1117 + 0 + + + 1.0 + Zhuang, S. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + abffcf93dc114332a181990ad56b7863 + 1118 + 0 + + + 1.0 + Zhuang, S. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + f894b0ae91eb412d93b6b06d4a73f350 + 1119 + 0 + + + 1.0 + Zhuang, S. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 75e06eb1f93c4ee38b782965ea905b5b + 1120 + 0 + + + 1.0 + Zhuang, S. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 70fd5c73bbe34d918b3dca3fc7294d28 + 1121 + 0 + + + 1.0 + Zhuang, S. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 8f1edc1d00764d6fb23859242c659deb + 1122 + 0 + + + 1.0 + Wang, Y. and Lipka, N. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + c8554314efb44679a898bbce08372abe + 1123 + 0 + + + 1.0 + Wang, Y. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + a41dbc4319f74bc995a93dbe0f4d9aee + 1124 + 0 + + + 1.0 + Wang, Y. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + ef76a41bf9bf45c893c475a7bd5a2938 + 1125 + 0 + + + 1.0 + Wang, Y. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 88f486cbb1904425a5fd5dfa268cf85d + 1126 + 0 + + + 1.0 + Wang, Y. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + ec10a8695b1a4e8787d9d29114e9d5ce + 1127 + 0 + + + 1.0 + Lipka, N. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 9397185bb4d7492b88eaa20fa10c0ae5 + 1128 + 0 + + + 1.0 + Lipka, N. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 2b1b9b1ed49c4ace91ff099752b8c0a5 + 1129 + 0 + + + 1.0 + Lipka, N. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 62c66c25992d4974829678313ed60b1d + 1130 + 0 + + + 1.0 + Lipka, N. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 26a889667b614ab890d863c4b8762e69 + 1131 + 0 + + + 1.0 + Rossi, R. A. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 30461855b0604128a4f10d0b348ce60f + 1132 + 0 + + + 1.0 + Rossi, R. A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + edef06de541f493f98d9281a704d785d + 1133 + 0 + + + 1.0 + Rossi, R. A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 0016a9dec22543e9b203f540860bf2e7 + 1134 + 0 + + + 1.0 + Siu, A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4f79558a259f4de58df5b022b68a459e + 1135 + 0 + + + 1.0 + Siu, A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + ca07919df74f4e5abfbd370c50eacc00 + 1136 + 0 + + + 1.0 + Zhang, R. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 94b8715919cd49d08ac0ce99b930ea53 + 1137 + 0 + + + 1.0 + Yang, Z. and Qi, P. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4a10c341918b4d888b8b0466bd101b1d + 1138 + 0 + + + 1.0 + Yang, Z. and Zhang, S. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 5173ce4188034717b9c90eef40b94932 + 1139 + 0 + + + 1.0 + Yang, Z. and Bengio, Y. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + cff3415902bf4745992473697570aef0 + 1140 + 0 + + + 1.0 + Yang, Z. and Cohen, W. W. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4c0cf727ec2843a288aa00b43f25b2de + 1141 + 0 + + + 1.0 + Yang, Z. and Salakhutdinov, R. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 87ece3a4dcc84c98a291c1138ae56544 + 1142 + 0 + + + 1.0 + Zheng, L. and Chiang, W.-L. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 143dc5f4cb4b4596900ee5158594b1b0 + 1143 + 0 + + + 1.0 + Zheng, L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 3c222c5fdfab46e1ac1352a0f85a5fdd + 1144 + 0 + + + 1.0 + Zheng, L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 8c13a465b483417691c9b8d40b913da3 + 1145 + 0 + + + 1.0 + Zheng, L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 51633d2f0adf4123a23eeb292d95e649 + 1146 + 0 + + + 1.0 + Zheng, L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 849136ae8c9f4f9589a989bfe4c4155d + 1147 + 0 + + + 1.0 + Zheng, L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 84b11b37d1dd4e75b4c453669fbd4df9 + 1148 + 0 + + + 1.0 + Zheng, L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 373c198a0ed2402cb885b8d9f9de92f3 + 1149 + 0 + + + 1.0 + Zheng, L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 64ec8c4eb0734d60a5287e3df62652bd + 1150 + 0 + + + 1.0 + Chiang, W.-L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 7636104f26794a4e9e74b2d6943c879d + 1151 + 0 + + + 1.0 + Chiang, W.-L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + fc4b406a34ea4b2d9f305600aab14ea3 + 1152 + 0 + + + 1.0 + Chiang, W.-L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 5400473bab9d4105a1517fdc55c58f17 + 1153 + 0 + + + 1.0 + Chiang, W.-L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 624a1e6ea1d248f8b5126527e82e76c0 + 1154 + 0 + + + 1.0 + Chiang, W.-L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 66e53a4f6fc740aaaa379aa63d15f0e9 + 1155 + 0 + + + 1.0 + Chiang, W.-L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e676d0167a3f43478a209ec9526c90df + 1156 + 0 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 44fa3d2247904198b1c776e060d35eb2 + 1157 + 0 + + + 1.0 + Sheng, Y. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + a135859c7d3d4d3596f1e4ab218eff8a + 1158 + 0 + + + 1.0 + Sheng, Y. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 5e6fd98384a24a34b80311842661e00a + 1159 + 0 + + + 1.0 + Sheng, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 3a10d5261d4240c7b05b6cdb7838ff24 + 1160 + 0 + + + 1.0 + Sheng, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + ec595c7b07e148dba900040a68ef0fdb + 1161 + 0 + + + 1.0 + Sheng, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 4b43619c5c6a4ea3826bfd3c06aa6e66 + 1162 + 0 + + + 1.0 + Sheng, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 3173671571f14c75bfb9141754424efa + 1163 + 0 + + + 1.0 + Wu, Z. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 1675e75d7b524d4ab236eeaefd2dc992 + 1164 + 0 + + + 1.0 + Wu, Z. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + bb906c8e869141aa9be12118dcd3d3b5 + 1165 + 0 + + + 1.0 + Wu, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 27c95f7d6c3d4732897ae7bffd7c5dc8 + 1166 + 0 + + + 1.0 + Wu, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + b63c467faf714acd8a006431faf7a141 + 1167 + 0 + + + 1.0 + Wu, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 19a5840a67e14c468f9f3d6851eaee5c + 1168 + 0 + + + 1.0 + Zhuang, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 7a58673144d849e7a784caee9d9d4e99 + 1169 + 0 + + + 1.0 + Zhuang, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 8076db94b7214fdf9e006ce5a7e1cbe2 + 1170 + 0 + + + 1.0 + Zhuang, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e733886404db4d46862bdddb2aee5211 + 1171 + 0 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + b75594a9d9c8404688a5cfe02272cdfc + 1172 + 0 + + + 1.0 + Lin, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + f253ff2311024729a758bb77b14bf72d + 1173 + 0 + + + 1.0 + Lin, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 1f0cb1e7fae64c238efb659d254d6221 + 1174 + 0 + + + 1.0 + Lin, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 6740be36e0e14774a5551a17db648a13 + 1175 + 0 + + + 1.0 + Li, D. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 0f926e9dfaae4615b16a794e984b85ae + 1176 + 0 + + + 1.0 + Li, D. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 47851446a4df4f5aa4505c999daaaaf7 + 1177 + 0 + + + 1.0 + Xing, E. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + c3e51a2782ba4f86b49b4038a316d9fb + 1178 + 0 + + + \ No newline at end of file diff --git a/graphfleet/output/graphindex/artifacts/clustered_graph.1.graphml b/graphfleet/output/graphindex/artifacts/clustered_graph.1.graphml new file mode 100644 index 000000000..bd8f59e69 --- /dev/null +++ b/graphfleet/output/graphindex/artifacts/clustered_graph.1.graphml @@ -0,0 +1,16186 @@ + + + + + + + + + + + + + + + + + + + PERSON + Darren Edge is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 0 + b45241d70f0e43fca764df95b2b81f77 + + + PERSON + Ha Trinh is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 1 + 4119fd06010c494caa07f439b333f4c5 + + + PERSON + Newman Cheng is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 2 + d3835bf3dda84ead99deadbeac5d0d7d + + + PERSON + Joshua Bradley is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 3 + 077d2820ae1845bcbb1803379a3d1eae + + + PERSON + Alex Chao is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 4 + 3671ea0dd4e84c1a9b02c5ab2c8f4bac + + + PERSON + Apurva Mody is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 5 + 19a7f254a5d64566ab5cc15472df02de + + + PERSON + Steven Truitt is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 6 + e7ffaee9d31d4d3c96e04f911d0a8f9e + + + PERSON + Jonathan Larson is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 7 + f7e11b0e297a44a896dc67928368f600 + + + ORGANIZATION + Microsoft Research is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + 3 + 8 + 1fd3fa8bb5a2408790042ab9573779ee + + + ORGANIZATION + Microsoft Strategic Missions and Technologies is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + 3 + 9 + 27f9fbe6ad8c4a8b9acee0d3596ed57c + + + ORGANIZATION + Microsoft Office of the CTO is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + 2 + 10 + e1fd0e904a53409aada44442f23a51cb + + + METHOD + RAG (Retrieval-Augmented Generation) is a developing research area with multiple established directions, including knowledge graph creation, completion, and extraction of causal graphs. It is a method used for generating responses in text generation tasks by retrieving relevant information from an external knowledge source to enable large language models to answer questions. This approach incorporates the retrieval of relevant data to augment text generation, producing direct responses in various text generation tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 26 + 1 + 31 + 11 + de988724cfdf45cebfba3b13c43ceede + + + TECHNOLOGY + LLM (Large Language Model) is a type of artificial intelligence model used for a variety of tasks in the field of Natural Language Processing and Information Retrieval. These tasks include generating and assessing text, entity extraction, summarization, understanding relationships in text, and automating human-like sensemaking and reasoning over large collections of documents. LLMs are also employed to generate intermediate answers and scores for text chunks, process these chunks to extract elements of a graph index, and automate the generation of questions for dataset evaluation. Additionally, LLMs can analyze and generate text based on retrieved information and queries, and they possess a context window that can be exceeded by external datasets. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,2c6ed90897310eea2f28e33fff1c32b0,6f33a085ff3304e5994f7fbb86c881a4,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 25 + 1 + 32 + 12 + 96aad7cb4b7d40e9b7e13b94a67af206 + + + METHOD + Graph RAG (Retrieval-Augmented Generation) is a sophisticated method that leverages the natural modularity of graphs to partition data for global summarization tasks. This approach integrates multiple stages and concepts, including knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS), to support human sensemaking over text corpora. It is particularly effective in providing comprehensive and structured overviews of public figures across various sectors of the entertainment industry, as well as generating answers for questions in the News article dataset. + +Graph RAG employs a high-level data flow and pipeline for processing and summarizing text, combining both global and local approaches to optimize token usage in text generation tasks. It uses community summaries to improve answer comprehensiveness and diversity while requiring fewer tokens compared to traditional source text summarization methods. This method has been shown to outperform naive RAG in terms of comprehensiveness and diversity in text generation tasks. + +A specific implementation of Graph RAG involves using four levels of graph communities, incorporating concepts from other systems such as self-memory and parallel generation of community answers. This multi-stage mechanism allows for the comparison of multiple conditions, enhancing the overall effectiveness of the summarization process. + +Graph RAG, launched by NebulaGraph, is a retrieval-augmented generation technology based on knowledge graphs. It combines global summarization with graph-based text indexing to answer questions over private text corpora, making it a versatile tool for various text analysis and summarization applications. + 086021a89900a39bcb62036981737bfa,21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,718017a4871c909420f84b85b8ba969d,833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19,fb3c48579608fa28be585ceb6cd2f0fe + 50 + 1 + 90 + 13 + c9632a35146940c2a86167c7726d35e9 + + + METHOD + QFS (Query-Focused Summarization) is a method used to generate summaries based on specific user queries + e8d83e6e7a7c0f57b218cef24976b745 + 50 + 1 + 1 + 14 + 9646481f66ce4fd2b08c2eddda42fc82 + + + CONCEPT + Community summaries are generated summaries of data clusters or communities, used to answer queries and tailored to the domain. These summaries are pre-generated for groups of closely-related entities, particularly in the Graph RAG approach, and are derived from community-generated content to compare with source texts. They are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks. Community summaries provide report-like insights into each community within a hierarchical structure, which is useful for understanding the dataset. They are generated from modular communities in the knowledge graph and cover different levels of each graph community hierarchy, including root-level communities in an entity-based graph index. Additionally, these summaries act as a kind of self-memory for generation-augmented retrieval. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + 66 + 1 + 26 + 15 + d91a266f766b4737a06b0fda588ba40b + + + CONCEPT + Global sensemaking questions are questions that require understanding and summarizing large datasets, often beyond the explicit content of the source texts + e8d83e6e7a7c0f57b218cef24976b745 + 3 + 16 + bc0e3f075a4c4ebbb7c7b152b65a5625 + + + METRIC + 1 million token range refers to the scale of datasets used in the evaluation of the Graph RAG approach + e8d83e6e7a7c0f57b218cef24976b745 + 1 + 17 + 254770028d7a4fa9877da4ba0ad5ad21 + + + TECHNOLOGY + Python is a programming language used for implementing both global and local Graph RAG approaches. Additionally, Python is utilized to implement the open-source version of the Graph RAG approach. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + 50 + 1 + 1 + 18 + 4a67211867e5464ba45126315a122a8a + + + URL + The URL "HTTPS://AKA.MS/GRAPHRAG" is the location where the open-source, Python-based implementation of Graph RAG approaches will be available. This URL serves as the repository for accessing the open-source implementation of the Graph RAG approach. + e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745 + 2 + 19 + 04dbbb2283b845baaeac0eaf0c34c9da + + + METHOD + Query-Focused Summarization (QFS) is a method used to generate summaries that are relevant to specific user queries. This summarization technique focuses on answering specific queries by utilizing the entire corpus of information available. It is designed to provide concise and relevant information based on the specific needs of the user, ensuring that the generated summaries are directly aligned with the queries posed. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + 70 + 1 + 5 + 20 + 1943f245ee4243bdbfbd2fd619ae824a + + + CONCEPT + An external knowledge source is a repository of information that can be accessed to retrieve relevant data for answering questions + e8d83e6e7a7c0f57b218cef24976b745 + 26 + 1 + 1 + 21 + 273daeec8cad41e6b3e450447db58ee7 + + + CONCEPT + A text corpus is a large collection of written texts used for analysis and research + e8d83e6e7a7c0f57b218cef24976b745 + 1 + 22 + e69dc259edb944ea9ea41264b9fcfe59 + + + CONCEPT + An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + 60 + 1 + 2 + 23 + e2f5735c7d714423a2c4f61ca2644626 + + + CONCEPT + Source documents are the original texts from which information is extracted, retrieved, or summarized. These documents serve as the foundational input texts for various processing tasks, including the Graph RAG (Retrieval-Augmented Generation) approach. In this context, source documents are critical for extracting and analyzing information, ensuring that the data used in computational linguistics and information retrieval tasks is accurate and comprehensive. + bc9e2c9e369c4108cf4f6dd5f60960f4,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + 60 + 1 + 5 + 24 + deece7e64b2a4628850d4bb6e394a9c3 + + + CONCEPT + A partial response is an intermediate answer generated from community summaries before being combined into a final response + e8d83e6e7a7c0f57b218cef24976b745 + 66 + 1 + 2 + 25 + e657b5121ff8456b9a610cfaead8e0cb + + + CONCEPT + A final response is the comprehensive answer generated after combining all partial responses + e8d83e6e7a7c0f57b218cef24976b745 + 66 + 1 + 1 + 26 + bf4e255cdac94ccc83a56435a5e4b075 + + + METRIC + COMPREHENSIVENESS is a metric used to evaluate the quality of generated responses by measuring how much detail an answer provides to cover all aspects and details of a question. It assesses the completeness and thoroughness of answers, ensuring that they encompass all relevant information. This metric is particularly important in evaluating the summarization approach, focusing on the completeness of the summary. In practical applications, such as evaluating Podcast transcripts and News articles, comprehensiveness has shown win rates between 72-83% and 72-80%, respectively. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + 53 + 1 + 9 + 27 + 3b040bcc19f14e04880ae52881a89c1c + + + METRIC + DIVERSITY is a metric used to evaluate the variety and richness of answers generated in response to a question. It measures how varied and rich an answer is in providing different perspectives and insights. This metric is particularly important in assessing the quality of summarization approaches, focusing on the variety of information included in the summary. DIVERSITY is applied to various types of content, including Podcast transcripts, where win rates range from 75-82%, and News articles, with win rates ranging from 62-71%. It is a crucial target quality for evaluating the effectiveness of different methods in generating diverse and informative responses. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + 50 + 1 + 7 + 28 + 3d6b216c14354332b1bf1927ba168986 + + + ACTIVITY + Human endeavors refer to activities and efforts across various domains that rely on reading and reasoning about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + 37 + 1 + 2 + 29 + 1c109cfdc370463eb6d537e5b7b382fb + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models designed to understand and generate human-like text, playing a crucial role in automating sensemaking in complex domains. Modern language models, such as GPT, Llama, and Gemini, leverage in-context learning to effectively summarize content. These models are integral to the field of Natural Language Processing and Information Retrieval, enabling sophisticated text analysis and generation capabilities. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + 36 + 1 + 11 + 30 + 3d0dcbc8971b415ea18065edc4d8c8ef + + + DOMAIN + Scientific discovery is a complex domain where sensemaking is applied to understand and generate new knowledge from scientific texts + f0306814bf64f5c9e79603fc6a52f4ea + 37 + 1 + 1 + 31 + 68105770b523412388424d984e711917 + + + DOMAIN + Intelligence analysis is a complex domain where sensemaking is applied to understand and generate insights from intelligence data + f0306814bf64f5c9e79603fc6a52f4ea + 37 + 1 + 1 + 32 + 85c79fd84f5e4f918471c386852204c5 + + + PROCESS + SENSEMAKING is the process of understanding and making sense of complex information. It involves understanding connections among people, places, and events to anticipate their trajectories and act effectively. This process is crucial for navigating and interpreting intricate data landscapes, enabling individuals and organizations to make informed decisions based on the relationships and patterns identified within the information. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 37 + 1 + 7 + 33 + eae4259b19a741ab9f9f6af18c4a0470 + + + DATA + TEXT CHUNKS are segments of text that are embedded into a vector space for analysis. These segments are extracted from source documents and are used for processing in the Graph RAG (Retrieval-Augmented Generation) approach. By embedding these text chunks into a vector space, they can be analyzed more effectively, facilitating various natural language processing and information retrieval tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + 60 + 1 + 5 + 34 + 3138f39f2bcd43a69e0697cd3b05bc4d + + + DATA + Element instances are identified and extracted instances of graph nodes and edges from text chunks. They represent individual occurrences of entities, relationships, and claims extracted from source texts. These specific pieces of information are tailored to the domain, providing a structured representation of the underlying data. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 59 + 1 + 4 + 35 + dde131ab575d44dbb55289a6972be18f + + + DATA + Element summaries are concise representations of element instances, tailored to the domain. They are descriptive texts created by the LLM to summarize entities, relationships, and claims extracted from source texts. These summaries provide detailed descriptions of nodes, edges, and covariates within a community, and are used to understand the structure and semantics of the dataset. In essence, element summaries serve as a tool to encapsulate and convey the intricate details of elements within a graph, facilitating a deeper comprehension of the dataset's structural dynamics and semantic relationships. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 59 + 1 + 10 + 36 + de9e343f2e334d88a8ac7f8813a915e5 + + + DATA + Graph communities are groups of elements, including nodes, edges, and covariates, detected within a graph index, primarily used for summarization. These communities consist of groups of nodes that exhibit stronger connections to each other than to nodes outside the group. This structural characteristic allows for the identification and analysis of densely connected subgraphs, which can be crucial for understanding the underlying relationships and dynamics within complex networks. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 68 + 1 + 3 + 37 + e2bf260115514fb3b252fd879fb3e7be + + + DATA + COMMUNITY ANSWERS are query-focused summaries derived from community summaries. These answers are generated in parallel from the community summaries and are designed to respond to user queries effectively. Essentially, COMMUNITY ANSWERS serve as responses that synthesize information from community summaries to provide concise and relevant answers to specific questions posed by users. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + 66 + 1 + 3 + 38 + b462b94ce47a4b8c8fffa33f7242acec + + + DATA + GLOBAL ANSWER is a comprehensive response generated from multiple community summaries to answer a user query. It is the final query-focused summary produced from all relevant community summaries. The final answer is generated by combining intermediate community answers based on their helpfulness scores. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 64 + 1 + 7 + 39 + 17ed1d92075643579a712cc6c29e8ddb + + + TIME + Indexing time refers to the time when the graph index is created and elements are summarized + f0306814bf64f5c9e79603fc6a52f4ea + 63 + 1 + 1 + 40 + 3ce7c210a21b4deebad7cc9308148d86 + + + TIME + Query time refers to the time when a query is made and the relevant summaries are generated + f0306814bf64f5c9e79603fc6a52f4ea + 63 + 1 + 1 + 41 + d64ed762ea924caa95c8d06f072a9a96 + + + PROCESS + Graph RAG pipeline is a process using an LLM-derived graph index to detect, extract, and summarize nodes, edges, and covariates in source documents + f0306814bf64f5c9e79603fc6a52f4ea + 63 + 1 + 7 + 42 + adf4ee3fbe9b4d0381044838c4f889c8 + + + DATA + NODES are entities detected in the graph index of source documents. They represent the individual elements or points in a graph. For instance, in the Podcast dataset, there are 8,564 nodes, while the News dataset contains 15,754 nodes. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + 21 + 1 + 3 + 43 + 32ee140946e5461f9275db664dc541a5 + + + DATA + EDGES are relationships detected in the graph index of source documents. They represent the connections or links between nodes in a graph. For instance, in the Podcast dataset, there are 20,691 edges, while the News dataset contains 19,520 edges. These edges are crucial for understanding the structural dynamics and relationships within the datasets, providing insights into how different nodes (such as topics, entities, or documents) are interconnected. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + 21 + 1 + 3 + 44 + c160b9cb27d6408ba6ab20214a2f3f81 + + + DATA + Covariates are additional attributes associated with extracted node instances in the graph index. They represent claims or additional information detected in the graph index of source documents. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 63 + 1 + 2 + 45 + 23527cd679ff4d5a988d52e7cd056078 + + + METHOD + LEIDEN is a community detection algorithm renowned for its efficiency in recovering hierarchical community structures. It is widely used to partition graphs into modular communities, effectively grouping elements within a graph index. The algorithm's ability to identify and organize these communities makes it a valuable tool in the analysis of complex networks, particularly within the domains of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 61 + 1 + 7 + 46 + f1c6eed066f24cbdb376b910fce29ed4 + + + METHOD + Retrieval-Augmented Generation (RAG) is an established approach in the field of Natural Language Processing and Information Retrieval, designed to answer user questions over entire datasets. This method involves retrieving relevant text regions to provide grounding for the generation task, thereby enhancing the accuracy and relevance of the generated responses. By combining retrieval and generation processes, RAG effectively synthesizes and presents pertinent information, making it a powerful tool for handling complex queries and large datasets. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + 63 + 1 + 4 + 47 + 83a6cb03df6b41d8ad6ee5f6fef5f024 + + + ORGANIZATION + Microsoft is a technology company whose Chief Technology Officer, Kevin Scott, actively participates in podcast conversations. The organization is deeply involved in automating sensemaking in scientific discovery through the use of large language models (LLMs). Notably, Microsoft conducted a study examining the impact of large language models, specifically GPT-4, on scientific discovery. + 1d07b4248c2655081c7af0e373bd70c9,833e7d67dcd30790b26b71c9b5306f6b,f0306814bf64f5c9e79603fc6a52f4ea + 38 + 1 + 3 + 48 + 147c038aef3e4422acbbc5f7938c4ab8 + + + PERSON + Ranade is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + 36 + 1 + 1 + 49 + b7702b90c7f24190b864e8c6e64612a5 + + + PERSON + Joshi is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + 36 + 1 + 1 + 50 + de6fa24480894518ab3cbcb66f739266 + + + PERSON + Klein is an author who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + 37 + 1 + 1 + 51 + 6fae5ee1a831468aa585a1ea09095998 + + + PERSON + Lewis is an author who contributed to the development of the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + 63 + 1 + 1 + 52 + ef32c4b208d041cc856f6837915dc1b0 + + + PERSON + Traag is an author who contributed to the development of the Leiden community detection method + f0306814bf64f5c9e79603fc6a52f4ea + 61 + 1 + 1 + 53 + 07b2425216bd4f0aa4e079827cb48ef5 + + + PUBLICATION + arXiv is a preprint repository where several significant papers in the field of Natural Language Processing and Information Retrieval have been published. It serves as a platform for electronic preprints (known as e-prints) that are approved for publication after moderation, but not full peer review. Notable papers published on arXiv include "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models," "Lost in the middle: How language models use long contexts," "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," "Llama 2: Open foundation and fine-tuned chat models," "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy," "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries," "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions," "Enhancing knowledge graph construction using large language models," "Is chatgpt a good nlg evaluator? a preliminary study," "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt," "Causal graph discovery with retrieval-augmented generation based large language models," "Knowledge graph prompting for multi-document question answering," "Text summarization with latent queries," "Retrieval-augmented generation for large language models: A survey," and "Knowledge graph-augmented language models for knowledge-grounded dialogue generation." This repository is a crucial resource for researchers to disseminate their findings rapidly and access the latest advancements in their fields. + 00e8e4e881bd0862022f4dfc913b900b,086021a89900a39bcb62036981737bfa,58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035,f0306814bf64f5c9e79603fc6a52f4ea,fc4b27d64f055b7fc30176ba110dd02e + 45 + 1 + 39 + 54 + 2670deebfa3f4d69bb82c28ab250a209 + + + PUBLICATION + Preprint refers to the version of the research paper that is under review and available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 41 + 1 + 4 + 55 + 404309e89a5241d6bff42c05a45df206 + + + CATEGORY + cs.CL is the category under which the research paper is classified on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 41 + 1 + 1 + 56 + b785a9025069417f94950ad231bb1441 + + + DATE + 24 Apr 2024 is the date when the research paper was submitted to arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 41 + 1 + 1 + 57 + 3b6cd96a27304614850709aba1c9598b + + + IDENTIFIER + 2404.16130v1 is the identifier for the research paper on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 41 + 1 + 1 + 58 + d54956b79dd147f894b67a8b97dcbef0 + + + DATA + Document collections refer to large sets of documents that are analyzed for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + 37 + 1 + 1 + 59 + 958beecdb5bb4060948415ffd75d2b03 + + + TECHNOLOGY + LLM PROMPTS are specific instructions given to large language models (LLMs) to tailor their responses to the domain of the dataset. These prompts are also used to extract elements from text chunks, ensuring that the LLMs provide relevant and precise information based on the given context. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 36 + 1 + 2 + 60 + b999ed77e19e4f85b7f1ae79af5c002a + + + METHOD + Community detection is a method used to identify groups of related elements within a graph. It involves the process of identifying communities within a graph, which are clusters of nodes that are more densely connected internally than with the rest of the network. This technique is crucial in understanding the structural dynamics and relationships within complex networks, such as those found in social networks, biological systems, and information retrieval systems. By uncovering these communities, researchers can gain insights into the underlying structure and function of the network, facilitating more effective analysis and interpretation of the data. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 68 + 1 + 3 + 61 + 48c0c4d72da74ff5bb926fa0c856d1a7 + + + METHOD + Domain-tailored summarization is a method used to create summaries that are specific to the domain of the dataset + f0306814bf64f5c9e79603fc6a52f4ea + 59 + 1 + 2 + 62 + 4f3c97517f794ebfb49c4c6315f9cf23 + + + PERSON + Klein et al. are authors who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + 37 + 1 + 1 + 63 + 1745a2485a9443bab76587ad650e9be0 + + + PERSON + Ranade and Joshi are authors who discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + 36 + 1 + 1 + 64 + 32e6ccab20d94029811127dbbe424c64 + + + PERSON + Lewis et al. are authors who developed the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + 63 + 1 + 1 + 65 + 94a964c6992945ebb3833dfdfdc8d655 + + + PERSON + Traag et al. are the authors who developed the Leiden algorithm, a method renowned for its efficiency in recovering hierarchical community structures. This algorithm is widely recognized in the field of Natural Language Processing and Information Retrieval for its ability to accurately detect and map out complex community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 61 + 1 + 1 + 66 + 1eb829d0ace042089f0746f78729696c + + + METHOD + QFS is a task framing that focuses on generating summaries based on specific queries, rather than just concatenating excerpts + fb3c48579608fa28be585ceb6cd2f0fe + 1 + 67 + 015e7b58d1a14b44beab3bbc9f912c18 + + + METHOD + A type of summarization that generates natural language summaries based on specific queries, rather than just extracting text + fb3c48579608fa28be585ceb6cd2f0fe + 3 + 68 + 26f88ab3e2e04c33a459ad6270ade565 + + + TECHNOLOGY + A neural network architecture that has shown substantial improvements in various summarization tasks + fb3c48579608fa28be585ceb6cd2f0fe + 3 + 69 + babe97e1d9784cffa1c85abc1e588126 + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + 35 + 1 + 3 + 70 + 1033a18c45aa4584b2aef6ab96890351 + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + 40 + 1 + 2 + 71 + c9b8ce91fc2945b4907fe35519339cac + + + TECHNOLOGY + GEMINI is a family of highly capable multimodal models, as described in an arXiv preprint. These models are known for their ability to perform in-context learning and summarization, making them a significant advancement in the field of Natural Language Processing and Information Retrieval. + 086021a89900a39bcb62036981737bfa,fb3c48579608fa28be585ceb6cd2f0fe + 36 + 1 + 2 + 72 + fa3c4204421c48609e52c8de2da4c654 + + + TECHNOLOGY + A knowledge graph is a structured representation of information, utilized in the Graph RAG approach for summarization. This structured representation of knowledge is specifically employed in the Graph RAG approach for global summarization, highlighting its role in organizing and integrating information to facilitate comprehensive and coherent summaries. + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + 50 + 1 + 2 + 73 + 53af055f068244d0ac861b2e89376495 + + + REFERENCE + Authors of a paper on Retrieval-augmented generation (RAG) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 63 + 1 + 1 + 74 + c03ab3ce8cb74ad2a03b94723bfab3c7 + + + REFERENCE + Author of a paper on query-focused summarization (QFS) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 75 + ed6d2eee9d7b4f5db466b1f6404d31cc + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 76 + fc01e9baa80e417c9206f941bb279407 + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 77 + 56d0e5ebe79e4814bd1463cf6ca21394 + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 78 + 7c49f2710e8b4d3b8dc9310834406ea5 + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 79 + c6d1e4f56c2843e89cf0b91c10bb6de2 + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 80 + 0adb2d9941f34ef7b2f7743cc6225844 + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 81 + 6b02373137fd438ba96af28f735cdbdb + + + REFERENCE + Authors of a paper on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 35 + 1 + 1 + 82 + 36a4fcd8efc144e6b8af9a1c7ab8b2ce + + + REFERENCE + "BROWN ET AL., 2020" refers to a publication by Brown et al. in 2020, which discusses in-context learning with few-shot examples. The authors of this paper are also known for their work on the GPT series of large language models. + bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 35 + 1 + 2 + 83 + fbeef791d19b413a9c93c6608286ab63 + + + REFERENCE + Authors of a paper on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 40 + 1 + 1 + 84 + d2b629c0396f4180a03e16ddf3818589 + + + REFERENCE + Authors of a paper on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 36 + 1 + 1 + 85 + 6102fc6619ed422ebc42588bfa97355d + + + REFERENCE + "KURATOV ET AL., 2024" refers to a publication by Kuratov and colleagues in 2024. The study discusses the recall degradation and potential for information loss in longer context windows of Large Language Models (LLMs). The authors explore the limitations of these extended context windows, providing insights into how the performance of LLMs can be affected when dealing with longer sequences of text. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 36 + 1 + 3 + 86 + 8d141c0b80f74b79a05eed7fe161fe49 + + + REFERENCE + "LIU ET AL., 2023" refers to a publication by Liu et al. in 2023, which discusses the recall degradation and potential for information loss in longer context windows of large language models (LLMs). The authors explore the limitations of LLM context windows, highlighting how extended contexts can lead to decreased recall accuracy and information retention. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 36 + 1 + 3 + 87 + e22d1d1cd8d14f12b81828d940f40d70 + + + TECHNOLOGY + COMMUNITY DETECTION ALGORITHMS are algorithms used to partition a graph into communities of nodes with stronger connections to one another. These algorithms are designed to identify modular communities of closely-related nodes within a graph, thereby revealing the underlying structure and relationships within the network. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + 61 + 1 + 6 + 88 + 9ab48505fb1b487babd0d1f6d3a3f980 + + + ALGORITHM + Louvain is a community detection algorithm used to partition graphs into modular communities + 21e52bc06a82796b1f4bcd73edda1f2a + 61 + 1 + 1 + 89 + 148fffeb994541b2b4b6dcefda7001a8 + + + DATASET + HOTPOTQA is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical resource for evaluating entity extraction prompts, particularly with advanced models like GPT-4-turbo. Additionally, HotPotQA is utilized to observe the behavior of text chunk extraction within the Graph RAG (Retrieval-Augmented Generation) approach, making it a versatile tool in the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4 + 39 + 1 + 3 + 90 + 89c08e793298442686292454a1abff31 + + + TECHNOLOGY + GPT-4-Turbo is a version of the GPT-4 model characterized by its large context size of 128k tokens, which is utilized in various analytical tasks. Specifically, GPT-4-Turbo is employed for entity extraction in evaluations, leveraging its extensive context capacity to enhance the accuracy and comprehensiveness of the analysis. This model is particularly suited for tasks within the Natural Language Processing and Information Retrieval domain, where handling large volumes of text and extracting relevant entities are critical. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + 39 + 1 + 2 + 91 + 0467928aa65e4a4fba62bdb1467e3a54 + + + DATASET + The "PODCAST TRANSCRIPTS" dataset is a comprehensive collection of compiled transcripts from podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders. This dataset is used for analysis and consists of 1669 text chunks, each containing 600 tokens with 100-token overlaps between chunks, amounting to approximately 1 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620,ebf5249c888e07fedce6572a4c03f88c + 20 + 1 + 19 + 92 + 43c3390303c6476cb65f584e37c3e81c + + + DATASET + The "NEWS ARTICLES" dataset is a comprehensive collection of news articles used for analysis. It serves as a benchmark dataset comprising news articles published from September 2013 to December 2023. The dataset spans a range of categories, including entertainment, business, sports, technology, health, and science. It consists of 3197 text chunks, each containing 600 tokens, with a 100-token overlap between chunks, amounting to approximately 1.7 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620 + 20 + 1 + 13 + 93 + fa14b16c17e3417dba5a4b473ea5b18d + + + METHOD + MAP-REDUCE is a method employed for text summarization by applying a map-reduce approach directly to source texts. It is particularly utilized for query-focused summarization of an entire corpus, enabling efficient processing and extraction of relevant information from large datasets. This technique leverages the map-reduce paradigm to distribute the computational workload, making it suitable for handling extensive text collections in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,973164fa90bf2b4ee267f4fd795916bf + 70 + 1 + 2 + 94 + 7cc3356d38de4328a51a5cbcb187dac3 + + + METRIC + "EMPOWERMENT" is a concept and metric used in the evaluation of various methods, with an average win rate of 51.3%. It measures how well an answer helps the reader understand and make informed judgments about a topic. Specifically, it evaluates the effectiveness of generated answers in empowering users by developing their understanding of broad issues and themes. Empowerment is a target quality in summarization approaches, focusing on the ability to help users reach an informed understanding. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,ebf5249c888e07fedce6572a4c03f88c + 23 + 1 + 6 + 95 + bef16fb5fd7344cca5e295b13ef3e0cd + + + METHOD + Naive RAG is a basic retrieval-augmented generation (RAG) method used as a baseline for comparison in text generation tasks. It converts documents to text, splits them into chunks, and embeds these chunks into a vector space for query matching. While it produces the most direct responses, it is outperformed by global approaches in terms of comprehensiveness and diversity. Naive RAG is also noted for listing public figures mentioned in entertainment articles, focusing on their professional achievements and personal lives. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c + 53 + 1 + 13 + 96 + bb9e01bc171d4326a29afda59ece8d17 + + + METHOD + A method for summarizing source texts using a map-reduce approach + 21e52bc06a82796b1f4bcd73edda1f2a + 50 + 1 + 1 + 97 + 3c063eea52e94164b70c99431ea30bae + + + OUTPUT + Questions generated to evaluate the summarization approach, focusing on understanding activities + 21e52bc06a82796b1f4bcd73edda1f2a + 50 + 1 + 1 + 98 + 252cc8452bfc4c2aa58cab68d8b61879 + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + 50 + 1 + 1 + 99 + 7e2c84548fb94ee395ba8588d8f2a006 + + + METRIC + TOKEN COSTS refer to the computational cost measured in tokens used in the summarization process. Specifically, in the context of the Graph RAG (Retrieval-Augmented Generation) approach, token costs denote the number of tokens required for processing text. This metric is crucial for evaluating the efficiency and scalability of text processing methods within the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 50 + 1 + 1 + 100 + f034618dde7948beb6dab30176d0fc87 + + + PROCESS + The high-level process of the Graph RAG approach and pipeline + 21e52bc06a82796b1f4bcd73edda1f2a + 50 + 1 + 1 + 101 + 5c41f96be13e49dba649454297834546 + + + PARAMETER + Design parameters are key settings and configurations in the Graph RAG approach. These parameters are crucial as they influence the design of the Graph RAG approach and pipeline, determining the effectiveness and efficiency of the overall system. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + PARAMETER + 50 + 1 + 1 + 102 + 7ea4afbf8a264f29af29950ce98105ba + + + METHOD + GLOBAL SUMMARIZATION is a method for summarizing information on a global scale. It aims to encapsulate the overall structure and semantics of a dataset, providing a comprehensive overview of information from large datasets or corpora. This technique is particularly useful in the field of Natural Language Processing and Information Retrieval, where it helps in distilling vast amounts of data into coherent and concise summaries, facilitating better understanding and analysis of the underlying information. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e4d9b12cf2b4c691c74019eefff4fb39 + 55 + 1 + 5 + 103 + 91ff849d12b24574b0691dbddf44968b + + + ATTRIBUTE + Modularity is an inherent quality of graphs that allows them to be partitioned into communities of closely-related nodes + 21e52bc06a82796b1f4bcd73edda1f2a + 50 + 1 + 1 + 104 + d73c1f2fb3094d8dace42ad2a76e9a52 + + + OUTPUT + Descriptions generated from modular communities in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + 66 + 1 + 1 + 105 + cdc8901e668749889bd49bebdc4ff1f6 + + + INPUT + A specific question or request for information that the summarization methods aim to answer + 21e52bc06a82796b1f4bcd73edda1f2a + 50 + 1 + 1 + 106 + 36084a9fab53433493f079e97e68bf65 + + + DATASET + A large collection of texts or documents used for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + 50 + 1 + 1 + 107 + eebcc7ec8e3e4df7aea83659bbdc2199 + + + OUTPUT + Intermediate answers generated from community summaries before being combined into a final global answer + 21e52bc06a82796b1f4bcd73edda1f2a + 66 + 1 + 2 + 108 + ceadf262ef834e9ab146b20650912cae + + + OUTPUT + The comprehensive answer generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + 66 + 1 + 1 + 109 + 7f65feab75424b53b24470d305ba331a + + + METHOD + A method that focuses on generating questions to understand activities from datasets + 21e52bc06a82796b1f4bcd73edda1f2a + 49 + 1 + 2 + 110 + fd9cb733b28d420cb5cef01e545a132c + + + INPUT + Brief descriptions of datasets used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + 49 + 1 + 1 + 111 + 0fbcca3f17c649a08aea64b5a7d9ef36 + + + DATASET + Datasets that represent real-world information, such as podcast transcripts and news articles + 21e52bc06a82796b1f4bcd73edda1f2a + 50 + 1 + 1 + 112 + 482027a59f32484c9c44fd700615c1b6 + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + 50 + 1 + 1 + 113 + de837ff3d626451282ff6ac77a82216d + + + METHOD + A method that summarizes the original source texts directly + 21e52bc06a82796b1f4bcd73edda1f2a + 50 + 1 + 1 + 114 + 460295fed3ae4cd39f9f274cec9c2506 + + + OUTPUT + LOW-LEVEL COMMUNITY SUMMARIES are a type of community summary used in the News dataset for analysis. These summaries provide a detailed overview of the source text and are generated from lower hierarchical levels of the community in the knowledge graph. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,bc9e2c9e369c4108cf4f6dd5f60960f4 + 60 + 1 + 3 + 115 + 553b285bba60460ab1ed8341ae61282b + + + OUTPUT + INTERMEDIATE-LEVEL COMMUNITY SUMMARIES are summaries that provide a mid-level overview of the source text. These summaries are generated from intermediate hierarchical levels of the community in the knowledge graph, offering a balanced perspective that captures essential details without overwhelming the reader with excessive information. This approach ensures that the summaries are both informative and concise, making them valuable for understanding the structural dynamics and relationships within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 60 + 1 + 2 + 116 + cec95bf17e7e4c939b56c9c6f402a29f + + + OUTPUT + Summaries generated from higher hierarchical levels of the community in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + 50 + 1 + 1 + 117 + 599164aead034bc19446efacc77554d2 + + + PROCESS, SYSTEM + The entity "PIPELINE" refers to a series of processes or steps used to analyze and summarize a dataset. Specifically, in the context of the Graph RAG approach, the pipeline denotes the sequence of steps and processes involved. This structured sequence is essential for systematically handling data, ensuring that each stage of the analysis is methodically executed to achieve accurate and comprehensive results. + 7fb7d9ce2da9c940a32afdd87d1d9e56,bc9e2c9e369c4108cf4f6dd5f60960f4 + 68 + 1 + 2 + 118 + bbf148ae4d48422f8fdef754cfa2b9e4 + + + DATA STRUCTURE, OUTPUT + The "GRAPH INDEX" is a data structure used in Retrieval-Augmented Generation (RAG) systems to organize and retrieve information. It is a self-generated index that enables Graph RAG by utilizing a graph structure to organize and retrieve data. This index is created from a graph structure and is employed for tasks such as query-focused summarization. The graph index includes various elements extracted from text chunks using Large Language Model (LLM) prompts. Additionally, it supports conditions C0-C3 and is created using generic prompts for entity and relationship extraction. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + 22 + 1 + 18 + 119 + de61b2670999433f807a6a1dc2b81e43 + + + DATA, UNIT + Entity references are mentions of entities within text chunks, extracted during the processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + 60 + 1 + 3 + 120 + 3e95dacfe57b4d57b5da4310ef2e157f + + + METRIC + Recall is a metric used to measure the completeness of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 60 + 1 + 1 + 121 + 1f1545308e9347af91fd03b94aadc21f + + + METRIC + Precision is a metric used to measure the accuracy of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 60 + 1 + 1 + 122 + 6ea81acaf232485e94fff638e03336e1 + + + TECHNIQUE, METHOD + FEW-SHOT EXAMPLES are specialized instances provided to the Large Language Model (LLM) to improve its performance in domains with specialized knowledge such as science, medicine, and law. These examples are tailored to the domain of the data used in the graph indexing process and serve as sample inputs for in-context learning. By tailoring the extraction prompt to the document corpus domain, few-shot examples enhance the LLM's ability to understand and process domain-specific information effectively. + 2c6ed90897310eea2f28e33fff1c32b0,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4 + 25 + 1 + 8 + 123 + d136b08d586d488f9e4188b524c85a29 + + + DATA, UNIT + Named entities are specific types of entities such as people, places, and organizations, extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 25 + 1 + 1 + 124 + cccfa151fedc4b218a8d96adc7dceabe + + + REFERENCE, PUBLICATION + A reference to a publication by Yang et al. in 2018, introducing the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + 39 + 1 + 1 + 125 + ce54725672a74ebcabe6127577dacb2b + + + METHOD, APPROACH + Techniques refer to the specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 50 + 1 + 1 + 126 + ea2b28ca1a974ffab4517811dc1d1e5c + + + ATTRIBUTE, CONFIGURATION + Implementation details are specific configurations and settings used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 50 + 1 + 1 + 127 + aff21f1da1654e7babdcf3fb0e4a75fc + + + PROCESS, METHOD + A single extraction round refers to one complete cycle of extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + 25 + 1 + 1 + 128 + dc2cc9016e3f49dbac7232f05cce794d + + + ATTRIBUTE, CONFIGURATION + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + 60 + 1 + 1 + 129 + 6ea0cef05f694dcea455478f40674e45 + + + METRIC, ISSUE + Recall degradation refers to the decrease in recall performance when using longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + 25 + 1 + 1 + 130 + 7ab5d53a872f4dfc98f3d386879f3c75 + + + PROCESS, METHOD + The extraction process involves identifying and extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + 25 + 1 + 2 + 131 + af1d0fec22114a3398b8016f5225f9ed + + + ATTRIBUTE, CONFIGURATION + Domain refers to the specific area of knowledge or field to which the document corpus belongs + bc9e2c9e369c4108cf4f6dd5f60960f4 + 60 + 1 + 1 + 132 + b07a7f088364459098cd8511ff27a4c8 + + + DATA, INPUT + Document corpus refers to the collection of documents being processed in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 60 + 1 + 2 + 133 + 8870cf2b5df64d2cab5820f67e29b9f1 + + + TECHNIQUE, METHOD + Default prompt is the standard set of instructions given to the LLM for extracting named entities + bc9e2c9e369c4108cf4f6dd5f60960f4 + 25 + 1 + 2 + 134 + cd130938a2844050be991af70baf5ee0 + + + TECHNIQUE, METHOD + Secondary extraction prompt is an additional set of instructions given to the LLM for extracting covariates + bc9e2c9e369c4108cf4f6dd5f60960f4 + 25 + 1 + 2 + 135 + 43544b99c3b04b059546198a0ae6366d + + + METHOD + A covariate prompt is used to extract additional attributes associated with detected entities, including claims linked to the entities + 2c6ed90897310eea2f28e33fff1c32b0 + 2 + 136 + a671bf7fea2f4514b6e96ba99127fafd + + + CONCEPT + Claims are statements or assertions linked to detected entities, including details such as subject, object, type, description, source text span, and start and end dates + 2c6ed90897310eea2f28e33fff1c32b0 + 7 + 137 + 525f41ea20274a05af4e52b625b473f3 + + + METHOD + Gleanings refer to multiple rounds of entity extraction to ensure that no entities are missed in the process + 2c6ed90897310eea2f28e33fff1c32b0 + 25 + 1 + 1 + 138 + 071a416efbec4f0886c19ac68f6d43cb + + + TECHNIQUE + Logit bias is a technique used to force a yes/no decision from the LLM during the entity extraction process + 2c6ed90897310eea2f28e33fff1c32b0 + 25 + 1 + 1 + 139 + 6d8473ef3b1042bf87178a611e3dbcc6 + + + CONCEPT + An entity node is a representation of an entity in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 59 + 1 + 1 + 140 + 30c9641543c24773938bd8ec57ea98ab + + + CONCEPT + A relationship edge is a representation of a relationship between entities in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 59 + 1 + 1 + 141 + 18b839da898e4026b81727d759d95c6a + + + CONCEPT + A claim covariate is an additional attribute or variable associated with a claim in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 59 + 1 + 1 + 142 + eeef6ae5c464400c8755900b4f1ac37a + + + CONCEPT + Communities of entities are groups of closely-related entities detected and summarized by the LLM + 2c6ed90897310eea2f28e33fff1c32b0 + 25 + 1 + 2 + 143 + 422433aa45804c7ebb973b2fafce5da6 + + + CONCEPT + The "NOISY GRAPH STRUCTURE" refers to a graph structure that may contain inconsistencies or errors, making it challenging to analyze. This type of graph often includes duplicate or inconsistent entity elements due to variations in text format. These inconsistencies can arise from various sources, such as data entry errors, differing data formats, or incomplete information, which complicate the process of extracting meaningful insights and relationships from the graph. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56 + 25 + 1 + 1 + 144 + 86505bca739d4bccaaa1a8e0f3baffdc + + + DOMAIN + Science is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + 25 + 1 + 1 + 145 + 1af9faf341e14a5bbf4ddc9080e8dc0b + + + DOMAIN + Medicine is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + 25 + 1 + 1 + 146 + 353d91abc68648639d65a549e59b5cf3 + + + DOMAIN + Law is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + 25 + 1 + 1 + 147 + 7ce637e4f35b42e3a9f8272cab69cd22 + + + ATTRIBUTE + Source text span is an attribute of a claim that indicates the specific portion of text from which the claim was extracted + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 148 + 4d999d7744b04a998475f8f8531589f0 + + + ATTRIBUTE + Start date is an attribute of a claim that indicates when the event or fact described in the claim began + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 149 + 9a6f414210e14841a5b0e661aedc898d + + + ATTRIBUTE + End date is an attribute of a claim that indicates when the event or fact described in the claim ended + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 150 + db541b7260974db8bac94e953009f60e + + + ATTRIBUTE + Description is an attribute of a claim that provides a detailed explanation of the claim + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 151 + f2ff8044718648e18acef16dd9a65436 + + + ATTRIBUTE + Subject is an attribute of a claim that indicates the main entity involved in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 152 + 00d785e7d76b47ec81b508e768d40584 + + + ATTRIBUTE + Object is an attribute of a claim that indicates the entity that is affected by the subject in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 153 + 87915637da3e474c9349bd0ae604bd95 + + + CONCEPT + A concept referring to an entity that has multiple name variations but is resilient to such variations due to sufficient connectivity to closely-related entities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1 + 154 + 8f1eba29f39e411188200bf0d14628ec + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models capable of understanding and generating human-like text. They are utilized for a variety of tasks, including the creation and completion of knowledge graphs, which are essential for structuring and interlinking information in a meaningful way. Additionally, LLMs serve as evaluators of natural language generation, assessing the quality and coherence of text produced by other AI systems. These models play a crucial role in the field of Natural Language Processing and Information Retrieval, contributing significantly to advancements in how machines comprehend and interact with human language. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf + 31 + 1 + 5 + 155 + 7282c73622b8408e97289d959faff483 + + + TECHNOLOGY + Structured representations of knowledge in the form of triples (subject, predicate, object) used for reasoning tasks + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 22 + 1 + 1 + 156 + 3deb220d31f74103aa44870a36a63220 + + + CONCEPT + Nodes in a graph that are of the same type and are described using rich descriptive text + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 2 + 157 + af7a1584dd15492cb9a4940e285f57fc + + + CONCEPT + Edges in a graph that represent relationships between entity nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 2 + 158 + 6e8d9029ce4e4ea182367173ab2c7bbf + + + METRIC + Weights assigned to edges in a graph, representing the normalized counts of detected relationship instances + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1 + 159 + cbf232211e7d4eb6abdbe182f71c2cf0 + + + CONCEPT + The "HIERARCHICAL COMMUNITY STRUCTURE" is a framework in which communities are organized in a hierarchy, with each level providing a partition of the graph nodes. This structure organizes data into a hierarchy of communities, facilitating a multi-level clustering approach. Hierarchical community structure is utilized to generate community summaries, offering a comprehensive method for understanding the relationships and structural dynamics within specialized communities. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39 + 65 + 1 + 7 + 160 + bb0cff774a4440b289cc6f3b929fe13c + + + CONCEPT + A division of graph nodes into mutually-exclusive, collectively-exhaustive communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 55 + 1 + 2 + 161 + ce55841ebfdd47008bab8c258f10372e + + + TECHNOLOGY + MULTIHOP-RAG is a benchmark dataset comprising news articles published from September 2013 to December 2023, covering a range of categories including entertainment, business, sports, technology, health, and science. It is specifically designed for open-domain question answering, targeting explicit fact retrieval. Additionally, MULTIHOP-RAG represents a specific implementation or variant of Retrieval-Augmented Generation (RAG) used in the context of graph communities. This dataset is also utilized for community detection and analysis, making it a versatile tool in the field of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,922778ce1cb2fdd6dbab1746c8795620 + 62 + 1 + 8 + 162 + 6090e736374d45fd84f0e4610a314f8f + + + PERSON + An author who has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 61 + 1 + 1 + 163 + 0e8d921ccd8d4a8594b65b7fd19f7120 + + + PERSON + Authors who have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 61 + 1 + 1 + 164 + 59c726a8792d443e84ab052cb7942b4a + + + CONCEPT + The entity "DATASET" refers to a collection of data used for various purposes such as analysis, summarization, and evaluation. This can include diverse types of data like podcast transcripts and news articles. Specifically, the term encompasses datasets used for evaluation purposes, including notable examples like the Podcast and News datasets. + 1d07b4248c2655081c7af0e373bd70c9,7fb7d9ce2da9c940a32afdd87d1d9e56,973164fa90bf2b4ee267f4fd795916bf + 3 + 165 + 4f2c665decf242b0bfcaf7350b0e02ed + + + CONCEPT + GLOBAL QUERIES refer to questions or inquiries that require comprehensive answers derived from multiple sources or datasets. These queries aim to retrieve information from a global perspective, covering the entire dataset. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 70 + 1 + 2 + 166 + 66cdf168f36d4a57a505028c97dc06e0 + + + CONCEPT + ROOT COMMUNITIES are the top-level clusters in a hierarchical community structure. These communities represent the highest level of organization within the hierarchy, serving as the primary divisions from which more specific sub-communities branch out. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 65 + 1 + 4 + 167 + 38f51478f41f48db9bee570859b6f43e + + + CONCEPT + SUB-COMMUNITIES are lower-level clusters within root communities in a hierarchical community structure, providing more detailed information. These sub-communities play a crucial role in breaking down the larger, more general root communities into more specific and focused groups, thereby facilitating a deeper and more granular understanding of the overall community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 65 + 1 + 5 + 168 + 896d2a51e8de47de85ba8ced108c3d53 + + + TECHNOLOGY + Detailed documents that provide information about specific subtopics within a community + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 65 + 1 + 1 + 169 + 14555b518e954637b83aa762dc03164e + + + CONCEPT + The division of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 68 + 1 + 2 + 170 + b1f6164116d44fe8b8f135d7f65b9e58 + + + CONCEPT + A system in which elements are ranked or organized in levels + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 68 + 1 + 3 + 171 + c8b2408617804483b620e1a6691ac90d + + + CONCEPT + LEVEL 0 represents the root-level communities in the hierarchical clustering with maximum modularity. It serves as the foundational layer in a hierarchical community structure, indicating the initial and most significant division of the dataset into distinct groups. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 68 + 1 + 2 + 172 + a5e0d1644eb547ba9a5c3211aac4631a + + + CONCEPT + LEVEL 1 is a sub-level in a hierarchical community structure, providing more detailed information about the internal organization. Specifically, Level 1 represents sub-communities within the root-level communities, thereby revealing the internal structure and dynamics of these larger groups. This level of granularity helps in understanding the intricate relationships and specialized interactions that occur within the broader community framework. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 68 + 1 + 2 + 173 + 5a28b94bc63b44edb30c54748fd14f15 + + + CONCEPT + A visual representation of graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 61 + 1 + 1 + 174 + f97011b2a99d44648e18d517e1eae15c + + + METHOD + The Leiden algorithm is a method used for detecting communities in large networks + 843fc5421e086120ffa1c75856ecf6cd + 62 + 1 + 1 + 175 + 35489ca6a63b47d6a8913cf333818bc1 + + + TOOL + OpenORD is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + 62 + 1 + 1 + 176 + 5d3344f45e654d2c808481672f2f08dd + + + TOOL + Force Atlas 2 is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + 62 + 1 + 1 + 177 + 6fb57f83baec45c9b30490ee991f433f + + + ELEMENT + Nodes represent entities in a graph, with size proportional to their degree + 843fc5421e086120ffa1c75856ecf6cd + 62 + 1 + 2 + 178 + 68762e6f0d1c41cd857c6b964a8e76c3 + + + ELEMENT + Edges represent connections between nodes in a graph + 843fc5421e086120ffa1c75856ecf6cd + 59 + 1 + 2 + 179 + 70634e10a5e845aa8c6a32fe7e8eb2b2 + + + ELEMENT + Covariates are variables that are linked to nodes and edges in a graph + 843fc5421e086120ffa1c75856ecf6cd + 62 + 1 + 2 + 180 + 04085f7cf46544b79597fc49286ff84d + + + CONCEPT + The LLM context window is the token limit within which summaries are added for processing by a language model + 843fc5421e086120ffa1c75856ecf6cd + 69 + 1 + 2 + 181 + d203efdbfb2f4b2a899abfb31cf72e82 + + + METHOD + Hierarchical clustering is a method of clustering data into a tree-like structure with multiple levels + 843fc5421e086120ffa1c75856ecf6cd + 65 + 1 + 2 + 182 + 6731a665561840c2898ce8c9788e4c88 + + + CONCEPT + The token limit is the maximum number of tokens that can be processed in a single context window by a language model + 843fc5421e086120ffa1c75856ecf6cd + 69 + 1 + 1 + 183 + 4026806fa92f4e849a59a7f5c9a45c79 + + + CONCEPT + Summary detail refers to the level of detail provided in a summary + 843fc5421e086120ffa1c75856ecf6cd + 66 + 1 + 1 + 184 + 68e0c60d2e8845d89d9d0ad397833648 + + + CONCEPT + Scope refers to the range or extent of information covered in a summary + 843fc5421e086120ffa1c75856ecf6cd + 66 + 1 + 1 + 185 + 101572f552b54e529fe7765c05168981 + + + CONCEPT + A "USER QUERY" is a question or inquiry posed by a user seeking information, which the system aims to answer. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd + CONCEPT + 64 + 1 + 2 + 186 + 60c58026b2764b40adffca6eaa31d6d9 + + + ELEMENT + Chunks are segments of community summaries divided into pre-specified token sizes + 843fc5421e086120ffa1c75856ecf6cd + ELEMENT + 66 + 1 + 1 + 187 + ad1595a78935472999444c9330e7730e + + + METRIC + Prominence is a metric used to prioritize community edges based on the combined degree of source and target nodes + 843fc5421e086120ffa1c75856ecf6cd + 2 + 188 + 735d19aea0744b2295556841c5c4c3fd + + + METRIC + Combined source and target node degree is a metric used to measure the overall prominence of community edges + 843fc5421e086120ffa1c75856ecf6cd + 1 + 189 + c725babdb14a485582f8fbdf95429030 + + + ELEMENT + Community edges are connections between nodes within a community, prioritized based on prominence + 843fc5421e086120ffa1c75856ecf6cd + 1 + 190 + a0047221896d418d849847d422fa4bb8 + + + CONCEPT + Sub-community summaries are shorter summaries of sub-communities used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + 59 + 1 + 1 + 191 + 98fc2ee593184c5a839454db4eec7013 + + + CONCEPT + Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + 66 + 1 + 1 + 192 + 80020a1da63042459e00266b2a605452 + + + CATEGORY + Community level refers to the different levels in the hierarchical community structure used to generate summaries + 843fc5421e086120ffa1c75856ecf6cd + 65 + 1 + 1 + 193 + 31a7e680c4d54101afe4c8d52d246913 + + + DATA + Chunks are segments of community summaries divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + 67 + 1 + 3 + 194 + 351abba16e5c448994c6daf48121b14d + + + METRIC + A score between 0-100 generated by the LLM to indicate how helpful an answer is in addressing the target question + 1d07b4248c2655081c7af0e373bd70c9 + 64 + 1 + 3 + 195 + 50ea7d3b69614bcdbfbff7ddbfbf3d34 + + + USER + A user looking for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + 6 + 196 + 004f40a5aeca48a1879db728eb12bcba + + + USER + A user incorporating current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + 6 + 197 + 4465efb7f6ed4dedad72a658184addd2 + + + TOPIC + A topic dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 198 + b0dd60e11dad4ff782623acf039b3948 + + + TOPIC + A topic discussing the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 199 + db8c43fa4df947b09e5754d3b1393ead + + + TOPIC + A topic discussing the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 200 + 5dabc4cd05da425cb194a04482bf0c29 + + + TOPIC + A topic discussing suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 201 + 9d08f285a7be4c79b8f359c51d51db37 + + + TOPIC + A topic discussing collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 202 + adffed660d154b519c1817e514e83096 + + + TOPIC + Current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 203 + b7e9c9ef572c445a9574ca571e41fb96 + + + TOPIC + A topic addressing the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 204 + dcb9f281cd6248c699e0ebb285a42a5e + + + TOPIC + Examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 205 + 072cdee531b74513984f49d99a8d64a0 + + + TOPIC + Insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 206 + 5ae335d9210a45fda3f92a9a028d6d9b + + + TOPIC + The importance of health literacy highlighted through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 207 + 5ac60a941a5b4934bdc43d2f87de601c + + + OUTPUT + Answers generated for each chunk of community summaries + 1d07b4248c2655081c7af0e373bd70c9 + 64 + 1 + 3 + 208 + d405c3154d0e48ce96fad4c28fe20590 + + + METRIC + The pre-specified size of tokens used to divide community summaries into chunks + 1d07b4248c2655081c7af0e373bd70c9 + 67 + 1 + 1 + 209 + 7923d8521c744bd9aab131c1aea91ffd + + + TECHNOLOGY + The "CONTEXT WINDOW" refers to a window of text used to generate answers, constrained by token size. The size of the context window is consistent across all conditions, ensuring uniformity in answer generation processes. + 1d07b4248c2655081c7af0e373bd70c9,973164fa90bf2b4ee267f4fd795916bf + 64 + 1 + 2 + 210 + 5bd156c87ec44e19ae6f8f62e6e50b9d + + + PERSON + Kevin Scott is the Chief Technology Officer (CTO) of Microsoft and actively participates in podcast conversations. His involvement in these discussions is documented and compiled in the dataset, highlighting his contributions to the field of technology and his role in shaping Microsoft's strategic direction. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + 20 + 1 + 2 + 211 + c1a146d7fb16429ea6d0aa2a55ee597f + + + PERSON + Individuals who are leaders in the technology industry and participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + 20 + 1 + 1 + 212 + ede9350632084da5b0b577ff799ab14b + + + INPUT + A specific activity or goal that the user aims to achieve using the datasets + 1d07b4248c2655081c7af0e373bd70c9 + 2 + 213 + ed559fb4ebde45518849ec803b350fa3 + + + INPUT + QUESTIONS refer to specific inquiries generated by the Large Language Model (LLM) based on the user's task and the target datasets. These questions are utilized in the analysis to evaluate the performance of different methods within the domain of Natural Language Processing and Information Retrieval. The generation and subsequent use of these questions are crucial for assessing the effectiveness and accuracy of various computational techniques and models. + 1d07b4248c2655081c7af0e373bd70c9,4c855404ee3d3c94aa2136f1513c666f + 4 + 214 + f422035f8b78417f98e4d116971cf9f3 + + + + + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 215 + c79d686eba044c5586c706cdc096817d + + + DATASET + MT-BENCH is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical tool in evaluating the performance of models in retrieving accurate and relevant information from a broad range of topics. MT-Bench is prominently featured in the academic paper titled "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," where it is utilized to assess the capabilities of large language models in a structured and rigorous manner. This dataset plays a significant role in advancing the field of Natural Language Processing and Information Retrieval by providing a standardized metric for model comparison and evaluation. + 922778ce1cb2fdd6dbab1746c8795620,b1bbda43309e8e0e2175ea034aa88e13 + DATASET + 46 + 1 + 12 + 216 + 0f70db1e598d463fbbcdd1e288bd9490 + + + PROCESS + The process through which people inspect, engage with, and contextualize data within the broader scope of real-world activities + 922778ce1cb2fdd6dbab1746c8795620 + PROCESS + 1 + 217 + b35c3d1a7daa4924b6bdb58bc69c354d + + + TECHNOLOGY + Retrieval-Augmented Generation systems used for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + TECHNOLOGY + 20 + 1 + 3 + 218 + a97e2ecd870944cfbe71c79bc0fcc752 + + + AUTHORS + Authors of a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 1 + 219 + 3e1b063bbfa9423d84e50311296d2f3c + + + AUTHORS + Authors of a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 1 + 220 + 9a8ce816ee954bdabd01ea2081538009 + + + AUTHORS + Authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 62 + 1 + 1 + 221 + 09f18f81442d4d6d93a90f0fac683f9b + + + AUTHORS + Authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 39 + 1 + 1 + 222 + e02be3e37ca0454883a4c1fd859c24bb + + + AUTHORS + Authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 46 + 1 + 1 + 223 + 6e0c81bef5364c988b21bf9b709d9861 + + + + + 922778ce1cb2fdd6dbab1746c8795620 + 1 + 224 + 1dbc51475cb04dafa4a8833a8378635e + + + PODCAST + "BEHIND THE TECH" is a podcast series featuring conversations between Kevin Scott and other technology leaders. It serves as a media platform associated with Kevin Scott, providing insights and discussions on various technological advancements and industry trends. + 833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620 + 0 + 225 + c12b9ebd8b4e42b7896822a32e3fa6eb + + + PERSON + Kevin Scott, Microsoft CTO, who participates in the podcast conversations compiled in the dataset + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 226 + 27505f6ade4b4e5f9316ffe9c34821f7 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 227 + 0ee7db2c6bea4630ba9f0c25e8a967ad + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 228 + 5a6c1d15424149f69052cd8d91fbff75 + + + DATASET + A benchmark dataset for open-domain question answering, targeting explicit fact retrieval + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 229 + d005bf75c31d4848ad7041f39651e59c + + + METRIC + N represents the number of test questions per dataset used in the evaluation + 973164fa90bf2b4ee267f4fd795916bf + 1 + 230 + 9b3eef8f3a3a45e6873838db95295b8a + + + METHOD + A method applying a map-reduce approach directly to source texts for summarization + 973164fa90bf2b4ee267f4fd795916bf + 70 + 1 + 3 + 231 + fdc954b454744820804d7798f3e0b5de + + + METHOD + A na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached + 973164fa90bf2b4ee267f4fd795916bf + 23 + 1 + 2 + 232 + 49c1383836934ec495c3b35769100a73 + + + CATEGORY + C0 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a root-level community summary, which is utilized to answer user queries by providing the fewest number of summaries. This category is essential for understanding the structural dynamics within the community, particularly in the domain of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 20 + 1 + 8 + 233 + 859dedcc3736439a8a563419f16cb3d8 + + + CATEGORY + C1 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a high-level community summary used to answer user queries, effectively representing sub-communities of C0. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 20 + 1 + 9 + 234 + 6078b9980a6c4dcd9198d151b833ead7 + + + CATEGORY + C2 is a category or condition used in the analysis, representing a specific subset of the data. It functions as an intermediate-level community summary used to answer user queries, representing sub-communities of C1. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 20 + 1 + 9 + 235 + f93cd6b8213e46dda67af7e5382e1bd2 + + + CATEGORY + C3 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a category or condition that encapsulates low-level community summaries, which are instrumental in answering user queries. These summaries represent sub-communities of C2, providing detailed insights into the structural dynamics and relationships within the broader community. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 20 + 1 + 9 + 236 + 496f17c2f74244c681db1b23c7a39c0c + + + METHOD + TS, or "Text Summarization," is a category or condition used in the analysis, representing a specific subset of the data. It is particularly focused on source text summarization within the analysis. TS employs a text summarization method that applies a map-reduce approach directly to source texts, facilitating efficient and scalable summarization processes. This category is integral to understanding and processing large volumes of text data, making it a crucial component in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 20 + 1 + 10 + 237 + da1684437ab04f23adac28ff70bd8429 + + + METHOD + "SS" is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a baseline condition and is associated with a na¨ıve RAG (Retrieval-Augmented Generation) approach. In this context, text chunks are retrieved and added to the context window until the token limit is reached. + 4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 23 + 1 + 4 + 238 + 4517768fc4e24bd2a790be0e08a7856e + + + CONCEPT + The prompts used for answer generation, which are the same across all conditions with minor modifications + 973164fa90bf2b4ee267f4fd795916bf + 64 + 1 + 1 + 239 + 545edff337344e518f68d1301d745455 + + + DATASET + The "PODCAST DATASET" is a collection of podcast transcripts utilized for both analysis and evaluation purposes. This dataset is specifically designed to support various analytical tasks, providing a rich source of textual data for researchers and practitioners in the field of Natural Language Processing and Information Retrieval. The transcripts within the dataset offer valuable insights and serve as a critical resource for evaluating different computational models and techniques. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + 21 + 1 + 15 + 240 + 9376ce8940e647a99e5e087514b88fa4 + + + DATASET + The "NEWS DATASET" is a collection of news articles utilized for both analysis and evaluation purposes. This dataset serves as a valuable resource for examining and assessing various aspects of news content, making it an essential tool in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + 21 + 1 + 15 + 241 + b38a636e86984600bb4b57c2e2df9747 + + + CONCEPT + METRICS in the context of Natural Language Processing and Information Retrieval are essential tools used to evaluate the performance of natural language generation. These metrics include both reference-based metrics, which compare generated texts to a set of reference texts, and qualities of the generated texts themselves. They are crucial in the analysis to assess the effectiveness of different methods in generating natural language, ensuring that the outputs are both accurate and of high quality. + 4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + 3 + 242 + 4bc7440b8f4b4e4cae65a5c49defa923 + + + REFERENCE + "WANG ET AL., 2023A" refers to a study conducted by Wang and colleagues in 2023, which highlights the effectiveness of Large Language Models (LLMs) in evaluation. This study is a significant contribution to the field, providing insights into the capabilities and performance of LLMs in various evaluative tasks. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 31 + 1 + 2 + 243 + 5d1b038ce8be4533b54dd79d6496de9b + + + REFERENCE + "ZHENG ET AL., 2024" refers to a study conducted by Zheng and colleagues in 2024. This study highlights the effectiveness of Large Language Models (LLMs) in evaluation processes. The research, authored by Zheng et al., provides significant insights into the capabilities and applications of LLMs within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 31 + 1 + 3 + 244 + ac6e5a44e0c04a4fa93589376fde4c34 + + + REFERENCE + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + 1 + 245 + 40e4ef7dbc98473ba311bd837859a62a + + + CONCEPT + The entity "CONDITIONS" refers to the different scenarios or variables that are compared in an experiment. Specifically, in the context of the analysis, these conditions include Graph RAG, text summarization, and semantic search RAG. These conditions are used to evaluate and compare various aspects of performance and effectiveness within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 50 + 1 + 3 + 246 + 222f0ea8a5684123a7045986640ec844 + + + CONCEPT + USER QUERIES refer to the inquiries made by users to retrieve information. These queries are answered using different methods and conditions, depending on the context and the specific requirements of the information retrieval process. + 973164fa90bf2b4ee267f4fd795916bf,e4d9b12cf2b4c691c74019eefff4fb39 + 20 + 1 + 6 + 247 + 668cf1fdfd644d39acc6350b86117ea2 + + + CONCEPT + Types of entities extracted during the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + 22 + 1 + 1 + 248 + 478e4c72d8fb46dd8cc9f0691c9878fd + + + METRIC + The "CONTEXT WINDOW SIZE" refers to the fixed size of the context window used in various stages of natural language processing and information retrieval tasks. For the final evaluation, the context window size is set to 8k tokens. During the analysis phase, different context window sizes are tested, including 8k, 16k, 32k, and 64k tokens. Additionally, in the graph indexing process, the context window size is set to 600 tokens. This variability in context window sizes highlights the importance of adapting the window size to the specific requirements of different tasks within the domain. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + 39 + 1 + 9 + 249 + 82b0446e7c9d4fc793f7b97f890e9049 + + + CONCEPT + The process of extracting information, with 1 gleaning for the Podcast dataset and 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + 21 + 1 + 2 + 250 + 8169efeea3ce473d9fd2f1c688126a1c + + + TECHNOLOGY + Natural Language Generation (NLG) is a subfield of artificial intelligence that focuses on generating human-like text from data + 322e02986c8724eedbcf3ebfa20b989c + 31 + 1 + 7 + 251 + c2d48b75af6a4d7989ccf9eceabd934e + + + METHOD + A method where a Large Language Model (LLM) is used to compare and evaluate competing outputs + 322e02986c8724eedbcf3ebfa20b989c + 31 + 1 + 1 + 252 + 5f1fc373a8f34050a5f7dbd8ac852c1b + + + METHOD + A method for evaluating the performance of Retrieval-Augmented Generation (RAG) systems, focusing on context relevance, faithfulness, and answer relevance + 322e02986c8724eedbcf3ebfa20b989c + 4 + 253 + 0c010fa3aeac4b28b2fbb8c2339c2521 + + + PUBLICATION + A reference to a study or paper authored by Es and others in 2023 + 322e02986c8724eedbcf3ebfa20b989c + 1 + 254 + c2999bdca08a478b84b10219875b285e + + + TOOL + A Large Language Model (LLM) used to evaluate and compare generated texts based on specific metrics + 322e02986c8724eedbcf3ebfa20b989c + 11 + 255 + 263d07354a1b4336b462024288f9bcd3 + + + METRIC + DIRECTNESS is a metric that measures how specifically and clearly an answer addresses a question. It is used to evaluate the straightforwardness of the generated answers. Additionally, it serves as a validity test metric to measure the directness of responses, with naive RAG (Retrieval-Augmented Generation) producing the most direct responses. + 322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 3 + 256 + f9005e5c01b44bb489f7112322fd1162 + + + DATA + An example of LLM-generated assessment shown in a table format + 322e02986c8724eedbcf3ebfa20b989c + 1 + 257 + d9ef017549724f4fbc4ff4ba6701dac0 + + + DATA + The entity "QUESTION" refers to a specific query used in the evaluation process, particularly as a metric to evaluate the generated responses by asking specific questions. This approach is commonly employed in the domain of Natural Language Processing and Information Retrieval to assess the quality and relevance of responses generated by various models or systems. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 2 + 258 + 33b9e826af3f43838c07c847b6349497 + + + ENTITY + Public figures are individuals who have gained fame or notoriety in various sectors such as entertainment, sports, and digital media. These individuals are well-known in the entertainment industry and are frequently mentioned across various articles. Their prominence in public discourse spans multiple domains, reflecting their influence and recognition in society. + 322e02986c8724eedbcf3ebfa20b989c,718017a4871c909420f84b85b8ba969d + 54 + 1 + 5 + 259 + dbe9063124d047dc8d6fcaeadcda038f + + + DATASET + ENTERTAINMENT ARTICLES is a collection of articles focused on the entertainment industry. This dataset consists of articles related to various aspects of the entertainment sector, providing a comprehensive resource for understanding trends, developments, and key topics within this field. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + 53 + 1 + 6 + 260 + c885166d0c454a748376b56279f96408 + + + DOMAIN + The **ENTERTAINMENT INDUSTRY** is a multifaceted sector that encompasses various forms of entertainment, including movies, music, television, sports, and digital media. This industry is characterized by its diverse range of content and mediums, which collectively contribute to its broad appeal and significant cultural impact. The entertainment industry plays a crucial role in shaping public opinion, trends, and cultural norms through its extensive reach and influence across different platforms and genres. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 19 + 261 + 586bccefb1e344289c1ee984e165de9c + + + METRIC + A metric indicating the highest level of development or achievement in a particular field + 322e02986c8724eedbcf3ebfa20b989c + 31 + 1 + 1 + 262 + a2201b8753ba4847ab0b22054e27d2c0 + + + METRIC + A metric indicating results that are comparable to or better than those of others in the same field + 322e02986c8724eedbcf3ebfa20b989c + 31 + 1 + 1 + 263 + b5ecd0553dd742f5813c9b855d548a41 + + + METRIC + A metric based on evaluations made by humans + 322e02986c8724eedbcf3ebfa20b989c + 31 + 1 + 1 + 264 + 89b2003e97804961805ea1886d078ebd + + + METRIC + Metrics that require a gold standard or reference answers for evaluation + 322e02986c8724eedbcf3ebfa20b989c + 32 + 1 + 2 + 265 + 6dd7f5f6b4544271a97f6a136f82fc3d + + + METHOD + An evaluation method that does not require reference answers + 322e02986c8724eedbcf3ebfa20b989c + 31 + 1 + 1 + 266 + eb01db8435554f2cbafe39a50f62f20a + + + METRIC + A metric that measures how relevant the generated text is to the given context + 322e02986c8724eedbcf3ebfa20b989c + 1 + 267 + 3d175ad1f0014cd4871eff4e86db9f88 + + + METRIC + A metric that measures how accurately the generated text reflects the source information + 322e02986c8724eedbcf3ebfa20b989c + 1 + 268 + c8e706fbdc90420d952deed03c4f04b4 + + + METRIC + A metric that measures how relevant the generated answer is to the question + 322e02986c8724eedbcf3ebfa20b989c + 1 + 269 + cf6115e69d6649cc99ef2bd11854ccfb + + + METHOD + A method involving multiple stages or steps + 322e02986c8724eedbcf3ebfa20b989c + 50 + 1 + 1 + 270 + 9ed7e3d187b94ab0a90830b17d66615e + + + DATA + The correct or ideal answers used as a benchmark in evaluations + 322e02986c8724eedbcf3ebfa20b989c + 32 + 1 + 2 + 271 + b4c7432f712849d7aba9dccbb77471ef + + + DATA + "SENSEMAKING QUESTIONS" are a class of questions used to evaluate the performance of Retrieval-Augmented Generation (RAG) systems. These questions are specifically designed to help users understand and make sense of complex information, as well as to validate the understanding and interpretation of data. By employing sensemaking questions, researchers and practitioners can assess how effectively a RAG system can retrieve and generate relevant information, thereby ensuring that the system aids in the comprehension and accurate interpretation of intricate datasets. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + 32 + 1 + 3 + 272 + 434e752b992c4e6a812557529315c5b9 + + + METHOD + A method where two items are directly compared against each other + 322e02986c8724eedbcf3ebfa20b989c + 1 + 273 + df79a27b9a4f42fd839c90bb8a79ad91 + + + DATA + TARGET METRICS are specific measures used to evaluate the performance of RAG systems. These metrics are aimed to be achieved or measured in the analysis and are the focus of an evaluation. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + 26 + 1 + 3 + 274 + 8f140fd7126f47b6b00307b0181509f9 + + + DATA + A metric used as a baseline or standard for comparison + 322e02986c8724eedbcf3ebfa20b989c + 2 + 275 + 40450f2c91944a81944621b94f190b49 + + + METRIC + A metric that measures the accuracy and reliability of a method or result + 322e02986c8724eedbcf3ebfa20b989c + 1 + 276 + 5b9fa6a959294dc29c8420b2d7d3096f + + + METRIC + A metric that measures the randomness or variability in a process + 322e02986c8724eedbcf3ebfa20b989c + 1 + 277 + b84d71ed9c3b45819eb3205fd28e13a0 + + + DATA + The average scores obtained from multiple evaluations + 322e02986c8724eedbcf3ebfa20b989c + 1 + 278 + b0b464bc92a541e48547fe9738378dab + + + PERSON + Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 53 + 1 + 6 + 279 + 44c65dda6fb7472dae36f6eea720ab47 + + + PERSON + Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 53 + 1 + 6 + 280 + 5d97ff82691c4482973d73d1860e4757 + + + PERSON + Britney Spears is a public figure frequently mentioned in entertainment articles, known for her significant contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 53 + 1 + 6 + 281 + 2567445079794d1e84f17abc48776002 + + + PERSON + Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his significant contributions to the music industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 53 + 1 + 6 + 282 + 392be891f8b649fabdc20e7bf549f669 + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in film and television + e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 1 + 283 + 0111777c4e9e4260ab2e5ddea7cbcf58 + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in music + e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 1 + 284 + 785f7f32471c439e89601ab81c828d1d + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in sports + e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 1 + 285 + 6768339b54084020aec27adcef8994ff + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in digital media and business + e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 1 + 286 + f09f381c319f4251847d1a4bb8cdcac1 + + + CATEGORY + A category of public figures in the entertainment industry who are involved in controversies + e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 1 + 287 + eec11f567e7f4943b157c3a657eb9a46 + + + METRIC + A metric used to determine the winner in the comparison of generated responses + e8c8f911135faf3ff35f24107eb3f99c + 53 + 1 + 3 + 288 + efef117839b64ce9adf614a461d41ba6 + + + METRIC + A metric used to evaluate the quality of LLM-generated responses + e8c8f911135faf3ff35f24107eb3f99c + 25 + 1 + 1 + 289 + 2171091ada0942d8ae7944df11659f6e + + + SECTOR + The entity "FILM" refers to a sector within the entertainment industry that encompasses movies and cinema. This sector includes public figures involved in the movie industry, such as actors, directors, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 2 + 290 + bcfdc48e5f044e1d84c5d217c1992d4b + + + SECTOR + The entity "TELEVISION" refers to a sector within the entertainment industry that encompasses TV shows and series. This sector includes public figures involved in TV shows, such as actors, hosts, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 2 + 291 + b232fb0f2ac14790b931d1e7fcddd8ad + + + SECTOR + MUSIC is a sector within the entertainment industry that encompasses musical performances and recordings. This sector includes public figures involved in the music industry, such as singers, musicians, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 53 + 1 + 6 + 292 + 1c16b22e18d3483b8d41b284754274e2 + + + SECTOR + The entity "SPORTS" refers to a sector within the entertainment industry that encompasses athletic events and competitions. This sector includes public figures involved in sports, such as athletes, coaches, and sports commentators. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 53 + 1 + 4 + 293 + 0080f96708cd4054a5f0986ca86889f4 + + + SECTOR + DIGITAL MEDIA is a sector within the entertainment industry that encompasses online content and social media. This sector includes public figures involved in online platforms, such as influencers, content creators, and digital marketers. These individuals play a significant role in shaping digital landscapes through their engagement with audiences and their ability to leverage various online tools and platforms for content dissemination and marketing purposes. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 2 + 294 + e683130322ac47708a852a5e51abb7c5 + + + CATEGORY + A category within the entertainment industry that includes stories and themes that shape culture + e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 1 + 295 + 71a0a8c1beb64da08124205e9a803d98 + + + CATEGORY + A category within the entertainment industry that includes popular movements and styles + e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 1 + 296 + f84314943bee4c859c9a62f268c9c216 + + + CATEGORY + A category within the entertainment industry that includes public conversations and debates + e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 1 + 297 + ba481175ee1d4329bf07757a30abd3a1 + + + CATEGORY + A category within the entertainment industry that includes formal discussions and communications + e8c8f911135faf3ff35f24107eb3f99c + 54 + 1 + 1 + 298 + 8d8da35190bf43c5878fa38f3eb4f3d2 + + + RESPONSE + Answer 1 provides a varied and rich response by covering a wide range of public figures from different sectors of the entertainment industry, including film, television, music, sports, gaming, and digital media. It offers insights into the contributions and influence of these figures, as well as controversies and their impact on public discourse. The answer also cites specific data sources for each mentioned figure, indicating a diverse range of evidence to support the claims. + 718017a4871c909420f84b85b8ba969d + 54 + 1 + 11 + 299 + 2fb7e14a3f124526bd7b24867fc18e81 + + + RESPONSE + "ANSWER 2" is a generated answer for the example question in the News article dataset. It focuses on a smaller group of public figures, primarily from the music industry and sports, and relies heavily on a single source for data. "ANSWER 2" provides concise explanations for the frequent mentions of specific public figures such as Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + 53 + 1 + 10 + 300 + 5c13c7d61e6c4bfe839f21e7ad3530a7 + + + METHOD + Naïve RAG is a baseline method for retrieval-augmented generation (RAG) that does not use advanced techniques. It is a basic form of RAG with certain drawbacks that advanced RAG systems aim to overcome. Naïve RAG is used to generate answers for questions in the News article dataset and to generate responses that directly list specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d,e4d9b12cf2b4c691c74019eefff4fb39,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19 + 26 + 1 + 4 + 301 + a621663edba64d99b7e50f1e53f32ee7 + + + DATASET + The "NEWS ARTICLE DATASET" is a collection of news articles utilized for various analytical purposes. This dataset is specifically employed for generating responses to questions about public figures in the entertainment industry, making it a valuable resource for both analysis and information retrieval tasks within this domain. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + 23 + 1 + 16 + 302 + 42be4e140061482ea509dd3e26189480 + + + TOPIC + Controversies are events or issues involving public figures that generate public debate and impact public discourse. + 718017a4871c909420f84b85b8ba969d + 54 + 1 + 2 + 303 + 4da4ef951ff340f1a3dd679de4be3341 + + + SECTOR + The gaming sector includes public figures involved in the gaming industry, including gamers, developers, and streamers. + 718017a4871c909420f84b85b8ba969d + 54 + 1 + 1 + 304 + 2f05fcce857e4a499ca4e89a3cefbcb3 + + + RESOURCE + Data sources are references or reports used to support claims about public figures and their influence. + 718017a4871c909420f84b85b8ba969d + 54 + 1 + 2 + 305 + b3aeb7ae009a4f52ae3ae4586e32fe11 + + + METHOD + Assessments generated by large language models (LLMs) to evaluate the answers produced by different methods + ebf5249c888e07fedce6572a4c03f88c + 23 + 1 + 1 + 306 + 089b9b9841714b8da043777e2cda3767 + + + DATASET + An example question used in the News article dataset for analysis + ebf5249c888e07fedce6572a4c03f88c + 23 + 1 + 1 + 307 + 38f1e44579d0437dac1203c34678d3c3 + + + DATA + The datasets used in the analysis, consisting of various text sources + 4c855404ee3d3c94aa2136f1513c666f + 2 + 308 + 1ca24718a96b47f3a8855550506c4b41 + + + METRIC + A metric used to compare the performance of different conditions in the analysis + 4c855404ee3d3c94aa2136f1513c666f + 1 + 309 + 9c980dfe3cab44b7a83408405edab0b6 + + + CATEGORY + A specific setup or scenario used in the analysis, such as C0, C1, C2, C3, TS, and SS + 4c855404ee3d3c94aa2136f1513c666f + 4 + 310 + f23484b1b45d44c3b7847e1906dddd37 + + + METRIC + WIN RATE is a measure used to evaluate the success rate of different methods in providing comprehensive and diverse answers. It represents the percentage of times a particular approach or method achieves a win in a given context. Specifically, it quantifies the percentage of times a condition outperformed another in the analysis. This metric is crucial in assessing the effectiveness of various strategies within the domain of Natural Language Processing and Information Retrieval, offering insights into the comparative performance of different techniques. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4 + 3 + 311 + 929f30875e1744b49e7b416eaf5a790c + + + METRIC + The condition that performed the best across all comparisons in the analysis + 4c855404ee3d3c94aa2136f1513c666f + 1 + 312 + 4920fda031804ce8a1073ace8e061ed6 + + + METRIC + The expected win rate of a condition when compared to itself, shown as 50% for reference + 4c855404ee3d3c94aa2136f1513c666f + 1 + 313 + 4b8aa4587c7344adac2cbfa69d5e40fa + + + METHOD + The use of large language models (LLMs) at the time of querying, evaluated in the analysis + 4c855404ee3d3c94aa2136f1513c666f + 39 + 1 + 1 + 314 + 52701d941dfb45359693baae8f267056 + + + METHOD + The "FINAL EVALUATION" is the last stage of the analysis where the best performing context window size was used. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + 39 + 1 + 3 + 315 + 31499ee6277a4d71b19cb5b6be554c69 + + + PROCESS + The process that resulted in the creation of graphs for the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + 21 + 1 + 1 + 316 + d99eabad5dfd47278692569d2a9395b1 + + + STRUCTURE + A data structure consisting of nodes and edges, used to represent the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + 21 + 1 + 4 + 317 + d53f15cb7f7845de91cc44ad44ff9f6e + + + METHOD + Approaches that consistently outperformed the naive RAG in comprehensiveness and diversity metrics + 36db32c37e1987e2c5863898ad882190 + 53 + 1 + 3 + 318 + 23becf8c6fca4f47a53ec4883d4bf63f + + + METRIC + The number of context units, such as community summaries or text chunks, used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + 20 + 1 + 4 + 319 + d0ffa3bcd1234258953ff4956d19f561 + + + METRIC + The term "TOKENS" refers to the number of individual words used in the analysis. The evaluation typically focuses on corpora in the region of 1 million tokens. This metric is crucial for understanding the scope and scale of the text data being analyzed, particularly in the fields of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,92e93fc6449756c0a60200636b297f65 + METRIC + 21 + 1 + 7 + 320 + ac41b77ba33c4c84877eb425aba03aa1 + + + METRIC + The percentage of the maximum token count used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + 21 + 1 + 4 + 321 + 5d3184dabfd647a5a7e565f72c60ff24 + + + METHOD + MAP-REDUCE SUMMARIZATION is a method for summarizing source texts using a map-reduce approach. This summarization technique is notably resource-intensive, necessitating the highest number of context tokens compared to other methods. The map-reduce framework, originally popularized for its efficiency in processing large-scale data, is adapted here to handle the complexities of text summarization, ensuring comprehensive and accurate extraction of key information from extensive source texts. + 36db32c37e1987e2c5863898ad882190,e4d9b12cf2b4c691c74019eefff4fb39 + 21 + 1 + 2 + 322 + 0ec262c2cfef4dd581f3655e5e496e31 + + + DATA + Summaries at the root level of the community hierarchy, requiring dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + 21 + 1 + 1 + 323 + 100c2fccd7f74d9281707082f062ba72 + + + DATASET + SOURCE TEXTS are the original texts from which summaries or analyses are derived. These texts serve as the foundational material used for comparison with community summaries in the analysis. + 6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39 + 55 + 1 + 2 + 324 + 378fc7636eeb4aabbfd40995a6960c64 + + + REFERENCE + A reference to a paper by Ram et al. in 2023 discussing RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + 26 + 1 + 1 + 325 + 80a04aa18cd649d584292f23b10c0727 + + + REFERENCE + "GAO ET AL., 2023" is a paper published in 2023 by Gao et al. that delves into advanced Retrieval-Augmented Generation (RAG) techniques, specifically where the index is a knowledge graph. The publication also touches upon naive RAG approaches, providing a comprehensive examination of both advanced and basic methodologies within the domain of Natural Language Processing and Information Retrieval. + 6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + 53 + 1 + 3 + 326 + 4e9ca18ccc1d4527a3bc035d07f5e162 + + + CATEGORY + Intermediate-level summaries are a type of community summary used in the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 21 + 1 + 1 + 327 + 5564257e89f1428486a64fcf52f49490 + + + CATEGORY + Root-level summaries are a type of community summary used in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 50 + 1 + 1 + 328 + 83c76fbd2a004d90a5b0a6736ffed61d + + + METRIC + Answer comprehensiveness is a measure used to evaluate the completeness of answers provided by different methods + 6f33a085ff3304e5994f7fbb86c881a4 + 50 + 1 + 1 + 329 + d9779c41e3c74fe0b26e23822a4b995b + + + TECHNOLOGY + Ad-hoc LLM use refers to the spontaneous use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + 25 + 1 + 1 + 330 + 9d7a563b3b2d405092c31f1fe08cff77 + + + TECHNOLOGY + Element extraction prompts are used to extract specific details in the Graph RAG index + 6f33a085ff3304e5994f7fbb86c881a4 + 50 + 1 + 1 + 331 + bd43f3d439a54781bd4b721a9a269b92 + + + CONCEPT, TECHNOLOGY + A mathematical space in which text chunks and queries are embedded to represent similar semantics + f35de4d9fb65f1d5a392064b20545c19 + 1 + 332 + adc0f95733e74351a891c4dadf650a52 + + + CONCEPT, DATA + Search inputs that are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + 1 + 333 + 225105a7be14447cb03186bd40756059 + + + TECHNOLOGY, METHOD + A type of RAG system that includes patterns for iterative and dynamic cycles of interleaved retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 26 + 1 + 1 + 334 + efce8a9d61254447a26aee99e53f0398 + + + TECHNOLOGY, CONCEPT + A concept related to generation-augmented retrieval that facilitates future generation cycles + f35de4d9fb65f1d5a392064b20545c19 + 56 + 1 + 2 + 335 + 4a75a9f0b18a48bea9c0601c0fc395c4 + + + TECHNOLOGY, METHOD + A method that facilitates future generation cycles by using self-memory + f35de4d9fb65f1d5a392064b20545c19 + 56 + 1 + 1 + 336 + e19287afe00a431f9a593a4827d1b448 + + + TECHNOLOGY, METHOD + A strategy for iterative retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 50 + 1 + 1 + 337 + f2c06f3a0c704296bf3353b91ee8af47 + + + TECHNOLOGY, METHOD + A federated strategy for retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 50 + 1 + 1 + 338 + f512103ed4624accac6cbbf90d7d250a + + + TECHNOLOGY, METHOD + A method that combines multiple concepts for summarizing multiple documents + f35de4d9fb65f1d5a392064b20545c19 + 58 + 1 + 2 + 339 + 2325dafe50d1435cbee8ebcaa69688df + + + TECHNOLOGY, METHOD + A method for answering questions that require multiple steps or "hops" to gather information + f35de4d9fb65f1d5a392064b20545c19 + 4 + 340 + 469aeef98cd1421fa123277b93d7b83a + + + TECHNOLOGY, METHOD + An approach that involves generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 52 + 1 + 2 + 341 + 2fb66f9a0de6406d83b61742a3b52cd6 + + + TECHNOLOGY, METHOD + A method for generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 52 + 1 + 1 + 342 + b0e6cfd979ea48b997019b059999d3c2 + + + TECHNOLOGY, METHOD + A method for answering multiple interpretations of ambiguous questions by generating a hierarchical structure + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 50 + 1 + 1 + 343 + ef00ec3a324f4f5986141401002af3f6 + + + TECHNOLOGY, METHOD + A process that involves using LLMs to create knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 25 + 1 + 1 + 344 + a542fd7aed7341468028928937ea2983 + + + TECHNOLOGY, METHOD + A process that involves using LLMs to complete existing knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 25 + 1 + 1 + 345 + 1c5e296a5ac541c1b5cac4357537c22d + + + TECHNOLOGY, METHOD + Graphs that represent causal relationships, which can be extracted using LLMs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 25 + 1 + 1 + 346 + 5ecf534a9ffe46e0b1c2144110c691c0 + + + REFERENCE, PUBLICATION + A reference to a publication by Cheng et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 50 + 1 + 1 + 347 + 4d183e7007624fcd98af96b9d752c16d + + + REFERENCE, PUBLICATION + A reference to a publication by Mao et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 50 + 1 + 1 + 348 + 718c507cb8ac49e6a35c251ac951b5ca + + + REFERENCE, PUBLICATION + A reference to a publication by Shao et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 50 + 1 + 1 + 349 + b45ef27279c043269b23b894461d7d8c + + + REFERENCE, PUBLICATION + A reference to a publication by Wang et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 50 + 1 + 1 + 350 + 10983a248cc448c59c94df4d1d0898f0 + + + REFERENCE, PUBLICATION + A reference to a publication by Su et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 50 + 1 + 1 + 351 + e2ec7d3cdbeb4dd086ae6eb399332363 + + + REFERENCE, PUBLICATION + A reference to a publication by Feng et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 50 + 1 + 1 + 352 + 67f10971666240ea930f3b875aabdc1a + + + REFERENCE, PUBLICATION + A reference to a publication by Trivedi et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 50 + 1 + 1 + 353 + 8b95083939ad4771b57a97c2d5805f36 + + + REFERENCE, PUBLICATION + A reference to a publication by Khattab et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 50 + 1 + 1 + 354 + 3c4062de44d64870a3cc5913d5769244 + + + REFERENCE, PUBLICATION + A reference to a publication by Sarthi et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 50 + 1 + 1 + 355 + 24652fab20d84381b112b8491de2887e + + + REFERENCE, PUBLICATION + A reference to a publication by Kim et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 50 + 1 + 1 + 356 + d4602d4a27b34358baa86814a3836d68 + + + REFERENCE, PUBLICATION + "TRAJANOSKA ET AL., 2023" refers to a paper by Trajanoska et al. published in 2023, which focuses on using Large Language Models (LLMs) for knowledge graph creation. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting innovative methodologies for leveraging advanced language models to construct and enhance knowledge graphs. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 25 + 1 + 2 + 357 + 36be44627ece444284f9e759b8cd25c6 + + + REFERENCE, PUBLICATION + "Yao et al., 2023" refers to a paper published by Yao and colleagues in 2023. The study focuses on the application of large language models (LLMs) for the task of knowledge graph completion. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting the potential of advanced LLMs to enhance the accuracy and efficiency of knowledge graph completion processes. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 25 + 1 + 2 + 358 + a64b4b17b07a44e4b1ac33580d811936 + + + REFERENCE, PUBLICATION + A reference to a publication by Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 25 + 1 + 1 + 359 + 423b72bbd56f4caa98f3328202c1c3c9 + + + TECHNOLOGY, METHOD + A system that combines multiple concepts for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 58 + 1 + 1 + 360 + 5c7ef01f46a94641bf1ae5cd25f8a538 + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 1 + 361 + aefde1f7617f4c0e9aed31db77f6d862 + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 1 + 362 + ad52ba79a84748a49067e53b1d5095f9 + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 1 + 363 + 289616058bf4495887292003b27ba216 + + + TECHNOLOGY, METHOD + Strategies used before the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + 0 + 364 + 7ffa3a064bce468082739c5a164df5a3 + + + TECHNOLOGY, METHOD + Strategies used during the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + 0 + 365 + ce36d1d637cf4a4e93f5e37ffbc6bd76 + + + TECHNOLOGY, METHOD + Strategies used after the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + 0 + 366 + eeb9c02c0efa4131b9e95d33c31019fc + + + TECHNOLOGY, METHOD + A pattern in Modular RAG systems for iterative and dynamic cycles of retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 0 + 367 + 7b2472c5dd9949c58828413387b94659 + + + TECHNOLOGY, METHOD + Cycles of generation that are facilitated by self-memory in Graph RAG + f35de4d9fb65f1d5a392064b20545c19 + 0 + 368 + bdddcb17ba6c408599dd395ce64f960a + + + PUBLICATION + A paper by Ban et al. published in 2023, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 26 + 1 + 1 + 369 + bc70fee2061541148833d19e86f225b3 + + + PUBLICATION + A paper by Zhang et al. published in 2024, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 26 + 1 + 1 + 370 + 0fc15cc3b44c4142a770feb4c037a6f7 + + + METHOD + A method where the index is a knowledge graph, developed by Baek et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 27 + 1 + 2 + 371 + a24e9df02e1b4b43bf6324b039e28285 + + + PUBLICATION + A paper by Baek et al. published in 2023, focusing on the KAPING method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 27 + 1 + 1 + 372 + ab3a5a6713244fd595a1ace978c3d960 + + + METHOD + A method where subsets of the graph structure are the objects of enquiry, developed by He et al. in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + 29 + 1 + 2 + 373 + 02a88c0d128e4586b2f1f64329786d3c + + + PUBLICATION + A paper by He et al. published in 2024, focusing on the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 29 + 1 + 1 + 374 + 1ca41537c47c4752a17a44d1d7086d96 + + + METHOD + A method where derived graph metrics are the objects of enquiry, developed by Zhang in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 34 + 1 + 2 + 375 + 7e0d14ca308b4796bdc675a64bd3a36e + + + PUBLICATION + A paper by Zhang published in 2023, focusing on the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 34 + 1 + 1 + 376 + 8323efc8e539419e9ca3c98e758f6609 + + + METHOD + A method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, developed by Kang et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 30 + 1 + 2 + 377 + a80c7c98c0b647f8b9f6f8cc09168e44 + + + PUBLICATION + A paper by Kang et al. published in 2023, focusing on the SURGE method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 30 + 1 + 1 + 378 + 2d66a15939294d21b83b3e277f0a4e46 + + + METHOD + A method where retrieved event-plot subgraphs are serialized using narrative templates, developed by Ranade and Joshi in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 28 + 1 + 2 + 379 + 47f6d6573cf34e1096c95e36251dd60c + + + PUBLICATION + A paper by Ranade and Joshi published in 2023, focusing on the FABULA method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 28 + 1 + 1 + 380 + 2fbd74d5ccca4be99c5257b3ac95cfba + + + PUBLICATION + A paper by Wang et al. published in 2023, focusing on a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 26 + 1 + 1 + 381 + a2b1621a3e424ae29a6a73f00edbeca3 + + + ORGANIZATION + LangChain is an organization that developed Langchain graphs and supports a variety of graph databases. + 71f6daf11e64e5273a3847d46bf228e1,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + 5 + 382 + ec45e1c400654c4f875046926486ded7 + + + ORGANIZATION + LlamaIndex is an organization that developed the LlamaIndex Knowledge Graph Index and supports a variety of graph databases. + 6cd82819982879bd164547d2773ba5c7,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + 3 + 383 + 047cd93e9d704c7d8dadb6e79f9458df + + + TECHNOLOGY + Neo4J is both a graph database format supported by various Retrieval-Augmented Generation (RAG) applications and an organization that developed Project NaLLM. The graph database format of Neo4J is widely recognized for its efficiency in handling complex relationships and structures, making it a valuable tool in the field of Natural Language Processing and Information Retrieval. As an organization, Neo4J has contributed significantly to the advancement of these domains through innovative projects like NaLLM, which further underscores its pivotal role in the community. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + 4 + 384 + 5b71ee73a5b6484495b2a0a75219426c + + + METHOD + A method that can create and reason over knowledge graphs in Neo4J format, developed by Neo4J in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + 1 + 385 + e1f524d4b9754ce2b64a0a4c8f73b854 + + + TECHNOLOGY + NebulaGraph is both a graph database format and an organization that has made significant contributions to the field of graph databases and retrieval-augmented generation (RAG) applications. As a graph database format, NebulaGraph is supported by various RAG applications, facilitating the efficient handling and querying of complex graph data structures. Additionally, NebulaGraph, as an organization, has pioneered the industry-first graph RAG, which integrates retrieval-augmented generation with large language models (LLMs) based on knowledge graphs. This innovation underscores NebulaGraph's role in advancing the capabilities of knowledge graph-based applications and enhancing the performance of LLMs in generating contextually relevant information. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + 4 + 386 + ae1fe1c014c54ec4bcdf10dbdaed5068 + + + METHOD + A method that can create and reason over knowledge graphs in NebulaGraph format, developed by NebulaGraph in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + 1 + 387 + 92646910ee624bd7909fac2b5c0232e3 + + + METHOD + A method for comparing fabrication rates, developed by Manakul et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 1 + 388 + 05913bee89a94bca88449249e35ba74d + + + PUBLICATION + "MANAKUL ET AL., 2023" refers to a paper by Manakul et al. published in 2023, which focuses on the SelfCheckGPT method. This work by Manakul and colleagues is centered around the development and application of SelfCheckGPT, a technique likely aimed at enhancing the performance and reliability of GPT models. The paper contributes to the field of Natural Language Processing and Information Retrieval by addressing specific challenges and proposing innovative solutions through the SelfCheckGPT method. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + PUBLICATION + 2 + 389 + 57b8930790c34dcba4a32c6be703ed78 + + + STAKEHOLDER + END USERS are individuals who are the final users of the system or analysis. They play a crucial role in validating sensemaking questions and target metrics, ensuring that the system or analysis meets the intended objectives and provides meaningful insights. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + STAKEHOLDER + 32 + 1 + 2 + 390 + 838c4498bc3c437f8d65428b580766a2 + + + CONCEPT + Considerations and compromises involved in building a graph index + 92e93fc6449756c0a60200636b297f65 + CONCEPT + 26 + 1 + 1 + 391 + 1b893f24eb98477aad6ce49c0f26737e + + + METRIC + The effectiveness of RAG systems, which varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + METRIC + 26 + 1 + 1 + 392 + 6573bc2af4f94596a3f4452a602d6fc4 + + + CONCEPT + Various forms of data used in RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + 26 + 1 + 1 + 393 + 0dddcca0e5df4b16bc03a51a2d2d8e16 + + + METRIC + The scale of datasets used in RAG systems, which affects performance + 92e93fc6449756c0a60200636b297f65 + METRIC + 26 + 1 + 1 + 394 + df40ad480a3c47299a6c8fad05349304 + + + PROCESS + The process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + PROCESS + 26 + 1 + 1 + 395 + fe98fb197d294b0b837aee8d5a98dfb1 + + + DATASET + Collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + DATASET + 26 + 1 + 1 + 396 + feb9ddd0ac2949178f26a36949aa5422 + + + CONCEPT + Different categories of questions used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + 26 + 1 + 1 + 397 + b4e4fa2e3dfc46e68d532d659b18d17d + + + METHOD + SelfCheckGPT is an approach used to compare fabrication rates in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 398 + f58813d090b947a48c1b4614b92c3ec3 + + + METHOD + A method for global summarization of source texts that does not use a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + 55 + 1 + 2 + 399 + 30a251bc3d04430d82b5a1a98c7b8c75 + + + RESOURCE + The amount of computational resources allocated for a task + e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 400 + 93e1d19f9bfa4c6b8962d56d10ea9483 + + + METRIC + The expected number of queries over the lifetime of a dataset + e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 401 + 8046335ba70b434aa3188392a746fd78 + + + DATA + Annotations that provide detailed information about the text + e4d9b12cf2b4c691c74019eefff4fb39 + 50 + 1 + 1 + 402 + 5c02b1ab32064c64a0f8b27b219e358a + + + METHOD + A method that uses embeddings to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + 24 + 1 + 3 + 403 + c5f77ba0c261408780db3d50346f16b7 + + + METHOD + RAG schemes that combine embedding-based matching with other approaches + e4d9b12cf2b4c691c74019eefff4fb39 + 51 + 1 + 2 + 404 + 453ecf5476f64f4a8d5020b95baf1314 + + + METHOD + Mechanisms used in map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + 57 + 1 + 2 + 405 + 6a1d83c9ce2b483dbd7de5ab3ae2487d + + + DATA + A hierarchical organization of communities + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 406 + 66c3dffb7d7a4fa8bb6b48a22ca917a6 + + + METHOD + A method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) for global text summarization + e4d9b12cf2b4c691c74019eefff4fb39 + 33 + 1 + 2 + 407 + 6f3dd1fd6d7f4df4af0656ed0525c92e + + + METRIC + The cost associated with the number of tokens used in a text generation task + e4d9b12cf2b4c691c74019eefff4fb39 + 33 + 1 + 1 + 408 + 711eb39432794b0a91110358dd536517 + + + TECHNOLOGY + An implementation of Graph RAG approaches using the Python programming language + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 409 + 0e00585b08044954a254116665400463 + + + PERSON + A person who contributed to the work mentioned in the acknowledgements + e4d9b12cf2b4c691c74019eefff4fb39 + 50 + 1 + 1 + 410 + db0147eff2204a20b5e5e6bec7a8bae5 + + + METRIC + The rates at which fabrications occur in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 411 + 67bb4f4678284819add02ba04f3b1103 + + + METRIC + The expected number of queries over the lifetime of a specific dataset + e4d9b12cf2b4c691c74019eefff4fb39 + 22 + 1 + 1 + 412 + 2033ec0487f04240abb3bdbe77b39087 + + + METRIC + The benefits or value obtained from using a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + 22 + 1 + 1 + 413 + f026fab8fec948ae9e7baa2ad715e6ef + + + METHOD + Different methods related to retrieval-augmented generation that utilize graph structures + e4d9b12cf2b4c691c74019eefff4fb39 + 22 + 1 + 1 + 414 + d0d7ed36d6f54b5d986dfd854096b728 + + + METHOD + Graph RAG approaches that operate in a more localized manner + e4d9b12cf2b4c691c74019eefff4fb39 + 50 + 1 + 1 + 415 + bf6a4c18f44042799eb7456a6b85b54a + + + DATA + Annotations made on the graph to provide additional information + e4d9b12cf2b4c691c74019eefff4fb39 + 24 + 1 + 1 + 416 + fac4a59c2278498d83f9f1b4231ad62e + + + DATA + Reports generated from community summaries + e4d9b12cf2b4c691c74019eefff4fb39 + 51 + 1 + 1 + 417 + d6d2b5862ddc4c4d87deee3423506817 + + + METHOD + An operation that aggregates information across multiple levels of a hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + 57 + 1 + 1 + 418 + 47d588d26e2b4cccb68fe2af4c147c8f + + + METHOD + A mechanism that allows for exploring detailed information by following higher-level summaries + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 419 + c0f2dc03d8df400db4997c1a0babd6ad + + + DATA + The trail of information that guides users to more detailed data + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 420 + 0211d61aae834229a3a1e004ff5cc658 + + + DATA + The top-level communities in a hierarchical structure + e4d9b12cf2b4c691c74019eefff4fb39 + 66 + 1 + 1 + 421 + ccbbbcc055c34709abcf103208c2c299 + + + DATA + A graph index organized around entities + e4d9b12cf2b4c691c74019eefff4fb39 + 50 + 1 + 1 + 422 + 989add81cf874018a569239b68d17ff2 + + + TECHNOLOGY + A publicly available implementation of a technology + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 423 + fd7d94fbab084bc380480abeef6bfade + + + PERSON + Alonso Guevara Fernández is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 21 + 424 + cfb915c95caf41c6a25e99a9f37f03a2 + + + PERSON + Amber Hoak is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 19 + 425 + 8815ed80f9b741dbb458d902024f34a4 + + + PERSON + Andrés Morales Esquivel is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 426 + dddb831546354e088d29aebd154e3a31 + + + PERSON + Ben Cutler is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 427 + 005d2154da754b21adcd90ac921bd5f7 + + + PERSON + Billie Rinaldi is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 428 + 711ba818354546cea69f1532b92a2f26 + + + PERSON + Chris Sanchez is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 429 + 5c4d8a8f9c104176b87d2bfdf04ae0bd + + + PERSON + Chris Trevino is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 430 + 5a781604f1fb4719b730f43f534627f6 + + + PERSON + Christine Caggiano is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 431 + ecdc1020b10e49ca869d399825e16fa3 + + + PERSON + David Tittsworth is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 432 + 0d8fde01d7234726a00d7e73e2e01d66 + + + PERSON + Dayenne de Souza is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 433 + 9c4bd60958fd4e09a6d5b9e2ab163b5a + + + PERSON + Douglas Orbaker is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 434 + 39d31f770cf740e78d526a2e1101a1db + + + PERSON + Ed Clark is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 435 + 9d282b2250f7408888504f1f93c202a8 + + + PERSON + Gabriel Nieves-Ponce is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 436 + c063484895794a0eaae1b0ff070ad4c9 + + + PERSON + Gaudy Blanco Meneses is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 437 + e8868920e21b4431aad16e86db977ecb + + + PERSON + Kate Lytvynets is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 438 + aea3378bfff842e5b3f4b7a4b55b3879 + + + PERSON + Katy Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 439 + d562223c17d948bf98e34b4d97dde932 + + + PERSON + Mónica Carvajal is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 440 + cde2d75c51d245879265b79d14b8699b + + + PERSON + Nathan Evans is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 441 + 44594467054849d4a1fadb46ddd51641 + + + PERSON + Richard Ortega is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 442 + 2918130221f94f4387da049b647bfe6a + + + PERSON + Rodrigo Racanicci is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 443 + fd139ac75b0e4777ab67b7423eaaa37f + + + PERSON + Sarah Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1 + 444 + a701c349eb7142d48ba7efad89caf9d2 + + + PERSON + Shane Solomon is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1 + 445 + e5d40a1b17f74b1db5d18279caedb04a + + + PUBLICATION + A technical report on GPT-4 published as an arXiv preprint + 086021a89900a39bcb62036981737bfa + 0 + 446 + de25d06733d04385825ee082792f5e52 + + + METHOD + A method for zero-shot knowledge graph question answering described in an arXiv preprint + 086021a89900a39bcb62036981737bfa + 0 + 447 + 32f6f11a7845416b8c6eb9fb0b382140 + + + METHOD + A method for harnessing large language models for advanced causal discovery from data + 086021a89900a39bcb62036981737bfa + 0 + 448 + 91407be8c3e54e23918d3a7183d962db + + + METHOD + A method incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models + 086021a89900a39bcb62036981737bfa + 0 + 449 + 3831134696584d83bbf676a6b3bfa8f9 + + + PERSON + J. Achiam is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 450 + 50e512a5dbe941f5af68bfdf74b1c3c0 + + + PERSON + S. Adler is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 451 + edc717747e904728b57185f5013461f9 + + + PERSON + S. Agarwal is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 452 + 8fba1fea719d49d380ac2d9c310d68b3 + + + PERSON + L. Ahmad is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 453 + 532da08f04f645708e747c57e9c4ee05 + + + PERSON + I. Akkaya is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 454 + 3cf0ab4cf14e47ddabd49d500a3dc488 + + + PERSON + F. L. Aleman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 455 + a39b72f8921f43ef8ef295c7cc8f7294 + + + PERSON + D. Almeida is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 456 + 9f5adbeb6cf04f089abe78d86cfa6aba + + + PERSON + J. Altenschmidt is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 457 + efb6350e65964659bc20396c0166b296 + + + PERSON + S. Altman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 458 + e095cc36da784300b27c6f8c60a96440 + + + PERSON + S. Anadkat is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 459 + c68893ca39d74ba08c6eb138f24441e1 + + + PERSON + R. Anil is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 460 + 472b23bb92834173b4118d101040c726 + + + PERSON + S. Borgeaud is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 461 + 81869985b45a4fefbbbb23ea118a3de4 + + + PERSON + Y. Wu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 462 + 42b8584c5a874eb08fbd61f0c18f3ca0 + + + PERSON + J.-B. Alayrac is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 463 + 824d93d9840a4b7c8b1f31bc6816b497 + + + PERSON + J. Yu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 464 + f209a808f1f04a5699601e672f4abd06 + + + PERSON + R. Soricut is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 465 + ccb335166f6c4564ac1c61549d8ded50 + + + PERSON + J. Schalkwyk is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 466 + cbe1a41a82aa4f268e8264568b25938f + + + PERSON + A. M. Dai is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 467 + 28e7639f55ce464c8a080cbb2c745fa2 + + + PERSON + A. Hauth is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 468 + 3f3a2d7aa1294116814f0b4d89baa23d + + + PERSON + J. Baek is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 2 + 469 + 3073b33926bd4f33807ffa3befacefaf + + + PERSON + A. F. Aji is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 2 + 470 + 2b916117691c4872a9c4e4888d4fe4ab + + + PERSON + A. Saffari is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 2 + 471 + 1f7b02bf486e4f42b23e9cb1a63207f3 + + + PERSON + T. Ban is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 1 + 472 + e744c118ae7f4638a01d060bbaedd6e9 + + + PERSON + L. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 1 + 473 + e1c1080c717d437996def1a41772d179 + + + PERSON + X. Wang is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 0 + 474 + 63fba9a7c47a4f14ac0bee6bc90d0fea + + + PERSON + H. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 0 + 475 + 6bfc2395b4f54a528a1ebac94a43acb8 + + + PERSON + T. Baumel is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 476 + 1cce5cebf437428eb1a60dffbdfa603f + + + PERSON + M. Eyal is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 477 + dc94039d6643460ca3c66150b9087129 + + + PERSON + M. Elhadad is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 478 + f197d75f159943f8a3ff441199790bc7 + + + PUBLICATION + The arXiv preprint identifier for the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0 + 479 + 4d8890c699684c9381105b03b0b41b03 + + + PUBLICATION + The arXiv preprint identifier for the Gemini paper + 086021a89900a39bcb62036981737bfa + 0 + 480 + b1658adfa43847eabad1437db235e858 + + + PUBLICATION + The arXiv preprint identifier for the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 0 + 481 + a1773cac7d4c4939aec965660e5015fe + + + PUBLICATION + The arXiv preprint identifier for the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 482 + 6a054cb59fb44cf494b93988b5f88833 + + + PERSON + Baumel, T. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 44 + 1 + 4 + 483 + e7b103a52e384e3e8bf14105223e7e82 + + + PERSON + Eyal, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 44 + 1 + 4 + 484 + 3f1042452c254cecaf7189e89162adc8 + + + PERSON + Elhadad, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 44 + 1 + 4 + 485 + fd31d549420744d1bd1a6b1112a9a6ba + + + PERSON + Blondel, V. D. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 486 + f7ab348030714072a277682b51f7c588 + + + PERSON + Guillaume, J.-L. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 487 + 2139b0906dc541e094138a978d070416 + + + PERSON + Lambiotte, R. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 488 + ff5466607e5d4453b1d833629292f664 + + + PERSON + Lefebvre, E. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 489 + 71f95003936e46a98d90757ffd845d40 + + + PUBLICATION + The journal where the paper "Fast unfolding of communities in large networks" was published + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 490 + bada987ea7da4c939393ee1c3d08ccd4 + + + PERSON + Brown, T. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9 + 491 + d0a274e7934d446fb91847bb53a961a6 + + + PERSON + Mann, B. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9 + 492 + 0a799eab61bc4e6b884db6689f9c2c4a + + + PERSON + Ryder, N. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9 + 493 + 8c34cd494a63438dac219c1dc0f73100 + + + PERSON + Subbiah, M. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 8 + 494 + c6f428af0c5e4f629902fd5455bf19ac + + + PERSON + Kaplan, J. D. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 495 + d1fd271d16c348019c2fcced762b35a2 + + + PERSON + Dhariwal, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 496 + ffa128c9c0c84d39bad1bba8cfa4adc5 + + + PERSON + Neelakantan, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 497 + 058f66cc356b43cc9433bd3c8d57fa46 + + + PERSON + Shyam, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 498 + ff74091eaba246698fcae59c21eec828 + + + PERSON + Sastry, G. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 499 + f6cbbf1b8f4b48a28a16e4dd8976b9bb + + + PERSON + Askell, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 500 + 757ca40654d5476aa949a26b733be8d4 + + + PUBLICATION + "ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS" is a prominent conference where significant papers in the field of Natural Language Processing and Information Retrieval are presented. Notable papers presented at this conference include "Language models are few-shot learners" and "Retrieval-augmented generation for knowledge-intensive NLP tasks." Additionally, it is also the journal where the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" was published. This conference and journal serve as key platforms for disseminating cutting-edge research in neural information processing systems. + 58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,b1bbda43309e8e0e2175ea034aa88e13 + 0 + 501 + 539d55e7c42e44b59d98f59fae3e0ee1 + + + PERSON + Cheng, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 502 + 3785eeadea9042bfb2e50f16c0397a12 + + + PERSON + Luo, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 503 + 48cd97f2297143e09d61ff2a8542c0c5 + + + PERSON + Chen, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 504 + ff95eb0d5f7f49b782027d5c7ae3c3fe + + + PERSON + Liu, L. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 505 + 086da554db5b4ad5806aedeb0024197c + + + PERSON + Zhao, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory"Zhao, D. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + PERSON + 45 + 1 + 2 + 506 + 216ee8a907a0466a88b27f8ada19ffa0 + + + PERSON + Yan, R. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 507 + 6fefb317687d4ac98efe39a52f3e190f + + + PERSON + Dang, H. T. is an author of the paper "Duc 2005: Evaluation of question-focused summarization systems" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 508 + 320d9d91238948a8be67972ccceab878 + + + PUBLICATION + The conference where the paper "Duc 2005: Evaluation of question-focused summarization systems" was presented + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 509 + bdcbcccadd474b3bbe9a8f56c811bab4 + + + PERSON + Es, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 42 + 1 + 2 + 510 + f127fc4d87f94794be89134406ba0694 + + + PERSON + James, J. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 42 + 1 + 2 + 511 + c27966a4e3be434686454204ac7b3ab4 + + + PERSON + Espinosa-Anke, L. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 42 + 1 + 2 + 512 + dab39f92d0ed468c80699f28c05c45fa + + + PERSON + Schockaert, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 42 + 1 + 2 + 513 + 3076f330d121489aa50964ce54a3b1ac + + + PERSON + Feng, Z. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 45 + 1 + 2 + 514 + c8e5d3afdcb54c8589e280f0c4a87417 + + + PERSON + Feng, X. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 45 + 1 + 2 + 515 + f3d30627e19245649e497ab49bf0fa30 + + + PERSON + Yang, M. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 45 + 1 + 2 + 516 + e3f1098c3d984bc7b5f30b9c0101f7a6 + + + PERSON + Qin, B. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 45 + 1 + 2 + 517 + 24b4a5f4db67418cbfa08c5316f0ab51 + + + PERSON + Fortunato, S. is an author of the paper "Community detection in graphs" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 518 + e4b707e3e6964197855b82fc66ef59e7 + + + PUBLICATION + The journal where the paper "Community detection in graphs" was published + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 519 + 109b8be5a8ee4180a1465cd23f019d7b + + + PERSON + Gao, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 520 + 49f771e31a0c4b35bc39e389f3623509 + + + PERSON + Xiong, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models. The paper provides a comprehensive survey of the methodologies and applications of retrieval-augmented generation, highlighting its significance in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 521 + aa946d4379694a74ba0da37e69d2810a + + + PERSON + Gao, X. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 522 + 268446fc52a54fd2837f73aeb3e0b74f + + + PERSON + Jia, K. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant area of research within the domains of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 523 + f6ddfa8491ff40d2839bb5b2e105df22 + + + PERSON + Pan, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 524 + db1295504da645b69d9786d54f233fed + + + PERSON + Bi, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 525 + 6ff4ed0dda4f4158af37be99f505565f + + + PERSON + Dai, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance and capabilities of large language models, a significant area of research within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 526 + 5d398b88ee4242a59c32feb188683ec3 + + + PERSON + Sun, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 527 + 0a784e00c9464bd3aeb830b908f73170 + + + PERSON + Wang, H. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 528 + b0966a0f455e44229e6c9705d57bfca9 + + + PUBLICATION + The arXiv identifier for the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 44 + 1 + 3 + 529 + 99761e9b89cc4060be3ed6b34532e7ff + + + PUBLICATION + The arXiv identifier for the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 42 + 1 + 4 + 530 + 8130a1a82bde46048952cf147690e630 + + + PUBLICATION + The arXiv identifier for the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 45 + 1 + 5 + 531 + 79c99026b7ef4946b9b8e0be841fd4c5 + + + PERSON + Goodwin, T. R. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 2 + 532 + fdcb1673254842f1935f53d0c38c467e + + + PERSON + Savery, M. E. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 1 + 533 + dcb3f4cc8abc46faabc193d9885e91d0 + + + PERSON + Demner-Fushman, D. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 1 + 534 + 3295be59128d451bb720c6688adc1e0b + + + CONFERENCE + COLING (International Conference on Computational Linguistics) is the conference where the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" was presented + 00e8e4e881bd0862022f4dfc913b900b + 0 + 535 + aca3eb8924ac494486fe0bfe892f7f2e + + + PERSON + He, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 536 + 66689accdd974295b7eb779e43578748 + + + PERSON + Tian, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 537 + 6b49c78aa1524609ab7aa74aeaa3e01d + + + PERSON + Sun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 538 + 7ff31ce54f424f0bbb297b0b3ba7c757 + + + PERSON + Chawla, N. V. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 539 + bac51e00d486420c8e91e824d8e17411 + + + PERSON + Laurent, T. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 540 + 4adee3aad6524a4aa4c4711c1ee05e64 + + + PERSON + LeCun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 541 + d034e4fd8ac849278e658daad1a1f033 + + + PERSON + Bresson, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 542 + 091e998370dd42d1b05ab0fcf6595a7e + + + PERSON + Hooi, B. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 543 + 1e6cabc18fab4c048281fd29d3044438 + + + PERSON + Jacomy, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 544 + dc08f6d7398b4b798a3bdccf508a2ad4 + + + PERSON + Venturini, T. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 545 + 1c7fd5af8d8041e186eae2431fc627cd + + + PERSON + Heymann, S. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 546 + b16eda56dcec40f2b3e109fb9246bee3 + + + PERSON + Bastian, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 547 + 43c68f9a86654a32a2215e23957ed184 + + + PUBLICATION + PLOS ONE is the journal where the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" was published + 00e8e4e881bd0862022f4dfc913b900b + 0 + 548 + 1ba06fe2e86140a59bbc4f4e969d0f71 + + + PERSON + Jin, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 549 + 36caa0a230c8422c8acb4dc62e35bb32 + + + PERSON + Yu, Z. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 550 + 09940fed9d154504948bba2df1789a50 + + + PERSON + Jiao, P. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 551 + 4d6608557eed49368a6d09c7c5c664c5 + + + PERSON + Pan, S. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 552 + eb7c93eeb9dc41aab57d29e97ebb4951 + + + PERSON + He, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 553 + 3b6e2ac584b64847b53828c9d779fed3 + + + PERSON + Wu, J. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 554 + e9b68002e035447baae848208cea5503 + + + PERSON + Philip, S. Y. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 555 + fe18353546824ca98294ce4be7b96e02 + + + PERSON + Zhang, W. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 556 + 0e9740e25f5a460c81318336e00ac880 + + + PUBLICATION + IEEE Transactions on Knowledge and Data Engineering is the journal where the paper "A survey of community detection approaches: From statistical modeling to deep learning" was published + 00e8e4e881bd0862022f4dfc913b900b + 0 + 557 + b7cd9a62710849778fdadced0d754687 + + + PERSON + Kang, M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 558 + 432a6b4962544200949421a96a405142 + + + PERSON + Kwak, J. M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 559 + d6700b360ac141d282cdb567414bf4ce + + + PERSON + Baek, J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 560 + c1b40a4039b44061a358e098867f7412 + + + PERSON + Hwang, S. J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 561 + 4643a7a319674adfb732b6f6122c7c64 + + + PERSON + Khattab, O. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 562 + 46e8056fb2ec4811ab33cb34a0dc9fb3 + + + PERSON + Santhanam, K. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 563 + 8b57a9f43a1942a49b58cf881835f974 + + + PERSON + Li, X. L. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 564 + f78b01b0d93948c283644ec58f7be74a + + + PERSON + Hall, D. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text, indicating its relevance within the domain of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 565 + 8dbe8f9867e4448f998416c18923eac4 + + + PERSON + Liang, P. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Liang, P. contributed to the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP," which explores the integration of retrieval and language models to enhance knowledge-intensive tasks in NLP. Additionally, Liang, P. authored the paper "Lost in the middle: How language models use long contexts," which investigates the utilization of extended contexts by language models. These contributions highlight Liang, P.'s significant role in advancing the understanding and application of language models in complex NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 7 + 566 + fe8ea8bf1395434393e04e8f7a33025f + + + PERSON + Potts, C. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 567 + 7d58b089bfc549e8951e91ad62541119 + + + PERSON + Zaharia, M. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 568 + 1fa6d3118bd846c8837b5fa9fb78f262 + + + PERSON + Kim, G. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 569 + 62c65bbae33c4ee9a21b61f6f454c4b4 + + + PERSON + Kim, S. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 570 + 30b7034c4468473f98ee18d00ee73b33 + + + PERSON + Jeon, B. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 571 + 00f78b85e5b84999a810e311e540037b + + + PERSON + Park, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 572 + 3e460d9f011d4b0b9ccaae7b6a5202de + + + PERSON + Kang, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 573 + 9d98dece22eb401aa1a5ce9c88c603f0 + + + PERSON + Klein, G. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 574 + 81446ea789b24eaf9eab02dc07c3d984 + + + PERSON + Moon, B. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 575 + 79f4b1c1b2be4cf7aa828846e20a4eb6 + + + PERSON + Hoffman, R. R. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 576 + de04830d6e414fd5b39a9e90769d9452 + + + PUBLICATION + The journal where the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" were published + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 577 + 69db426b97714835bf4937180774787a + + + PERSON + Koesten, L. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 578 + 9c7bc862339d4a5bb21ee5154d9b33bb + + + PERSON + Gregory, K. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 579 + 17bad53a0ebe4569839e5e151ff78593 + + + PERSON + Groth, P. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 580 + 53d98f08e7c74158b7318357b6c660b3 + + + PERSON + Simperl, E. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 581 + cd601f77419c403889aadeee591915b5 + + + PUBLICATION + The journal where the paper "Talking datasets–understanding data sensemaking behaviours" was published + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 582 + 0f564ebd53e940fba9d16674ac7bc038 + + + PERSON + Kuratov, Y. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 583 + 7deb75816e4f473480e0c79ae99b5bf4 + + + PERSON + Bulatov, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 584 + 7f85b181f1184f77aeb3ea2155cf4027 + + + PERSON + Anokhin, P. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 585 + d148b2b2033048618f1a090a492a40a5 + + + PERSON + Sorokin, D. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 586 + 4d839a10353e4144a26563b0966721d5 + + + PERSON + Sorokin, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 587 + 521a862bb196488389f17c0b0f4b6f4d + + + PERSON + Burtsev, M. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 588 + 22ea3328fb6343f4ad2862495ea27640 + + + TECHNOLOGY + Langchain graphs is a technology developed by LangChain + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 589 + 3f9a2a2c1c0a424e8b4980ea9d48bdbe + + + PERSON + Laskar, M. T. R. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" and also contributed to the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models." Both works highlight Laskar's expertise in leveraging transformer models and transfer learning techniques to enhance the performance of query-focused abstractive text summarization, demonstrating a significant contribution to the field of Natural Language Processing and Information Retrieval. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 5 + 590 + aa2ec452728a4703ae1bdabe85b6c079 + + + PERSON + Hoque, E. is an author of two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning to improve the effectiveness of transformer models in query-focused abstractive summarization tasks. Both works contribute to advancing the understanding and application of transformer models in specialized summarization contexts. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 5 + 591 + c5ddb31e0a9c4b2683e4631283dd505b + + + PERSON + Huang, J. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 592 + 07d8eeb549044ac88d2e788c146a0ef1 + + + PUBLICATION + The conference where the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" was presented + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 593 + 47df2815030c4f1c99facd5cf2482526 + + + PUBLICATION + arXiv preprint refers to a preprint of a paper that is available on the arXiv repository + 71f6daf11e64e5273a3847d46bf228e1 + 18 + 594 + ae521508bdc244f99c4fce4ab5214c79 + + + EVENT + The 33rd Canadian Conference on Artificial Intelligence, held in Ottawa, ON, Canada, from May 13–15, 2020 + 6cd82819982879bd164547d2773ba5c7 + 2 + 595 + 6315b4bf135c40358823ed7e4e4060e2 + + + EVENT + The 2020 edition of the Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + 1 + 596 + 33905debec1a45ecae1c65daac1d854c + + + PUBLISHER + Springer is the publisher of the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + 1 + 597 + bfbe904780fe47daad1a04126b12923c + + + PERSON + Huang, J. X. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + 3 + 598 + 0614f00e932c4cd0b53928053811ebc1 + + + PUBLICATION + The journal where the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" was published + 6cd82819982879bd164547d2773ba5c7 + 3 + 599 + 9ef487dd0b574b108c60a56d6a2f146c + + + PERSON + Lewis, P. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9 + 600 + 4067269e7f6943cdbc299ce02b7eadbd + + + PERSON + Perez, E. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9 + 601 + 094a736ba43c4da48c556437f47f88d1 + + + PERSON + Piktus, A. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9 + 602 + 563c2af32bb3476299e9b24a646097ab + + + PERSON + Petroni, F. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks"Petroni, F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + PERSON + 9 + 603 + d59b49eb94ce442d89907e90c5d3a44e + + + PERSON + Karpukhin, V. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 5 + 604 + 8ea7cef407df48098046551e303e1c64 + + + PERSON + Goyal, N. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 5 + 605 + 186e60d2176547bf84e5bf87bd16bb40 + + + PERSON + Küttler, H. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 606 + e65017091c8d4c7daa45b6c8414e0465 + + + PERSON + Lewis, M. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 607 + a0f326b9597b49dda6563e9208316117 + + + PERSON + Yih, W.-T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 608 + bff3db70f9af4f2c87a93df48ecbb6bc + + + PERSON + Rocktäschel, T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 609 + bf91f36307cb43e1ab1e967cb3ba8274 + + + PERSON + Liu, N. F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 610 + cd58a8740ba54d86a77db9bb9544ef0d + + + PERSON + Lin, K. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 611 + e96d3475d43b42a781b297ae7e650afe + + + PERSON + Hewitt, J. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 612 + 1ce76a5547854d458878bd445f0ccbd6 + + + PERSON + Paranjape, A. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 613 + 11e4325f59394ff1bc89892f79288702 + + + PERSON + Bevilacqua, M. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 614 + 71743537a07c440ea1710a269da8b538 + + + PERSON + Liu, Y. is an author of the paper "Hierarchical transformers for multi-document summarization" + 6cd82819982879bd164547d2773ba5c7 + 0 + 615 + 1389192ce5464be6b3b5749bc9536709 + + + PERSON + Lapata, M. is an author known for significant contributions to the field of Natural Language Processing and Information Retrieval. Notably, Lapata, M. has authored the paper "Hierarchical transformers for multi-document summarization," which explores advanced techniques in summarizing information from multiple documents using hierarchical transformer models. Additionally, Lapata, M. has contributed to the paper "Text summarization with latent queries," which delves into innovative methods for summarizing text by leveraging latent query representations. These works highlight Lapata, M.'s expertise and active research in the domain of text summarization, showcasing a focus on developing sophisticated models and methodologies to enhance the efficiency and accuracy of summarization tasks. + 6cd82819982879bd164547d2773ba5c7,fc4b27d64f055b7fc30176ba110dd02e + 2 + 616 + b349041c0be64c62b964ab1234e055e6 + + + TECHNOLOGY + LlamaIndex Knowledge Graph Index is a technology developed by LlamaIndex + 6cd82819982879bd164547d2773ba5c7 + 0 + 617 + 969e1ea0b1e443a68e9a65dfef91d161 + + + PERSON + Manakul, P. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + 0 + 618 + 8e09e7cfea7d405db8b22ae2f836ccb1 + + + PERSON + Liusie, A. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + 0 + 619 + 490583524d394bf79289c5fe34f7dcf1 + + + PERSON + Gales, M. J. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + 0 + 620 + d7db38bb599c42cab7066f3fdd282282 + + + PERSON + Mao, Y. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 621 + efd87a59d01e47c8adc02f63ef2c5c3e + + + PERSON + He, P. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 622 + 80e3ce3de41e4601823a333e22b7bb3f + + + PERSON + Liu, X. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 623 + 50eabc166e8944a49197e79c32f27597 + + + PERSON + Shen, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Shen, Y.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 0 + 624 + 5197a3fb02ef4677abd1900aa87e4efa + + + PERSON + Gao, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 625 + 887f444240bb474da23cdfb6abf7a998 + + + PERSON + Han, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 626 + 5d29053f2ce74442aa1855b327ef3bb7 + + + PERSON + Chen, W. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Chen, W.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 627 + 7e40cd12839a4577a95e33d785147a31 + + + PERSON + Martin, S. is an author of the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing a comprehensive, open-source solution for the layout of large graphs, which is a critical task in the visualization and analysis of complex networks. The toolbox aims to facilitate the understanding and interpretation of large-scale graph data, making it a valuable resource for researchers and practitioners in fields such as computational linguistics, information retrieval, and data science. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 628 + 8fe58de8a04f4f8f807c77fb41829a3a + + + PERSON + Brown, W. M. is an author of the paper "Openord: An open-source toolbox for large graph layout." + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 629 + a9f50861273c4bb697d868a9d049d392 + + + PERSON + KLAVANS, R. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 630 + be4820f29fd942b282049fa49697b4ed + + + PERSON + Boyack, K. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on the development and application of Openord, a comprehensive open-source toolbox designed for the layout of large graphs. The paper likely discusses the methodologies, algorithms, and practical implementations of the toolbox, contributing to the fields of graph theory and data visualization. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 631 + 6deaefe707f84b3dbda979dea0d095ac + + + EVENT + The conference where the paper "Openord: An open-source toolbox for large graph layout" was presented + 833e7d67dcd30790b26b71c9b5306f6b + EVENT + 0 + 632 + d053ea9432a24fb192e8d6aa993b0caa + + + TECHNOLOGY + GPT-4 is a large language model used in Microsoft's study on scientific discovery + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + 38 + 1 + 1 + 633 + a3e683d294ed42a28d60d09a36cbeb54 + + + TECHNOLOGY + Project NaLLM is a project developed by Neo4J + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + 1 + 634 + 39887ca8567141d5b857b87a2bca4086 + + + PERSON + Newman, M. E. is the author of the paper "Modularity and community structure in networks" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 1 + 635 + 8df8563ab0394ee9a91b89dea7d59404 + + + PUBLICATION + The journal where the paper "Modularity and community structure in networks" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + 1 + 636 + 12398f70065143839d812fd42ac4b2e7 + + + PERSON + Ram, O. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 637 + 74d43d20f251441baf8e3db64fedca43 + + + PERSON + Levine, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 638 + 1b7a22f76f7741e8b140bdc3d8856d76 + + + PERSON + Dalmedigos, I. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 639 + b823ba1bfe944fa9887edd8faf8a5f17 + + + PERSON + Muhlgay, D. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 640 + d0bfb473fdc64643954cdb4675e2f389 + + + PERSON + Shashua, A. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 641 + a4db1b2a9c3e4d2d838725f8166c36b4 + + + PERSON + Leyton-Brown, K. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 642 + 8dae140578c841ae9373cbc607c4a6e6 + + + PERSON + Shoham, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 643 + b215cc33cf40434f87f284ff8f3506a4 + + + PUBLICATION + The journal where the paper "In-context retrieval-augmented language models" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + 0 + 644 + c1ff9d8e1b8745d6860c34ce26122d79 + + + PERSON + Ranade, P. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 1 + 645 + 9d1e6ca9ae8e4e068fb74631a633b20b + + + PERSON + Joshi, A. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 1 + 646 + 1d7b0deca7674777bf76c163ac065845 + + + PERSON + Sarthi, P. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 5 + 647 + 03afe9988f864c9fa501bfbf043f74c0 + + + PERSON + Abdullah, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 5 + 648 + 4084f614af494fa8ab73095fb5b6b07b + + + PERSON + Tuli, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 5 + 649 + 3ce25564af6e47f390a0b16b6f9433a1 + + + PERSON + Khanna, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 3 + 650 + 78213664d0eb45d1a9239ba4b85b10f7 + + + PERSON + Goldie, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 3 + 651 + 1226e4a4077b4b3a970db4d2509b590c + + + PERSON + Manning, C. D. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" and the paper "Raptor: Recursive abstractive processing for tree-organized retrieval". These contributions highlight Manning's involvement in advancing the fields of Natural Language Processing and Information Retrieval, particularly in the areas of multi-hop question answering and recursive abstractive processing. + 833e7d67dcd30790b26b71c9b5306f6b,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 4 + 652 + b4c7de7a824a4a71b9f52193d2f1a10d + + + PERSON + Scott, K. is associated with "Behind the Tech" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 0 + 653 + b609f1939dae4c7383c7d199bb3c7dc3 + + + PERSON + Shao, Z. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 0 + 654 + aeee2f443dfb4e3ea80af6ae1d9197ce + + + PERSON + Gong, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 0 + 655 + 8c46d37bc26e4d4dbd37d6ee26867bc6 + + + PERSON + Huang, M. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + 2 + 656 + 58a8fa7f29e347bdb9689b70b065a779 + + + PERSON + Duan, N. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + 2 + 657 + fae3fe31deb141ab93143ac411f1eaaa + + + PERSON + Su, D. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 658 + a2cb46c226b94831853a5d28c5d94b0a + + + PERSON + Xu, Y. is an author of multiple academic papers in the field of Natural Language Processing and Information Retrieval. Notably, Xu, Y. contributed to the paper titled "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," which addresses the management of scholarly information related to COVID-19 through advanced question answering and summarization techniques. Additionally, Xu, Y. co-authored the paper "Text summarization with latent queries," which explores innovative methods for text summarization by leveraging latent queries. These contributions highlight Xu, Y.'s expertise and active involvement in developing sophisticated systems for information retrieval and summarization. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 7 + 659 + d3511ecd27cd4166bdb39e757e275300 + + + PERSON + Yu, T. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 660 + de3b561f5cce4c83bccb39180e362c97 + + + PERSON + Siddique, F. B. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 661 + 5bfefaa0fce04002851733337bed714c + + + PERSON + Barezi, E. J. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 662 + b5fed5609f154df58c6a9f74e55fc0ba + + + PERSON + Fung, P. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 663 + 91ae5251eaab4c08afe6cd4cbefcaa6b + + + PERSON + Tang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 664 + bbdd53a15e99452a9deff05d1de2d965 + + + PERSON + Yang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 665 + 532bf54d5a924ff48aee254970efb914 + + + PERSON + Touvron, H. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9 + 666 + 2489232bd2bb492babe00617e7290282 + + + PERSON + Martin, L. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 7 + 667 + d2ed972353af4d1db74702638bfdbb58 + + + PERSON + Stone, K. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 668 + 575befc8d64c47eb95af8b1096e02963 + + + PERSON + Albert, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 669 + d6e6366617e04b0ba6732fd1d2d76429 + + + PERSON + Almahairi, A. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 670 + b4c4354c8edb40db984942799fe0c8b1 + + + PERSON + Babaei, Y. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 671 + 170507a64973429f818067b80506d428 + + + PERSON + Bashlykov, N. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 672 + fd9b298e6aea4685bbb2064b05fcda79 + + + PERSON + Batra, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 673 + eeecb159cc8a4c8989f8da0f3df09f2a + + + PERSON + Bhargava, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 674 + 70f22b1d7336492dbade94b8edefe457 + + + PERSON + Bhosale, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 675 + 66e098dc431146e19fc4bc2ea37efbd9 + + + PERSON + Traag, V. A. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 676 + 932e213c57134098a07073febd51dcc2 + + + PERSON + Waltman, L. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 677 + 9593428ad36746ae8af6d8ce639834ef + + + PERSON + Van Eck, N. J. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 678 + 1bcaeb58479d42a6963a073c09f3f397 + + + PUBLICATION + Scientific Reports is the journal where the paper "From Louvain to Leiden: guaranteeing well-connected communities" was published + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 679 + 1ef0c1c59ce946668ccf1a6a4f5ab7cc + + + PERSON + Trajanoska, M. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 680 + d734746e3d6146f780af91827e578dfd + + + PERSON + Stojanov, R. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 681 + 21ed913271614cbeb1b754cdbbef13af + + + PERSON + Trajanov, D. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 682 + 1505dfebbfb04652b0ba57de1a251d67 + + + PERSON + Trivedi, H. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 683 + 907ec65076e5494a8631efffb81b3178 + + + PERSON + Balasubramanian, N. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 684 + 2dc7f6b230db452190a09643ca3d5ec0 + + + PERSON + Khot, T. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 685 + c20ecfc93b3a4875ade5c92cfe4b94a1 + + + PERSON + Sabharwal, A. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 686 + 4bc7dc91ede345dfb63d7d4f7ac3554f + + + PERSON + Wang, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 48 + 1 + 8 + 687 + 0b2b815c9f834aaaac0c341097def9ba + + + PERSON + Liang, Y. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 48 + 1 + 8 + 688 + 424ae71c56024094a02e6fd9bfcfbb04 + + + PERSON + Meng, F. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 48 + 1 + 8 + 689 + 400d10f2ee1d49be9a66efa34dada0e6 + + + PERSON + Sun, Z. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 48 + 1 + 8 + 690 + 91deb9f152264e958d106d481ff2e1ee + + + PERSON + Shi, H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 48 + 1 + 8 + 691 + 586cf02da9494088aed9b3419725638f + + + PERSON + Li, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through their work on evaluating language models. Specifically, Li, Z. has co-authored the paper titled "Is ChatGPT a Good NLG Evaluator? A Preliminary Study," which explores the effectiveness of ChatGPT as a natural language generation evaluator. Additionally, Li, Z. has co-authored another paper, "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which examines the performance of large language models in evaluative roles using specific benchmarking tools. These contributions highlight Li, Z.'s active involvement in advancing the understanding and assessment of language models within the academic community. + 8d87efac8c50cf20cdf26bf61e5e2035,b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + 48 + 1 + 20 + 692 + 229d85a2783e4a2991f17d2ab5750af7 + + + PERSON + Xu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 48 + 1 + 9 + 693 + b7f97d1909a3433abef8ca8e9334fafa + + + PERSON + Qu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 48 + 1 + 9 + 694 + b7fdfffc38b94bf7872eabe9b022c8fd + + + PERSON + Zhou, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 48 + 1 + 9 + 695 + 6242e0c237a348908d0256ea790a0211 + + + PERSON + Wang, S. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" and also contributed to the paper "Is chatgpt a good nlg evaluator? a preliminary study." These works indicate Wang, S.'s involvement in cutting-edge research within the fields of federated search, retrieval augmented generation, and natural language generation evaluation, showcasing a focus on both the technical and evaluative aspects of Natural Language Processing and Information Retrieval. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 47 + 1 + 4 + 696 + 7cc9f26737e1442595e53253e98015ef + + + PERSON + Khramtsova is an author mentioned in the text + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 697 + 1868fec1493643208dbdcad7bc97dfa0 + + + PERSON + H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 48 + 1 + 4 + 698 + a87aa935dccf49cd98b40fb5afe7ad5c + + + PERSON + Khramtsova, E. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 47 + 1 + 4 + 699 + 36870a3393f6413e9bf647168eb6977a + + + PERSON + Zhuang, S. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through multiple academic papers. Notably, Zhuang, S. co-authored the paper titled "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," which explores the evaluation of federated search systems within the framework of retrieval-augmented generation. Additionally, Zhuang, S. co-authored another significant paper, "Judging llm-as-a-judge with mt-bench and chatbot arena," which delves into the assessment of large language models (LLMs) using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Zhuang, S.'s active involvement in advancing research in federated search and the evaluation of LLMs. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + 46 + 1 + 15 + 700 + 4fe3ff52700c491f8cc650aadb4d7cb0 + + + PERSON + Zuccon, G. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 47 + 1 + 4 + 701 + f1f6f6435a444e388d67e16e847afca6 + + + PERSON + Wang, Y. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 43 + 1 + 6 + 702 + 0af2ca1c090843ea92679fd14c1fbc9a + + + PERSON + Lipka, N. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 43 + 1 + 6 + 703 + 1b06d3e53ffd4771952fbef04d1e666c + + + PERSON + Rossi, R. A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 43 + 1 + 6 + 704 + b8e966b34cba4b11b9995106767212ba + + + PERSON + Siu, A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 43 + 1 + 6 + 705 + f6de923de6474d2cab6a9c2f0d81fa59 + + + PERSON + Zhang, R. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 43 + 1 + 6 + 706 + 6915637e8d124fdc8473111d501e3703 + + + PERSON + Derr, T. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 43 + 1 + 6 + 707 + 2233f31929194eac89333ce8731a5584 + + + PERSON + Yang, Z. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 708 + 61f1dc4267314470ac820b6a46c61f7b + + + PERSON + Qi, P. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 709 + f0c578614b224345974c3e4c110878af + + + PERSON + Zhang, S. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 710 + 7ffb88ebc729492c897ccfb569d7f6d0 + + + PERSON + Bengio, Y. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 711 + 60dce7d8bc1b4729a038178a400b9a59 + + + PERSON + Cohen, W. W. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 712 + 4cbb4e238c5b4656803fb9b4b6c3512e + + + PERSON + Salakhutdinov, R. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 713 + 652873bcd6d5432187e5deafc4fc5211 + + + CONFERENCE + The conference where the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" was presented + fc4b27d64f055b7fc30176ba110dd02e + 0 + 714 + 78f9b30c08134ac5abb4f4e0bff0f7f2 + + + PERSON + Yao, J.-g. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + 0 + 715 + f33e4e897b1e422bb516e8a2c941d9dc + + + PERSON + Wan, X. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + 0 + 716 + fac4e1553a9840e990bbfff46e64ff27 + + + PERSON + Xiao, J. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + 0 + 717 + 029a55d327ee4fb3a8314b36d52bdf34 + + + PUBLICATION + The journal where the paper "Recent advances in document summarization" was published + fc4b27d64f055b7fc30176ba110dd02e + 0 + 718 + 5a636c894c384532bff66212cf9d5824 + + + PERSON + Yao, L. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models"Yao, L. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 45 + 1 + 1 + 719 + a9c468ef78704e9aabfc0317a5b1b42d + + + PERSON + Peng, J. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 0 + 720 + 5df80c25d33a4d148a14aa614343cc6b + + + PERSON + Mao, C. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 0 + 721 + 6a87f06ed55a46f29b24f77e548a3f1d + + + PERSON + Luo, Y. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 0 + 722 + 0daf88ac4ec94cbb868e27e956c6d7f1 + + + PERSON + Zhang, J. is an author of the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 45 + 1 + 1 + 723 + 9ed120043e6247be9965e4904920991b + + + PERSON + Zhang, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 45 + 1 + 1 + 724 + 94d81d7de9254ae4b3b16fcc69aa22ea + + + PERSON + Gan, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 45 + 1 + 1 + 725 + 60c9212246f84ae5b6ab254127a39262 + + + PERSON + Wang, C. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 45 + 1 + 1 + 726 + 0f8d0c36a4274526a9eddedae5e63881 + + + PERSON + Zheng, L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zheng, L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Zheng, L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools like MT-Bench and Chatbot Arena. These contributions highlight Zheng, L.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR domains. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 46 + 1 + 12 + 727 + 6aedd377efbe4f07ae42e546996e7bfa + + + PERSON + Chiang, W.-L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Chiang, W.-L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Chiang, W.-L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Chiang, W.-L.'s active involvement in advancing the understanding and capabilities of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 46 + 1 + 12 + 728 + 1aa8484562784f378851c33843c89687 + + + PERSON + Sheng, Y. is an author known for contributing to the field of Natural Language Processing and Information Retrieval. Notably, Sheng, Y. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Sheng, Y. has contributed to the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Sheng, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic and technical community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 46 + 1 + 12 + 729 + f1a65d05dd5d456b889217020475ef80 + + + PERSON + Wu, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Wu, Z. co-authored the paper titled "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Wu, Z. is also credited with co-authoring the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Wu, Z.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 46 + 1 + 12 + 730 + c077d92b48b6477db91e1a0460600f52 + + + PERSON + Zhuang, Y. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zhuang, Y. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness of knowledge graphs. Additionally, Zhuang, Y. has also authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Zhuang, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the domain. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 46 + 1 + 12 + 731 + 5ca888df9b884e54accdd2ff29d125c1 + + + PERSON + Lin, Z. is an author of the paper "Exploring large language models for knowledge graph completion" and also contributed to the paper "Judging LLM-as-a-judge with MT-Bench and Chatbot Arena." These works indicate Lin, Z.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the application of large language models for tasks such as knowledge graph completion and the evaluation of language models in judgment tasks. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 46 + 1 + 12 + 732 + 8290a6212d6c4430ae0056c7e8eccd5f + + + PERSON + Li, D. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant research. Notably, Li, D. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Li, D. has also co-authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Li, D.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 46 + 1 + 12 + 733 + 14f8ac195fdb4e06a0b9ebc6ef391180 + + + PERSON + Xing, E. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Xing, E. contributed to the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Xing, E.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 46 + 1 + 12 + 734 + 667ee58a79194316ae2b82eadd3fc575 + + + TECHNOLOGY + Chatbot Arena is a platform or tool used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 46 + 1 + 11 + 735 + b0e3ee2324054c88adacdf80db13278f + + + 1.0 + Darren Edge and Ha Trinh co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 28b7457ca5dc4a38a488946a3f8e207e + 0 + 1 + + + 1.0 + Darren Edge and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8029a14d15404e6db95ddf5e2bf9fc15 + 1 + 1 + + + 1.0 + Darren Edge and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 389314ca89d445888c8d4985864dd733 + 2 + 1 + + + 1.0 + Darren Edge and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 87fe1462b9064d5692641ab48e826301 + 3 + 1 + + + 1.0 + Darren Edge and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + a55175ac57014df696ca09d0def9604b + 4 + 1 + + + 1.0 + Darren Edge and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 1766e8858d7b45ed97f71cb5a39e96ea + 5 + 1 + + + 1.0 + Darren Edge and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 6191e014f3f64e46a0777063ed4ac19a + 6 + 1 + + + 1.0 + Darren Edge is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + 21b0499cf14342269c46170c291d0535 + 7 + 1 + + + 1.0 + Ha Trinh and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + c1ef05b38b3f4d59888150fc0dd26826 + 8 + 1 + + + 1.0 + Ha Trinh and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 74cb9b3510e84498b9aee0d904316e8b + 9 + 1 + + + 1.0 + Ha Trinh and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 043d764b2e1b4d1294651ff938df5391 + 10 + 1 + + + 1.0 + Ha Trinh and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 31f2170fef004f3281c533a4a60dc3f3 + 11 + 1 + + + 1.0 + Ha Trinh and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 57f186c5c2754483ba66750e98222f95 + 12 + 1 + + + 1.0 + Ha Trinh and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 4b3fc569d91f4a7aa6501ad4fcf67b7a + 13 + 1 + + + 1.0 + Ha Trinh is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + cbc1667556f84a5eadf867a823e6986c + 14 + 1 + + + 1.0 + Newman Cheng and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + a876d1ab79864396bc47a039225fd5c7 + 15 + 1 + + + 1.0 + Newman Cheng and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + c09f67d4f25448c99f7c0552c30b7706 + 16 + 1 + + + 1.0 + Newman Cheng and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + c0866306dc8c4da2a8a81c0c3a78b657 + 17 + 1 + + + 1.0 + Newman Cheng and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 3884c37eb13a4c9097ee2c5be4eeefaf + 18 + 1 + + + 1.0 + Newman Cheng and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 90764eb2cab74cffb1c7d72d28b965cc + 19 + 1 + + + 1.0 + Newman Cheng is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + 01abe16e67c241a887aa62abe22d155c + 20 + 1 + + + 1.0 + Joshua Bradley and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 37049be0a2c240c6a06acf9339237b8b + 21 + 1 + + + 1.0 + Joshua Bradley and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + e785c52881704d95bf4ec03d2720f8ae + 22 + 1 + + + 1.0 + Joshua Bradley and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 654689c65613476b9905d7afb3809cd2 + 23 + 1 + + + 1.0 + Joshua Bradley and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 15dfb45a6ffa4d34ad72cfe4b3c5cc0d + 24 + 1 + + + 1.0 + Joshua Bradley is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + 427c3b7458f148d8bace1b768e2b5b7c + 25 + 1 + + + 1.0 + Alex Chao and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 95d506750fd94e72bbd9cf2d3fe18e28 + 26 + 1 + + + 1.0 + Alex Chao and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + bf0138ccbcc740089a55fd0c24897360 + 27 + 1 + + + 1.0 + Alex Chao and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 83cd5df42643494396b00d6cb6376def + 28 + 1 + + + 1.0 + Alex Chao is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + 909d28e443fd4e0bac189373125c8309 + 29 + 1 + + + 1.0 + Apurva Mody and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + efbc2439e5034801af83ac1a0b440535 + 30 + 1 + + + 1.0 + Apurva Mody and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + b9a2ef791a064f038cac2059ebea1138 + 31 + 1 + + + 1.0 + Apurva Mody is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + 1ce2b24bc93442148dc2240d3c6223b1 + 32 + 1 + + + 1.0 + Steven Truitt and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 804c1e94e7974332a817931363ddb643 + 33 + 1 + + + 1.0 + Steven Truitt is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + 32dc0b572ad84c75a64a2007788eb981 + 34 + 1 + + + 1.0 + Jonathan Larson is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + f5c11a5ac94e40068bca8be178a6bcd6 + 35 + 1 + + + 4.0 + RAG (Retrieval-Augmented Generation) and LLM (Large Language Models) are closely intertwined in the domain of Natural Language Processing and Information Retrieval. RAG is employed to enhance the capabilities of LLMs by enabling them to retrieve pertinent information from external knowledge sources. This symbiotic relationship allows LLMs to generate and assess text more effectively. Specifically, RAG leverages the power of LLMs to access and utilize relevant data, thereby augmenting the overall performance and accuracy of text generation tasks. + e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + ea28ff7f127e4677a913952595dce2f5 + 36 + 1 + + + 7.0 + Graph RAG is a specific implementation of RAG that combines the strengths of RAG with graph-based text indexing. This method leverages the natural modularity of graphs to partition data, facilitating global summarization. As a specialized approach within the RAG framework, Graph RAG enhances the capabilities of RAG by integrating graph structures to improve the efficiency and effectiveness of text data processing and summarization. + 21e52bc06a82796b1f4bcd73edda1f2a,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 192a6d23595045f38b0d46a3d8e52fd6 + 37 + 1 + + + 1.0 + Query-Focused Summarization is a task that RAG fails to address effectively + e8d83e6e7a7c0f57b218cef24976b745 + ef67c9fc60284b50aa15ac655b06a155 + 38 + 1 + + + 1.0 + RAG retrieves relevant information from an external knowledge source + e8d83e6e7a7c0f57b218cef24976b745 + cc8201cce1024b5192056fe8e98fda22 + 39 + 1 + + + 1.0 + Naive RAG is a specific implementation of RAG + e8c8f911135faf3ff35f24107eb3f99c + 97e097f9022540b88ab7c13d2805c25f + 40 + 1 + + + 1.0 + Ram et al., 2023 discusses RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + 829a6299a5fa4e7b8ff4020020a0be05 + 41 + 1 + + + 2.0 + Naïve RAG is a basic form of RAG + f35de4d9fb65f1d5a392064b20545c19 + dde2742459c24fb4a91172aa5c1a7620 + 42 + 1 + + + 2.0 + Modular RAG is an advanced form of RAG + f35de4d9fb65f1d5a392064b20545c19 + 323979a67d79498fa271acdf8cd1a0c2 + 43 + 1 + + + 2.0 + LLMs are used in various RAG tasks such as knowledge graph creation and completion + 92e93fc6449756c0a60200636b297f65 + c7e8b188b45841a0a1bcb22f3445ea6e + 44 + 1 + + + 2.0 + The paper by Trajanoska et al. discusses using LLMs for knowledge graph creation, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 5a4ad077106a4a3f951f43d2e01499b0 + 45 + 1 + + + 2.0 + The paper by Yao et al. discusses using LLMs for knowledge graph completion, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + a7ec8df038d7461689d28f1bdea84d9b + 46 + 1 + + + 2.0 + The paper by Ban et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 8ddefa32e2ed4eaf8f76d17a676f74f3 + 47 + 1 + + + 2.0 + The paper by Zhang et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 95ec30ce8dbe4ca28714e3e3735da8f3 + 48 + 1 + + + 2.0 + The paper by Gao et al. discusses advanced RAG where the index is a knowledge graph + 92e93fc6449756c0a60200636b297f65 + 259e7f5e2ec04418937513413b6d51d1 + 49 + 1 + + + 2.0 + KAPING is a method where the index is a knowledge graph, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 2289f06dd3804a3c84371dda0bab091e + 50 + 1 + + + 2.0 + G-Retriever is a method where subsets of the graph structure are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 34ff8ef897804691842071f9ff78708e + 51 + 1 + + + 2.0 + Graph-ToolFormer is a method where derived graph metrics are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + a7401447d994439993da7cc57f127649 + 52 + 1 + + + 2.0 + SURGE is a method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 754b0f2616064b18abb90f409ef0539a + 53 + 1 + + + 2.0 + FABULA is a method where retrieved event-plot subgraphs are serialized using narrative templates, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + acd35bb6b3cb4979a3f3fb68a86b3b05 + 54 + 1 + + + 2.0 + The paper by Wang et al. discusses a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 9e1e7f67ba044c7fbf64723af1ade58e + 55 + 1 + + + 2.0 + Sensemaking questions are used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + 57e16ff087a84b8ebd70de1e7e534225 + 56 + 1 + + + 2.0 + The evaluation of RAG systems focuses on corpora in the region of 1 million tokens + 92e93fc6449756c0a60200636b297f65 + bbf4007dc9c0486b8ea76d616045467a + 57 + 1 + + + 2.0 + Trade-offs are considerations involved in building a graph index for RAG systems + 92e93fc6449756c0a60200636b297f65 + 9535f4d754044e128cd3951a9d2e3702 + 58 + 1 + + + 2.0 + A graph index is a data structure used in RAG systems to organize and retrieve information + 92e93fc6449756c0a60200636b297f65 + e1ed13e29ee946d4aaafac50aaa3b68f + 59 + 1 + + + 2.0 + Performance of RAG systems varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + eb961d47a30c4870a1134b4a4672a8b2 + 60 + 1 + + + 2.0 + Different data types are used in RAG systems + 92e93fc6449756c0a60200636b297f65 + 5b019e8652264136b95306bac70a2e25 + 61 + 1 + + + 2.0 + Dataset sizes affect the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + 325fc9e2b37043b7af9f6ad338b09469 + 62 + 1 + + + 2.0 + Evaluation is the process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + 6bb11aa08b414232b5b45f10f5766f62 + 63 + 1 + + + 2.0 + Corpora are collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + 179737fd23c943babdfae01ac5c6bfc3 + 64 + 1 + + + 2.0 + Different question types are used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + 65a31e4da283411fb7c971f63d606723 + 65 + 1 + + + 2.0 + Target metrics are specific measures used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + de31810d43174a52aa2f31b72f4542f5 + 66 + 1 + + + 4.0 + Graph RAG utilizes Large Language Models (LLMs) to construct a graph-based text index, enabling the generation of summaries and the answering of queries. In this approach, LLMs play a crucial role in analyzing and generating text based on the information retrieved through the graph structure. Additionally, LLMs leverage the Graph RAG framework to provide comprehensive overviews of public figures in the entertainment industry. This integration of LLMs within Graph RAG enhances the system's ability to process and synthesize complex text data, making it a powerful tool for information retrieval and natural language processing tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + a6ae1d99330443fcacb06ace15a0d937 + 67 + 1 + + + 1.0 + Text chunks are processed using LLM to extract elements of a graph index + bc9e2c9e369c4108cf4f6dd5f60960f4 + 5174cdabb6024de0975762d3a80b059f + 68 + 1 + + + 1.0 + LLM is used to extract elements of a graph index from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + e379fba901174b529250169e62d98c09 + 69 + 1 + + + 2.0 + LLM (Large Language Model) and Few-Shot Examples are closely related in the context of Natural Language Processing and Information Retrieval. Few-shot examples are provided to the LLM for in-context learning, which helps tailor the extraction prompt. This technique is particularly useful for improving the performance of the LLM in specialized domains. By leveraging a small number of examples, the LLM can better understand and adapt to specific tasks, thereby enhancing its overall effectiveness in extracting and processing information within those specialized areas. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4 + 81ee8bb20bbb4d37bc0db642f1c75b8e + 70 + 1 + + + 1.0 + LLM extracts named entities from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 93b4aa6ce6e44123a861d4c3b3d509a2 + 71 + 1 + + + 1.0 + Kuratov et al. (2024) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + ee8414e314f547eeb369849cdb51bac2 + 72 + 1 + + + 1.0 + Liu et al. (2023) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + 9f77aa8888bd4f94abba8a77c4b0565c + 73 + 1 + + + 1.0 + LLM prompts are instructions given to the LLM for extracting elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + dcf33412678340319e7ec8f7be267ef9 + 74 + 1 + + + 1.0 + Recall degradation occurs with longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + 4aa1e0fa00c048939a5d006bfd305fb4 + 75 + 1 + + + 1.0 + The extraction process involves using LLM to identify and extract elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 03053ab4a9054384a5f5e88d28841621 + 76 + 1 + + + 1.0 + Default prompt is the standard set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + 9fd0f20997d541bca46c4ec9843a5d0f + 77 + 1 + + + 1.0 + Secondary extraction prompt is an additional set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + 27168beee1ff456696c330c9c3b3259f + 78 + 1 + + + 1.0 + The LLM uses covariate prompts to extract additional attributes associated with detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + e1c20e06aeac436788a9c6e918bcb844 + 79 + 1 + + + 1.0 + The LLM uses multiple rounds of gleanings to ensure no entities are missed + 2c6ed90897310eea2f28e33fff1c32b0 + 344417f626ef4da4be4539ef4037bf3f + 80 + 1 + + + 1.0 + Logit bias is used to force a yes/no decision from the LLM during entity extraction + 2c6ed90897310eea2f28e33fff1c32b0 + 8b1fff87c350475fb1d411a26c3c5b0c + 81 + 1 + + + 1.0 + The LLM extracts element instances from source texts + 2c6ed90897310eea2f28e33fff1c32b0 + 898a9458adfb4c13a1eafacf6a1068f6 + 82 + 1 + + + 1.0 + The LLM detects and summarizes communities of entities + 2c6ed90897310eea2f28e33fff1c32b0 + 5448f05781de44ea96e3dea40b285842 + 83 + 1 + + + 1.0 + LLM generates intermediate answers and scores for each chunk + 1d07b4248c2655081c7af0e373bd70c9 + 76b1e69904b84d09ba05c4b7efc48f32 + 84 + 1 + + + 1.0 + LLM generates a helpfulness score for each answer + 1d07b4248c2655081c7af0e373bd70c9 + 3f5590a604894d268603b4b27c3348b5 + 85 + 1 + + + 2.0 + LLM is used to generate questions for evaluating the Podcast Transcripts dataset + 922778ce1cb2fdd6dbab1746c8795620 + 68f998c9c8c34bb7a994de5a998bb9a0 + 86 + 1 + + + 2.0 + LLM is used to generate questions for evaluating the News Articles dataset + 922778ce1cb2fdd6dbab1746c8795620 + aafc13d02ade40adae13d3bee241817a + 87 + 1 + + + 1.0 + LLM uses Naive RAG to list public figures mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 81a4818e5cf84ea085abf09de385c86e + 88 + 1 + + + 1.0 + LLM-generated responses are evaluated using assessment metrics + e8c8f911135faf3ff35f24107eb3f99c + b69851bf63e34ced83827b0021628543 + 89 + 1 + + + 1.0 + LLM-generated responses are evaluated using specific questions + e8c8f911135faf3ff35f24107eb3f99c + b83a4e11bfa64559954327714b73293f + 90 + 1 + + + 1.0 + Ad-hoc LLM use involves the use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + de23b974cc90497eb4363e26d931a57c + 91 + 1 + + + 2.0 + LLMs are used for knowledge graph creation + f35de4d9fb65f1d5a392064b20545c19 + a9de65176e234a9f9073b8df9d675e90 + 92 + 1 + + + 2.0 + LLMs are used for knowledge graph completion + f35de4d9fb65f1d5a392064b20545c19 + 09a1bd11eb9347a9b466edad1a562cc5 + 93 + 1 + + + 2.0 + LLMs are used for the extraction of causal graphs + f35de4d9fb65f1d5a392064b20545c19 + 11d74eab1dcb4fcba7c45def5f0ee22d + 94 + 1 + + + 2.0 + LLMs are used for knowledge graph creation as per Trajanoska et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 4f6a6fd018a948f4bd0e630266b8bf61 + 95 + 1 + + + 2.0 + LLMs are used for knowledge graph completion as per Yao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 17dbfbecfaf0436bb11ed8f867c0caa1 + 96 + 1 + + + 2.0 + LLMs are used for the extraction of causal graphs as per Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + 2b1ec99684574c2ab26bb050d5b57a4d + 97 + 1 + + + 1.0 + Graph RAG is proposed as a method to combine the strengths of RAG and QFS + e8d83e6e7a7c0f57b218cef24976b745 + 1ccce5d1892a4b6995bbaec22882d34d + 98 + 1 + + + 7.0 + Graph RAG is an approach that utilizes community summaries in various capacities to enhance its functionality. Specifically, community summaries are generated within the Graph RAG framework and are employed to generate partial responses. These summaries are also used to compare against source texts, serving as a form of self-memory for the system. By leveraging community summaries, Graph RAG aims to improve the comprehensiveness and diversity of answers. Additionally, Graph RAG incorporates summaries of root-level communities within an entity-based graph index, further enriching its analytical capabilities. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 51cd93f89fbe4bcf883cdb2ca6774cd6 + 99 + 1 + + + 1.0 + Graph RAG is designed to handle global sensemaking questions over large datasets + e8d83e6e7a7c0f57b218cef24976b745 + 5f353b18fadb438f95ba0ea8feae137c + 100 + 1 + + + 2.0 + Graph RAG is implemented in Python. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + 947d70dd14b34cf398a1ab6dbdc51161 + 101 + 1 + + + 1.0 + The open-source implementation of Graph RAG will be available at this URL + e8d83e6e7a7c0f57b218cef24976b745 + 90f5597a558a4652bded9001a4ec2e56 + 102 + 1 + + + 1.0 + Graph RAG uses an entity knowledge graph to index text + e8d83e6e7a7c0f57b218cef24976b745 + 9532cf83e9324ea0a46e5ac89bac407d + 103 + 1 + + + 3.0 + Graph RAG is an approach that is evaluated for its comprehensiveness, a target quality used to assess its effectiveness. The method aims to improve the comprehensiveness of generated answers, ensuring that the information provided is thorough and complete. This evaluation metric is crucial in determining the success of Graph RAG in producing detailed and accurate responses. + 21e52bc06a82796b1f4bcd73edda1f2a,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + 8919fa72a9e74d1daff801e8f4c15b2b + 104 + 1 + + + 3.0 + Graph RAG is an approach in the domain of Natural Language Processing and Information Retrieval that focuses on improving the diversity of generated answers. Diversity, in this context, is a target quality used to evaluate the performance of the Graph RAG approach. By enhancing the diversity of responses, Graph RAG aims to provide a broader range of answers, thereby improving the overall effectiveness and robustness of the system. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745 + bef38889bb86413895d7dd25b4c3137c + 105 + 1 + + + 3.0 + Graph RAG uses a knowledge graph for global summarization + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + f770bc07cecf4aba8fe2d2c33fdc5542 + 106 + 1 + + + 1.0 + Community detection algorithms are used in the Graph RAG approach to partition graphs + 21e52bc06a82796b1f4bcd73edda1f2a + 13cd49512d5642989c2c72bb5e674807 + 107 + 1 + + + 1.0 + Podcast transcripts are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + e5c5c87a281b43868c344ff60f44c100 + 108 + 1 + + + 1.0 + News articles are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + a562ffbe986247b7943990e7151f4d69 + 109 + 1 + + + 2.0 + Graph RAG is evaluated using the target quality of Empowerment. Empowerment is specifically utilized to assess Graph RAG's capability in aiding users to achieve an informed understanding. This evaluation metric underscores the importance of user comprehension and the effectiveness of the Graph RAG approach in facilitating informed decision-making processes. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4 + 7ea0bc1467e84184842de2d5e5bdd78e + 110 + 1 + + + 2.0 + Graph RAG is compared to naive RAG in the evaluation. In this comparison, Graph RAG outperformed naive RAG on comprehensiveness and diversity. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + 056f23eb710f471393ae5dc417d83fd9 + 111 + 1 + + + 1.0 + Graph RAG is compared to global map-reduce summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + e1ae27016d63447a8dfa021370cba0fa + 112 + 1 + + + 1.0 + Query-focused summarization is a method used in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + f8c10f61a8f344cea7bdafa2d8af14b8 + 113 + 1 + + + 1.0 + Activity-centered sensemaking questions are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + aa7d003f25624e19bc88d3951d4dc943 + 114 + 1 + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 1c97184ce5ea4049be417a3fd125357b + 115 + 1 + + + 2.0 + The "Graph RAG" approach is evaluated in terms of its performance by considering "Token Costs." Token costs are measured to assess the efficiency of the Graph RAG method, indicating that the computational expense associated with processing tokens is a critical factor in determining the overall effectiveness of this approach. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 13a044c404394c34af1e9b07c48aa985 + 116 + 1 + + + 1.0 + Data flow describes the high-level process of the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 69ef1ac7b1f44372979149e82ecbc860 + 117 + 1 + + + 3.0 + Design parameters are key settings in the Graph RAG approach and significantly influence the Graph RAG approach and pipeline. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 6e26ce67bacc4fa089296843463f69ad + 118 + 1 + + + 1.0 + Graph RAG uses global summarization to summarize information from a large dataset + 21e52bc06a82796b1f4bcd73edda1f2a + ae0d3104647f4e6ab3ec2cf8e60be5ca + 119 + 1 + + + 1.0 + Graph RAG aims to answer specific queries + 21e52bc06a82796b1f4bcd73edda1f2a + 49e24b5f2c1d40d7857afe327db4f554 + 120 + 1 + + + 1.0 + Graph RAG uses a corpus for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + 587f39a32e93412395d9c22ad0ac2f94 + 121 + 1 + + + 1.0 + Activity-centered sensemaking is used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 8d9ded5fc9cf4c4faba8c6c8cd50e2f4 + 122 + 1 + + + 1.0 + Real-world datasets are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 595a841aa6034c93bd3dc55681e17710 + 123 + 1 + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + d0e58b78e8e84a0c8796e707b1f95f65 + 124 + 1 + + + 1.0 + Graph RAG is compared to source text summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + 215fcc6a3b5e452da123aa7f9ef0cbc9 + 125 + 1 + + + 1.0 + Low-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 0d0fc5d4ecb548079b28979186f19bf6 + 126 + 1 + + + 1.0 + Intermediate-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + e7d3fe0f87ff47f5a4c8d9572d27245a + 127 + 1 + + + 1.0 + High-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 6f7165b558ae427ca14b2b16d1e8e204 + 128 + 1 + + + 1.0 + The Graph RAG approach involves a specific pipeline for processing and summarizing text + bc9e2c9e369c4108cf4f6dd5f60960f4 + 2ec093d2a76d45f88ec508e45ba8c6a3 + 129 + 1 + + + 1.0 + Techniques are specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 16d5a528d6374612b87a5656e8d95193 + 130 + 1 + + + 1.0 + Implementation details are specific configurations used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 40293e74dbc643e8ab6546dff759ac7c + 131 + 1 + + + 2.0 + Graph RAG is a specific implementation of RAG systems + 922778ce1cb2fdd6dbab1746c8795620 + 1834b753dc7f4a8b98c2317a551b56ee + 132 + 1 + + + 2.0 + Graph RAG is a system that utilizes root-level community summaries, denoted as C0, to answer user queries. C0 represents these root-level community summaries within the Graph RAG analysis, serving as a foundational element in the system's ability to map out relationships and understand the structural dynamics within specialized communities. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + d9b127eab2f64e338d7adcd186786a45 + 133 + 1 + + + 1.0 + Graph RAG uses high-level community summaries (C1) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + a18f7c9f58ca49d6acf18e1ca69d3033 + 134 + 1 + + + 1.0 + Graph RAG uses intermediate-level community summaries (C2) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + f3c3dd44cf50495c81e362174991242e + 135 + 1 + + + 2.0 + Graph RAG utilizes low-level community summaries, represented by C3, to answer user queries. C3 plays a crucial role in the Graph RAG analysis by providing detailed summaries of community structures, which are essential for effectively addressing user inquiries. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + 86c2b3749a3c4342bbb3a8c70c3a76a0 + 136 + 1 + + + 2.0 + Graph RAG is a key entity in the analysis, serving both as a condition being compared and as a tool for comparing multiple conditions. This dual role highlights its significance in the study, where it functions not only as a subject of comparison but also as a methodological framework for evaluating other conditions. The analysis likely involves a detailed examination of various conditions, with Graph RAG playing a central role in facilitating these comparisons. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 571f65acb3134490932feeb91b01cca3 + 137 + 1 + + + 1.0 + Graph RAG uses different levels of graph communities to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + d3faf86c153f440eaa410305b3dc6617 + 138 + 1 + + + 1.0 + The Graph RAG mechanism uses an LLM evaluator for head-to-head comparison + 322e02986c8724eedbcf3ebfa20b989c + f85786004b0540349192d2ca05b15264 + 139 + 1 + + + 1.0 + Graph RAG is a multi-stage mechanism + 322e02986c8724eedbcf3ebfa20b989c + cf56bfc9fa7d47fe9cb553dd09f2b412 + 140 + 1 + + + 1.0 + Graph RAG mentions Taylor Swift as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + a077dbcd38b644f6929cf05272c2fb9d + 141 + 1 + + + 1.0 + Graph RAG mentions Travis Kelce as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + d2659a32b9de406eb750a35d078c9774 + 142 + 1 + + + 1.0 + Graph RAG mentions Britney Spears as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + 0b26876307ad4cc48839b61a21a1d03a + 143 + 1 + + + 1.0 + Graph RAG mentions Justin Timberlake as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + c68e6c694a554256846d12178ddb12dc + 144 + 1 + + + 1.0 + Graph RAG is determined to be the winner based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + ff25ce2e8ace4bdcb765c863b483852b + 145 + 1 + + + 1.0 + Answer 1 was generated using the Graph RAG method, which provides a comprehensive and structured overview of public figures across various sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + 757e402cb7ee4601ac1bc8c4fafb5207 + 146 + 1 + + + 1.0 + Graph RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + 62e8f5f04cd04384b246291cef3a9e4d + 147 + 1 + + + 1.0 + Graph RAG is compared with source texts for answer comprehensiveness and diversity + 6f33a085ff3304e5994f7fbb86c881a4 + c04abbd5e59b4c64b023908f6db05498 + 148 + 1 + + + 1.0 + TS represents source text summarization in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 6bb9bed2e39c4e31a81f12479af3d16c + 149 + 1 + + + 1.0 + Root-level summaries are used in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 26c926c6016d4639b05427f01ba629f5 + 150 + 1 + + + 1.0 + Answer comprehensiveness is used to evaluate the performance of Graph RAG + 6f33a085ff3304e5994f7fbb86c881a4 + 8f6872eeb81b432b91405d327636113c + 151 + 1 + + + 1.0 + Win rate is used to measure the success rate of Graph RAG in providing comprehensive and diverse answers + 6f33a085ff3304e5994f7fbb86c881a4 + ac80a99fda2b488285d29596dd4d1471 + 152 + 1 + + + 1.0 + Element extraction prompts are used in Graph RAG to retain specific details in the index + 6f33a085ff3304e5994f7fbb86c881a4 + 67d6a3481e4b419292247cef5cd5b737 + 153 + 1 + + + 2.0 + Graph RAG incorporates the concept of self-memory + f35de4d9fb65f1d5a392064b20545c19 + 904cd052ec194654bb72f4027e43daa3 + 154 + 1 + + + 2.0 + Graph RAG incorporates the concept of iterative retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + 7e88fd2e835147fbb71866612735e8d4 + 155 + 1 + + + 2.0 + Graph RAG incorporates the concept of federated retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + 029d1a8c3b184aa5bb21228f40cd12fd + 156 + 1 + + + 2.0 + Graph RAG incorporates concepts used in multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + a1ebc53a0bc74a0eb6dbdd18cf3c88cd + 157 + 1 + + + 2.0 + Graph RAG incorporates concepts used in multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + a51d063ad4c744049edb359eb88407b7 + 158 + 1 + + + 2.0 + Graph RAG uses a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + b67268f90338474e8e53b9a6715b6833 + 159 + 1 + + + 2.0 + Graph RAG incorporates the concept of a tree of clarifications + f35de4d9fb65f1d5a392064b20545c19 + acb53370e72b4430a752d9ea18c17352 + 160 + 1 + + + 3.0 + Graph RAG utilizes a self-generated graph index. This self-generated graph index is a crucial component of Graph RAG, enabling it to efficiently manage and retrieve information within its graph-based framework. The use of a self-generated graph index suggests that Graph RAG has an inherent capability to construct and maintain its indexing structure, which likely enhances its performance and adaptability in handling complex data relationships. + e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + a8738c7de11543df930169741381c252 + 161 + 1 + + + 2.0 + Graph RAG incorporates concepts from Gao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 3f8b5b2727924ba0b62e6286063b6861 + 162 + 1 + + + 2.0 + Graph RAG incorporates concepts from Cheng et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + bb5010633113442eaf814852995cfa22 + 163 + 1 + + + 2.0 + Graph RAG incorporates concepts from Mao et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + 9eb8c635538243a690366f8bc1de34e0 + 164 + 1 + + + 2.0 + Graph RAG incorporates concepts from Shao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 050c5b770d51409cb40f9c52f02d1329 + 165 + 1 + + + 2.0 + Graph RAG incorporates concepts from Wang et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + 9e12f514d26d48dfab65807568a6cff9 + 166 + 1 + + + 2.0 + Graph RAG incorporates concepts from Su et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + 97f98b1623104f48aa93196a1f7dede2 + 167 + 1 + + + 2.0 + Graph RAG incorporates concepts from Feng et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 87718ef799a34104b6ef9c2df6621cbc + 168 + 1 + + + 2.0 + Graph RAG incorporates concepts from Trivedi et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + f64e87431d674f298c533f6878458b95 + 169 + 1 + + + 2.0 + Graph RAG incorporates concepts from Khattab et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + e6d44d0db58f42799a02eacbd6b14543 + 170 + 1 + + + 2.0 + Graph RAG incorporates concepts from Sarthi et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + 64961fbc3a1641378be10bcb3b0955e1 + 171 + 1 + + + 2.0 + Graph RAG incorporates concepts from Kim et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 59bcc2ec512c4c1ba44272446b419230 + 172 + 1 + + + 2.0 + Graph RAG generates community answers in parallel + f35de4d9fb65f1d5a392064b20545c19 + 8f39ae56f8b54b1b94faf04dbd0b9d11 + 173 + 1 + + + 1.0 + Graph RAG is compared to a graph-free approach for global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + f3018b934ac241639a33c925c24bc507 + 174 + 1 + + + 1.0 + Graph RAG is compared to map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + ddedfd5179e64700adced4803c75cdba + 175 + 1 + + + 1.0 + Graph RAG uses rich text annotations + e4d9b12cf2b4c691c74019eefff4fb39 + 07d501edd4614e1d9d08d01b702688a3 + 176 + 1 + + + 1.0 + Graph RAG uses a hierarchical community structure + e4d9b12cf2b4c691c74019eefff4fb39 + f745075dedcf444daa9370cf32403d31 + 177 + 1 + + + 1.0 + Graph RAG can operate using embedding-based matching + e4d9b12cf2b4c691c74019eefff4fb39 + 1ef48284d238405f94190125092a3e28 + 178 + 1 + + + 1.0 + Graph RAG can be part of hybrid RAG schemes + e4d9b12cf2b4c691c74019eefff4fb39 + 8806b817446447e3b50f5bc85ff497e1 + 179 + 1 + + + 1.0 + Graph RAG can employ map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + 7303ee20690449db8c168df3fe008bc5 + 180 + 1 + + + 1.0 + Graph RAG can extend operations across the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + 2f1c535a14b14758bf1cacca81c74878 + 181 + 1 + + + 1.0 + Alonso contributed to the work on Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + 3b78cc7ce8224afcab3e4bbe550cde10 + 182 + 1 + + + 1.0 + Graph RAG includes local graph RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + 29ec9dd9f5864170a7e75c46c11c0090 + 183 + 1 + + + 1.0 + Graph RAG uses an entity-based graph index + e4d9b12cf2b4c691c74019eefff4fb39 + 7893ee15f0e941cbacad8cc1feaacbaf + 184 + 1 + + + 2.0 + NebulaGraph launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs + 833e7d67dcd30790b26b71c9b5306f6b + f53397f743ca4d7397c0a694fe787da0 + 185 + 1 + + + 1.0 + Community summaries are used to generate partial responses + e8d83e6e7a7c0f57b218cef24976b745 + 0041db9da3694ad397f37c76f8477770 + 186 + 1 + + + 1.0 + Community summaries are created from graph communities + f0306814bf64f5c9e79603fc6a52f4ea + a7c2a64e06374091adce74adb36801ab + 187 + 1 + + + 2.0 + Community answers are created from community summaries. These answers are generated by synthesizing information from the summaries provided by the community, ensuring that the responses are comprehensive and reflective of the collective knowledge and insights within the community. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 107568a67cac472c89dfce4bbe11157c + 188 + 1 + + + 1.0 + Domain-tailored summarization is used to create community summaries + f0306814bf64f5c9e79603fc6a52f4ea + 3d78aa9d14714ac189e4020f78b15d24 + 189 + 1 + + + 1.0 + Community descriptions are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + ce0366abadef410d9b65e2bfbbf0b0f9 + 190 + 1 + + + 1.0 + Partial answers are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + f4370806deb84d0eb7e85e742e7d4bbf + 191 + 1 + + + 1.0 + Community summaries are created for each level in the hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56 + c92392d168c2443e8ed7b04992d0c92b + 192 + 1 + + + 1.0 + Community summaries are useful for understanding the global structure and semantics of the dataset + 7fb7d9ce2da9c940a32afdd87d1d9e56 + b5800c807edd4087a2420007272d15d0 + 193 + 1 + + + 1.0 + Community summaries are used to answer global queries + 843fc5421e086120ffa1c75856ecf6cd + aa247540e90d4a7abc5bca6fafaaffa1 + 194 + 1 + + + 1.0 + Community summaries are generated from root communities + 843fc5421e086120ffa1c75856ecf6cd + 34537afa1e954e08bdb52ead3a49e2f3 + 195 + 1 + + + 1.0 + Community summaries are generated from sub-communities + 843fc5421e086120ffa1c75856ecf6cd + ae043af0299f4b32a98cf187efd2a5db + 196 + 1 + + + 1.0 + Community summaries are added to the LLM context window until the token limit is reached + 843fc5421e086120ffa1c75856ecf6cd + 6016863be3414d5a92397f2d45fdfd78 + 197 + 1 + + + 1.0 + Global answers are generated from community summaries + 843fc5421e086120ffa1c75856ecf6cd + a9b900821b8444d69f432da08a77539f + 198 + 1 + + + 1.0 + The level of summary detail affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + 1fee51d6f4614127a3e1cc80d018506e + 199 + 1 + + + 1.0 + The scope of information affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + 00dc2c0748214e52bc799ca3e25204e9 + 200 + 1 + + + 1.0 + Community summaries are used for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + 42d1a9e749ad40daa34c7b0b695f8751 + 201 + 1 + + + 2.0 + Community summaries are divided into chunks of pre-specified token size + 843fc5421e086120ffa1c75856ecf6cd + 20de9a1af6ab4e88acf003cb7be0217c + 202 + 1 + + + 1.0 + Summary detail and scope affect the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + f3229f10a5a54cb1b91a26ffa6ee77a3 + 203 + 1 + + + 1.0 + Community summaries are divided into chunks + 1d07b4248c2655081c7af0e373bd70c9 + 5154b4a4f3ac43729703c69fccb54633 + 204 + 1 + + + 1.0 + Community summaries are prepared to answer user queries + 1d07b4248c2655081c7af0e373bd70c9 + 2091070e709e45f5ae56d40a9da45520 + 205 + 1 + + + 1.0 + Intermediate answers are generated from community summaries + 1d07b4248c2655081c7af0e373bd70c9 + 09045ef5c4314dde9a631a206274563f + 206 + 1 + + + 1.0 + Community summaries are part of the graph community hierarchy + 36db32c37e1987e2c5863898ad882190 + 1b9baa98ede84164883e8cdcbc7000c1 + 207 + 1 + + + 1.0 + Community summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + e4f3fcc475a74756925b730caffcb70d + 208 + 1 + + + 1.0 + Community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 7f3d5282303f4fc3a009e04f7de0ad84 + 209 + 1 + + + 1.0 + Summaries of root-level communities are used in Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + 1219a14eaf5f49ab84c9287ebf58db7a + 210 + 1 + + + 1.0 + Global sensemaking questions are evaluated over datasets in the 1 million token range + e8d83e6e7a7c0f57b218cef24976b745 + efaa386bd5e9454b87e1851cd8b28ac3 + 211 + 1 + + + 1.0 + Global sensemaking questions are directed at an entire text corpus + e8d83e6e7a7c0f57b218cef24976b745 + 073241be9b6a4952ad01dd14b94fb89c + 212 + 1 + + + 1.0 + The Python-based implementation of Graph RAG approaches will be available at this URL + e4d9b12cf2b4c691c74019eefff4fb39 + f7ac6bc4a9ca4250ad29a3adb5d08657 + 213 + 1 + + + 1.0 + Query-focused summarization is used to produce the global answer + f0306814bf64f5c9e79603fc6a52f4ea + ac2ee54e75a2492c8db372dadfccd083 + 214 + 1 + + + 1.0 + Map-reduce is used for query-focused summarization of an entire corpus + 21e52bc06a82796b1f4bcd73edda1f2a + ee895ad0b8cd40c29465e8527748d847 + 215 + 1 + + + 1.0 + Query-focused summarization is used for answering global queries + 7fb7d9ce2da9c940a32afdd87d1d9e56 + fe38c996c2d64bc899eabd6389034075 + 216 + 1 + + + 1.0 + An entity knowledge graph is derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + c0e28ae832c94405b8ddd4d2ad978be5 + 217 + 1 + + + 2.0 + In the domain of Natural Language Processing and Information Retrieval, "SOURCE DOCUMENTS" and "TEXT CHUNKS" are closely related entities. Text chunks are extracted from source documents, specifically for processing in the Graph RAG (Retrieval-Augmented Generation) approach. This method involves breaking down the source documents into smaller, manageable pieces, or text chunks, which can then be effectively utilized in various computational processes, enhancing the efficiency and accuracy of information retrieval and generation tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 7a4573a19ef94e25b4480cb4d953ae7a + 218 + 1 + + + 1.0 + Intermediate-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + 05f6639803524537b67a7f2b0c66ad23 + 219 + 1 + + + 1.0 + Low-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + 21bfd14cbc1f4cbc8ac59f7fd8c75b31 + 220 + 1 + + + 1.0 + Document corpus consists of source documents being processed + bc9e2c9e369c4108cf4f6dd5f60960f4 + c19cf2d7b067421990ab9f3acec9e736 + 221 + 1 + + + 1.0 + Partial responses are summarized to generate a final response + e8d83e6e7a7c0f57b218cef24976b745 + 3e1981b9301c4d339a9228ae7a089a04 + 222 + 1 + + + 1.0 + The LLM evaluator assesses answers based on the comprehensiveness metric + 322e02986c8724eedbcf3ebfa20b989c + 0948efa844814529b4c023aacbc23d64 + 223 + 1 + + + 1.0 + Naive RAG is evaluated for comprehensiveness + e8c8f911135faf3ff35f24107eb3f99c + fcdc0cc5ff93453eb0b94b9254760999 + 224 + 1 + + + 1.0 + Comprehensiveness is a metric used to determine the decision + e8c8f911135faf3ff35f24107eb3f99c + 0ec4ad4398a8457ab3d71bd2561858dc + 225 + 1 + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + 3c06988555334a389eab093f98679e85 + 226 + 1 + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + 81ceb8db419b4697ad24e9d7f46422ff + 227 + 1 + + + 1.0 + The smallest context window size (8k) was universally better for comprehensiveness + 4c855404ee3d3c94aa2136f1513c666f + fd05d8198d0947b39b8fa1b16f3ecf5f + 228 + 1 + + + 1.0 + The final evaluation prioritized comprehensiveness in answers + 4c855404ee3d3c94aa2136f1513c666f + d984f08ad62f47ab9aabb9aeec1b245e + 229 + 1 + + + 1.0 + Global approaches achieved higher comprehensiveness win rates + 36db32c37e1987e2c5863898ad882190 + 43603c7868164ac38c659bce7a77f45a + 230 + 1 + + + 1.0 + The LLM evaluator assesses answers based on the diversity metric + 322e02986c8724eedbcf3ebfa20b989c + 54a20cc6062d4b7193d023b6ff20461f + 231 + 1 + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + 6bb190069a704ccca3d8e1648a384185 + 232 + 1 + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + 47d2036509bf408095ab440bd052ac24 + 233 + 1 + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on diversity + 4c855404ee3d3c94aa2136f1513c666f + c20e6b1418a140389c31c7b71a6eba0c + 234 + 1 + + + 1.0 + The final evaluation prioritized diversity in answers + 4c855404ee3d3c94aa2136f1513c666f + ad96e5294247465a9c7d5ea8161dc305 + 235 + 1 + + + 1.0 + Global approaches achieved higher diversity win rates + 36db32c37e1987e2c5863898ad882190 + 25c968bf5a4f48369fded6c260f71540 + 236 + 1 + + + 1.0 + Human endeavors rely on sensemaking to understand and reason about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + 269b441634a144219f539202309bc9fb + 237 + 1 + + + 1.0 + Human endeavors rely on analyzing document collections for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + d0baf2392635468db7f5657f89eb2024 + 238 + 1 + + + 1.0 + LLMs are used to automate sensemaking in complex domains + f0306814bf64f5c9e79603fc6a52f4ea + 4f29bcf5377d4c9f94ff3f8ca2f8d941 + 239 + 1 + + + 1.0 + Microsoft uses LLMs for automating sensemaking in scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + e7072a582d9b4c1ea8b171ee940d4d6e + 240 + 1 + + + 1.0 + Ranade uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + cb6fcf84e3d04ef59b01f97ac94823a1 + 241 + 1 + + + 1.0 + Joshi uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + 97a21db5f5954e2c8868b298a3f0090e + 242 + 1 + + + 1.0 + LLM prompts are used to tailor the responses of large language models + f0306814bf64f5c9e79603fc6a52f4ea + c8f3e6cadcf34c8fafe8987e4a9b66f8 + 243 + 1 + + + 1.0 + Ranade and Joshi discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + 26c9c44e5059429bb8abc3308bc6c814 + 244 + 1 + + + 2.0 + GPT is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + 7cea9903153f43b895c0b23d25bc90a3 + 245 + 1 + + + 2.0 + Llama is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + b54436ccc23745c88d24edcc3fdd8ed1 + 246 + 1 + + + 2.0 + Gemini is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + 977c895bb98d4136a76e8749533154b6 + 247 + 1 + + + 2.0 + Kuratov et al., 2024, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + 8d75cfea884248aba1f372de5e1b82a9 + 248 + 1 + + + 2.0 + Liu et al., 2023, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + 90f4ee186bcd4996ad8002888569fffc + 249 + 1 + + + 1.0 + Sensemaking is applied in the domain of scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + 4bb78401581b4240b0967309e96af00b + 250 + 1 + + + 1.0 + Sensemaking is applied in the domain of intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + a18dd9ea4143411cb32e261db056cf0c + 251 + 1 + + + 1.0 + Klein defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + cd8d9795f540413390927ea2a9e77c26 + 252 + 1 + + + 1.0 + Klein et al. defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + 2917f3b478b04ffcacd4b47602f4d0f5 + 253 + 1 + + + 2.0 + Element instances are extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 3984bd063b384901862e68506c77cc68 + 254 + 1 + + + 1.0 + Entity references are extracted from text chunks during processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + 4137a2c7dd884bc2a8469b7fa937346c + 255 + 1 + + + 1.0 + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + 60b6bf585ccc477d830d4b69b8c7b62a + 256 + 1 + + + 2.0 + Element instances are converted into element summaries by the LLM (Large Language Model). Element summaries are created from element instances, indicating a transformation process facilitated by the LLM. This process involves the LLM taking detailed element instances and generating concise element summaries, which encapsulate the essential information from the instances. + 2c6ed90897310eea2f28e33fff1c32b0,f0306814bf64f5c9e79603fc6a52f4ea + 4330f73cb78a4bb39a384eb29112201b + 257 + 1 + + + 1.0 + Covariates are additional attributes associated with extracted element instances + bc9e2c9e369c4108cf4f6dd5f60960f4 + 45c4ed77967746e485ec9e52c0dcc0d2 + 258 + 1 + + + 1.0 + Domain-tailored summarization is used to create element summaries + f0306814bf64f5c9e79603fc6a52f4ea + 17c2cc25d00347c3bf2422d4f7a4ad7e + 259 + 1 + + + 1.0 + Element summaries include descriptions of entity nodes + 2c6ed90897310eea2f28e33fff1c32b0 + 0057fb2ddc0e4088ae5099b7ffa137da + 260 + 1 + + + 1.0 + Element summaries include descriptions of relationship edges + 2c6ed90897310eea2f28e33fff1c32b0 + d67d67cc3698438db76eb4a7f75e1ea0 + 261 + 1 + + + 1.0 + Element summaries include descriptions of claim covariates + 2c6ed90897310eea2f28e33fff1c32b0 + c23761290af24cf29adc1ee8644bdad0 + 262 + 1 + + + 1.0 + Element summaries are used to understand the structure and semantics of graph communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + de51b828ce1f442bbb19a7b20bce9dda + 263 + 1 + + + 1.0 + Element summaries include descriptions of nodes + 843fc5421e086120ffa1c75856ecf6cd + 4a3ff6a3471945fd8c7fd5c171c56d56 + 264 + 1 + + + 1.0 + Element summaries include descriptions of edges + 843fc5421e086120ffa1c75856ecf6cd + 31bb84eb2a834dabacc0ed51af4fcefd + 265 + 1 + + + 1.0 + Element summaries include descriptions of covariates + 843fc5421e086120ffa1c75856ecf6cd + 5070012e83e7442381bcba1cdacdb7d8 + 266 + 1 + + + 1.0 + Sub-community summaries are used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + 5eda9074df124f5497f17b61badd52ac + 267 + 1 + + + 2.0 + Community detection is a technique used to identify graph communities. Graph communities are groups of nodes within a graph that are more densely connected to each other than to the rest of the graph. This process of identifying such communities is crucial for understanding the structural dynamics and relationships within complex networks, particularly in the domain of Natural Language Processing and Information Retrieval. By leveraging community detection algorithms, researchers can uncover hidden patterns and insights within large datasets, facilitating more effective data analysis and interpretation. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 4cf4107b0e2842778aaa658a1a85f3b3 + 268 + 1 + + + 1.0 + Global answer is created from community answers + f0306814bf64f5c9e79603fc6a52f4ea + 7f4857f94b4e4e49be7236a42071e167 + 269 + 1 + + + 2.0 + Global answers are generated in response to user queries + 843fc5421e086120ffa1c75856ecf6cd + d21a1fef903f4a399bd3cd366aad3c9e + 270 + 1 + + + 1.0 + Global answer is generated by sorting intermediate answers based on helpfulness scores + 1d07b4248c2655081c7af0e373bd70c9 + fc596a598ff74a4c843e405b597551b5 + 271 + 1 + + + 1.0 + Intermediate answers are combined to form the global answer + 1d07b4248c2655081c7af0e373bd70c9 + e2aacff6b4404574b818e7a3ece57b5b + 272 + 1 + + + 1.0 + The final context window is used to generate the global answer + 1d07b4248c2655081c7af0e373bd70c9 + 2ec5cae98c7a485881f0680fbca6d67f + 273 + 1 + + + 1.0 + Graph RAG pipeline operates at indexing time + f0306814bf64f5c9e79603fc6a52f4ea + c87b815d61af448596d3194a804b57b3 + 274 + 1 + + + 1.0 + Graph RAG pipeline operates at query time + f0306814bf64f5c9e79603fc6a52f4ea + 2f92fc82c3b74417896bad3bd8e61f5e + 275 + 1 + + + 1.0 + Nodes are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + fb61c68efe5b4d69a9623e531e7c639c + 276 + 1 + + + 1.0 + Edges are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + dc61e34c1ca8419e923aeeff7d83d949 + 277 + 1 + + + 1.0 + Covariates are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + 697fb824eef34759852f1d5588921aec + 278 + 1 + + + 1.0 + Leiden method is used in the graph RAG pipeline for community detection + f0306814bf64f5c9e79603fc6a52f4ea + b872fcc5b18a4f32b976f4693f22e88e + 279 + 1 + + + 1.0 + Graph RAG pipeline uses the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + 64be9b98299f4d349e0f4358685ca235 + 280 + 1 + + + 1.0 + The Podcast dataset graph consists of 8564 nodes + 36db32c37e1987e2c5863898ad882190 + 8302a03f6ede471bb955c0bbf44a4b3c + 281 + 1 + + + 1.0 + The News dataset graph consists of 15754 nodes + 36db32c37e1987e2c5863898ad882190 + a02263dd89964a1c8ab2d0e9aba0f4eb + 282 + 1 + + + 1.0 + The Podcast dataset graph consists of 20691 edges + 36db32c37e1987e2c5863898ad882190 + 6b7aa6ce4cac4edbaaab831286e67e5e + 283 + 1 + + + 1.0 + The News dataset graph consists of 19520 edges + 36db32c37e1987e2c5863898ad882190 + 655d40ea08e348ad94ae49785797da90 + 284 + 1 + + + 1.0 + Traag contributed to the development of the Leiden method + f0306814bf64f5c9e79603fc6a52f4ea + 254cea99330f4f2aa062c771146da7ea + 285 + 1 + + + 2.0 + Traag et al. are the authors of the Leiden algorithm and developed the Leiden method. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + a2836232227c4e3383d166db860cb2a3 + 286 + 1 + + + 2.0 + Leiden is a specific type of community detection algorithm used in various analytical pipelines. It is designed to identify and map out the structural dynamics within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. The algorithm is known for its efficiency and accuracy in detecting community structures, making it a valuable tool for researchers and practitioners in the field. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + 8a9247ee9bac45bdbf69c9d0bb8419b5 + 287 + 1 + + + 1.0 + Leiden is known for its ability to recover hierarchical community structures efficiently + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 757a0f78fcdd4bf6b8326a75fcee9e15 + 288 + 1 + + + 1.0 + The Leiden algorithm is used to detect graph communities in the MultiHop-RAG + 7fb7d9ce2da9c940a32afdd87d1d9e56 + b5235cb24b8f440389f250ebd5b6e2f8 + 289 + 1 + + + 1.0 + Figure 3 shows graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + bdee1849252749efa2e671ed87641f61 + 290 + 1 + + + 1.0 + Lewis contributed to the development of the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + 057641c1476247958d8c357e17095d8e + 291 + 1 + + + 1.0 + Lewis et al. developed the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + b61dfd0b24664f37af4046bdf0cb7b19 + 292 + 1 + + + 2.0 + Lewis et al., 2020, are the authors who established the RAG approach + fb3c48579608fa28be585ceb6cd2f0fe + 0bc00f14e6194df7b0fe9ef9ba28d34f + 293 + 1 + + + 1.0 + Kevin Scott is the CTO of Microsoft + 1d07b4248c2655081c7af0e373bd70c9 + b823c5d22037423da919eee6c35c4c8b + 294 + 1 + + + 2.0 + Microsoft conducted a study on the impact of large language models on scientific discovery using GPT-4 + 833e7d67dcd30790b26b71c9b5306f6b + cd7f555e4ab948ba94bade14e262ff84 + 295 + 1 + + + 1.0 + Preprint is available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 86cd53087b2542f898d6cecca31e6145 + 296 + 1 + + + 1.0 + Baumel, T. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 5dc3480806b04fdd8089a3be46e22540 + 297 + 1 + + + 1.0 + Eyal, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 50c91820a91f488d8606198540aba894 + 298 + 1 + + + 1.0 + Elhadad, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + a38eace89e7e40de8f007fde24597e9e + 299 + 1 + + + 1.0 + Es, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 5d75097d065e4b049a1678deab40949b + 300 + 1 + + + 1.0 + James, J. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + c277134d380a42cd886a14a953554792 + 301 + 1 + + + 1.0 + Espinosa-Anke, L. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + b680be879404440885b1d3af5b9af583 + 302 + 1 + + + 1.0 + Schockaert, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 4cc609b1a64a442aac6b72078a315ac6 + 303 + 1 + + + 1.0 + Feng, Z. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + a3ee323c9c9a4f81b5907030122b80d2 + 304 + 1 + + + 1.0 + Feng, X. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 19aa5f0b738c4f4a96668c80c3e93331 + 305 + 1 + + + 1.0 + Zhao, D. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + f8402b10349f4db888ac4fb6fd81723a + 306 + 1 + + + 1.0 + Yang, M. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 5927f9089289429da4adf2bbd5641e44 + 307 + 1 + + + 1.0 + Qin, B. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 60724b8b268044b69a4b3d939f1757e2 + 308 + 1 + + + 1.0 + LangChain is an organization that has published on arXiv + 71f6daf11e64e5273a3847d46bf228e1 + d931685d35e149909472f736114ca62f + 309 + 1 + + + 1.0 + Wang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 3f5e9927a4114a958d75f5ed313526a8 + 310 + 1 + + + 1.0 + Khramtsova, E. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 4728bf0cb7564bbd85c90ceaa846f290 + 311 + 1 + + + 1.0 + Zhuang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + cdac6338c3234797a0d3a32cd68d1b2e + 312 + 1 + + + 1.0 + Zuccon, G. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 372f78df13f9452b84d898c703a1ba95 + 313 + 1 + + + 1.0 + Wang, Y. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 7af06d2b32a941a4b044579a7c423371 + 314 + 1 + + + 1.0 + Lipka, N. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + a10b8fad74744ae981747dadf7234b78 + 315 + 1 + + + 1.0 + Rossi, R. A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0cb2118ecc87439a91409deef7ef9830 + 316 + 1 + + + 1.0 + Siu, A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + ea27218042d640fd81c23eb64aff6b46 + 317 + 1 + + + 1.0 + Zhang, R. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 9e5d626681094933abf87cf797f2fa46 + 318 + 1 + + + 1.0 + Derr, T. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 545358ff14f84601a22e9f39f5ef1534 + 319 + 1 + + + 1.0 + Xu, Y. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 1b0e7dbc7c5944a7833f6540bde1fa4f + 320 + 1 + + + 1.0 + Lapata, M. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0c0f2d8c623949f1ae89c67d0753aeab + 321 + 1 + + + 1.0 + Zhang, J. published the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 20c3844c80a140ac97b62dc444feee41 + 322 + 1 + + + 1.0 + Zhang, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + c5fac1bea509464d9dc934275d938039 + 323 + 1 + + + 1.0 + Gan, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 45b64fbddd8f4abdb86a9c3c6f53f802 + 324 + 1 + + + 1.0 + Yao, L. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0e504b58cbda4d9188050bc43004c01f + 325 + 1 + + + 1.0 + Wang, C. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + c06bd37e120e4af49ec8bd6ce399473b + 326 + 1 + + + 1.0 + Zheng, L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 5d507985f2f540d8a1fa2d1191eae2a8 + 327 + 1 + + + 1.0 + Chiang, W.-L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 8e0b5b4011d74bbb8dc09fa05d88369c + 328 + 1 + + + 1.0 + Sheng, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 5d8184f5d52040d8bb67d1a6b889e9fe + 329 + 1 + + + 1.0 + Wu, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + b3bf669489ae4913bb60ddfe50e41697 + 330 + 1 + + + 1.0 + Zhuang, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0eba9d55a3ff46298665a0c292e2237f + 331 + 1 + + + 1.0 + Lin, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 55e3f4a200eb4619ae2b6efb645464d1 + 332 + 1 + + + 1.0 + Li, D. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + c44324c171674d00a743413042e9b944 + 333 + 1 + + + 1.0 + Xing, E. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 4bdaba79a3274241ab98e27aeaf98f57 + 334 + 1 + + + 1.0 + Preprint is classified under cs.CL on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 7c8c464ed7044a7896adfeb35f58a04d + 335 + 1 + + + 1.0 + Preprint was submitted on 24 Apr 2024 + f0306814bf64f5c9e79603fc6a52f4ea + 5fa2eec73bec481b85eba22ea7a2a927 + 336 + 1 + + + 1.0 + Preprint has the identifier 2404.16130v1 on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + e6aa5eedca984c56b5fa5e179127951d + 337 + 1 + + + 1.0 + Community detection results in the partition of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1c4bd4ba4ef64a93acd55faa8fd97ca9 + 338 + 1 + + + 1.0 + The pipeline includes a step for community detection + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 5b85c70d578c4d67b5cb4743552bd559 + 339 + 1 + + + 2.0 + Dang, 2006, is the author who established the QFS approach + fb3c48579608fa28be585ceb6cd2f0fe + 956113fb770840c38bce65bb5832f988 + 340 + 1 + + + 2.0 + Baumel et al., 2018, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + 785bb55e79954b0c84a4a53cd7f0b454 + 341 + 1 + + + 2.0 + Laskar et al., 2020, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + 1239281fd3774b91a99358c9c1e6ee1c + 342 + 1 + + + 2.0 + Yao et al., 2017, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + 32b29a842b224f4c99fa1d5c764efc9a + 343 + 1 + + + 2.0 + Goodwin et al., 2020, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + f5ae7dc11fd64822a3a15e7d3839031a + 344 + 1 + + + 2.0 + Laskar et al., 2022, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + e1e254e67719488894eaa3553112a8cf + 345 + 1 + + + 2.0 + Liu and Lapata, 2019, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + ebdd79169d7d41b99faf09b039a66204 + 346 + 1 + + + 2.0 + Achiam et al., 2023, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + e036534e17b24dd2895167a20873230f + 347 + 1 + + + 2.0 + Brown et al., 2020, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + a00bc5e4be634b08b1f084b6a07abafd + 348 + 1 + + + 2.0 + Touvron et al., 2023, are the authors who worked on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + ce8241c964724429bb361b7b53867007 + 349 + 1 + + + 2.0 + Anil et al., 2023, are the authors who worked on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + 61cd7f168f7f44d6a23415e9497f1e65 + 350 + 1 + + + 1.0 + Modularity is an inherent quality of knowledge graphs + 21e52bc06a82796b1f4bcd73edda1f2a + 3be77a7b57e34c55acc1f1dfbc64ee10 + 351 + 1 + + + 1.0 + Brown et al. (2020) discuss in-context learning with few-shot examples + bc9e2c9e369c4108cf4f6dd5f60960f4 + 751c564f8ff6444d9d4c8de4a677e655 + 352 + 1 + + + 1.0 + Kuratov et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + 96963c158fb64680bded290f442ff9aa + 353 + 1 + + + 1.0 + Liu et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + bdbfbde5dd244447a2a0674b30ae3e8f + 354 + 1 + + + 1.0 + Louvain is a type of community detection algorithm + 21e52bc06a82796b1f4bcd73edda1f2a + f970bfe31db74929abff6ea38e5d18e6 + 355 + 1 + + + 1.0 + Community detection algorithms are used to partition the graph index into communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 6f0c2a8b79e6406a8ab7a20864ae2ce2 + 356 + 1 + + + 1.0 + Fortunato has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 7b09e60e33f44ffdab9c656c5b9c1d50 + 357 + 1 + + + 1.0 + Jin et al. have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 8b7beab7c0a143aea7bffc31df7528d5 + 358 + 1 + + + 1.0 + HotPotQA dataset is used to evaluate the entity extraction prompt with gpt-4-turbo + 21e52bc06a82796b1f4bcd73edda1f2a + d03eb34a0612420680555ab9f10d03d5 + 359 + 1 + + + 1.0 + Yang et al. (2018) introduced the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + b066746cdff7440c8a3591f0c098201d + 360 + 1 + + + 2.0 + Yang et al. are the authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + 1e2eded8ef7b4b458c33fbc2d36c4380 + 361 + 1 + + + 1.0 + GPT-4-Turbo was tested with varying context window sizes + 4c855404ee3d3c94aa2136f1513c666f + c59e3e931b0f4cf888c2eb70857ee753 + 362 + 1 + + + 1.0 + Tech journalist uses podcast transcripts to look for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + 305b80bb4df5488b8a34129daeeae0c7 + 363 + 1 + + + 3.0 + Kevin Scott is a participant in the podcast conversations compiled in the Podcast Transcripts dataset. His conversations are included as part of the podcast transcripts, contributing to the overall content and discussions captured within this dataset. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + 66fa0de756da440bad8da583306410c4 + 364 + 1 + + + 1.0 + Technology leaders participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + d1e9c550a0e74c48ae81c319f26ccafc + 365 + 1 + + + 2.0 + RAG systems are used to evaluate the Podcast Transcripts dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + 3730b5d759ba4fd28a54af0a02151f09 + 366 + 1 + + + 2.0 + C0 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 82b7f7c27e2348f880c94ffb80942de7 + 367 + 1 + + + 2.0 + C1 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 0980c4f558654466b4d691d0cb7ce16d + 368 + 1 + + + 2.0 + C2 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + f1e47cf5daa441649c3474c3339bb704 + 369 + 1 + + + 2.0 + C3 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 0964dcfbff934c92af8961155673ac7f + 370 + 1 + + + 2.0 + TS is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 51b82bcdffe04056bad1c082c3830047 + 371 + 1 + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + c62bb148852b49a98e2779ca23a0919d + 372 + 1 + + + 1.0 + SS is a category used in the analysis of podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + 72b5a0c357c24b739084d501b9354bc1 + 373 + 1 + + + 1.0 + Units are used to measure the context in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + c827b62ebf134e55a3ccf0b63f976870 + 374 + 1 + + + 1.0 + Tokens are used to measure the word count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + b51ef388758845e880e736309ae791e3 + 375 + 1 + + + 1.0 + % Max is used to measure the percentage of maximum token count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + 0a841cd4b6664423b033f22e3a80f33c + 376 + 1 + + + 1.0 + Both are datasets used in the analysis + 36db32c37e1987e2c5863898ad882190 + 16911c51c65b42f8a2d04c05f45b2c58 + 377 + 1 + + + 1.0 + Educator uses news articles to incorporate current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + fc3f77f29574410d991a2aa333950bf6 + 378 + 1 + + + 2.0 + RAG systems are used to evaluate the News Articles dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + 4f847eb72cbe48678d5634dcf93fc0e2 + 379 + 1 + + + 1.0 + C0 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + 829e64159ae04301982e88e93a2f0e49 + 380 + 1 + + + 1.0 + C1 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + cf37d3d4bc154f65b3d79c831c587763 + 381 + 1 + + + 1.0 + C2 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + 4b4fce341d554012bc73d7886860749e + 382 + 1 + + + 1.0 + C3 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + 9f6e7a08bd814d19b45fac58928027f8 + 383 + 1 + + + 1.0 + TS is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + ff9410fed5e64c04a875e040e3d182b2 + 384 + 1 + + + 1.0 + Units are used to measure the context in news articles + 36db32c37e1987e2c5863898ad882190 + 1161272728914953b568f384d7a9f2f1 + 385 + 1 + + + 1.0 + Tokens are used to measure the word count in news articles + 36db32c37e1987e2c5863898ad882190 + f09c82eb89944ae9846df82135123b90 + 386 + 1 + + + 1.0 + % Max is used to measure the percentage of maximum token count in news articles + 36db32c37e1987e2c5863898ad882190 + d221b743a51d464b87de3b72b85f6b59 + 387 + 1 + + + 1.0 + Map-reduce is the method used in the text summarization condition + 973164fa90bf2b4ee267f4fd795916bf + 9fd31a28e1384b40a9d1658a765871cd + 388 + 1 + + + 1.0 + The LLM evaluator assesses answers based on the empowerment metric + 322e02986c8724eedbcf3ebfa20b989c + 0119f233c8394b9584e55fadcce173f0 + 389 + 1 + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for news articles + ebf5249c888e07fedce6572a4c03f88c + 5c20b469b92446dabb1b68976807be7c + 390 + 1 + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on empowerment + 4c855404ee3d3c94aa2136f1513c666f + 2c2392247a35456da663adfcffd12e73 + 391 + 1 + + + 1.0 + Empowerment has an average win rate of 51.3% + 36db32c37e1987e2c5863898ad882190 + 167a32ff67ce4471baa8cf019ee7c17b + 392 + 1 + + + 1.0 + Naive RAG mentions Taylor Swift as a public figure + e8c8f911135faf3ff35f24107eb3f99c + 3280fc12ef414827838e6ac7089f0618 + 393 + 1 + + + 1.0 + Naive RAG mentions Travis Kelce as a public figure + e8c8f911135faf3ff35f24107eb3f99c + 556fba72a0854ce4831f6cfea6fd035e + 394 + 1 + + + 1.0 + Naive RAG mentions Britney Spears as a public figure + e8c8f911135faf3ff35f24107eb3f99c + 8e2e6eeed5a04c9f80efbcfc624ced95 + 395 + 1 + + + 1.0 + Naive RAG mentions Justin Timberlake as a public figure + e8c8f911135faf3ff35f24107eb3f99c + ea6d546f1caa4b4aaacdad8b8af195ec + 396 + 1 + + + 1.0 + Naive RAG is determined to be the loser based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + 267ce44e6dae43ee94d0d375ec08ef17 + 397 + 1 + + + 1.0 + Global approaches consistently outperformed the naive RAG + 36db32c37e1987e2c5863898ad882190 + b37e5d15f3154ee39df016b8eac8de66 + 398 + 1 + + + 1.0 + Naive RAG produces the most direct responses + 36db32c37e1987e2c5863898ad882190 + e13eb574e885414b80f0b66992767ef2 + 399 + 1 + + + 1.0 + SS represents naive RAG in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 93f4140f654e41ccba908c6f6dc65f17 + 400 + 1 + + + 1.0 + Gao et al., 2023 discusses naive RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + a102d091986749ef90b45d411e707bef + 401 + 1 + + + 1.0 + Community partitions enable divide-and-conquer global summarization + 7fb7d9ce2da9c940a32afdd87d1d9e56 + cd6ae38a5a6742899d14f4a064f42c19 + 402 + 1 + + + 1.0 + Global summarization can be performed using a graph-free approach + e4d9b12cf2b4c691c74019eefff4fb39 + fe18688bd4ef44d1a184ec6d1451a5cf + 403 + 1 + + + 1.0 + Source texts are used in global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + 0f1282bdfedb4f6e8765007a90dd2959 + 404 + 1 + + + 1.0 + Final global answer is generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + 540af5c5d4cd41ceb29c40c5fb02e2fe + 405 + 1 + + + 1.0 + Short descriptions are used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + bbf83708095f47019eaee93d6879bc77 + 406 + 1 + + + 1.0 + Low-level community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 245a56f01d1b48a7b4d88ed0e354155a + 407 + 1 + + + 1.0 + The use of rich descriptive text for homogeneous nodes in a graph index aligns with the capabilities of LLMs + 7fb7d9ce2da9c940a32afdd87d1d9e56 + d3aa564fb4eb430a8ca6813ca76bfff6 + 408 + 1 + + + 1.0 + Graph indexes differ from typical knowledge graphs in their use of rich descriptive text instead of concise knowledge triples + 7fb7d9ce2da9c940a32afdd87d1d9e56 + d9b948357d96419ca135065ce1c360ef + 409 + 1 + + + 1.0 + The graph index supports condition C0 + 973164fa90bf2b4ee267f4fd795916bf + 20a79ddd91ba48e4bb7bc194c79baaf6 + 410 + 1 + + + 1.0 + The graph index supports condition C1 + 973164fa90bf2b4ee267f4fd795916bf + b95728a0b96b405cbccafa6c12fd8722 + 411 + 1 + + + 1.0 + The graph index supports condition C2 + 973164fa90bf2b4ee267f4fd795916bf + 5d6dc034d2014e8c930fde69c31b99cf + 412 + 1 + + + 1.0 + The graph index supports condition C3 + 973164fa90bf2b4ee267f4fd795916bf + 127cbb53940f4efa8e1807b4452375ba + 413 + 1 + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + f1ea6ef9539043ab887bcce22ccf9625 + 414 + 1 + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + b50c4f053f0546029c4095b7b93aa05e + 415 + 1 + + + 1.0 + The graph index was created using generic prompts for entity and relationship extraction + 973164fa90bf2b4ee267f4fd795916bf + 0cea7f7a7fab49339cdd6fb02d0d183e + 416 + 1 + + + 1.0 + Few-shot examples tailored to the domain of the data were used in the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + 5b89f0d8101c419b86e1959cca2db848 + 417 + 1 + + + 1.0 + The graph indexing process used a context window size of 600 tokens + 973164fa90bf2b4ee267f4fd795916bf + cdb407fc600b45caa6f94f82e89d2e4f + 418 + 1 + + + 1.0 + The decision to build a graph index depends on the expected number of lifetime queries per dataset + e4d9b12cf2b4c691c74019eefff4fb39 + 7f4905fcb43e4d6ca23e6d2b40f6958e + 419 + 1 + + + 1.0 + The decision to build a graph index depends on the value obtained from it + e4d9b12cf2b4c691c74019eefff4fb39 + f5ad4fe84df544c69db25f0e30c6eace + 420 + 1 + + + 1.0 + The decision to build a graph index depends on the value obtained from other graph-related RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + 237a46cc973b41dc9af4190c71c5c9e1 + 421 + 1 + + + 1.0 + Recall measures the completeness of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + aaa27aa0b1024e3aa3c87a6ec821a348 + 422 + 1 + + + 1.0 + Precision measures the accuracy of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + f0a28fe3f68546dba7850815f7933275 + 423 + 1 + + + 1.0 + Few-shot examples are used to tailor the default prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + 45b59feba7134bc18632cb42530c189a + 424 + 1 + + + 1.0 + Few-shot examples are used to tailor the secondary extraction prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + 7747cd2048f94d378e83265b9561d921 + 425 + 1 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of science + 2c6ed90897310eea2f28e33fff1c32b0 + c4e9532dbc734264a0e3e827bc8014c6 + 426 + 1 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of medicine + 2c6ed90897310eea2f28e33fff1c32b0 + 003e5d505a01434596c6d65ff20b0bdf + 427 + 1 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of law + 2c6ed90897310eea2f28e33fff1c32b0 + f79358f3535045d9aad3b828df59293b + 428 + 1 + + + 1.0 + A single extraction round is part of the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + 7d375c18c1e2415faecd9f7397068a32 + 429 + 1 + + + 1.0 + Domain refers to the specific area of knowledge of the document corpus + bc9e2c9e369c4108cf4f6dd5f60960f4 + dfa0e847a6704c93a0fe014b01858ff7 + 430 + 1 + + + 1.0 + Covariate prompts are used to extract claims linked to detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + 9e91823feb174cd1b6a3bf8d0a5cb86b + 431 + 1 + + + 1.0 + Source text span is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + ad76c8dc8dd94412a5e79005cf8e0f2f + 432 + 1 + + + 1.0 + Start date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 26a03482961e41918ea049018080af7a + 433 + 1 + + + 1.0 + End date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 1cfd220ff4d2493ca4b92d725d171d32 + 434 + 1 + + + 1.0 + Description is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 97738fe0830d405ba53598b5cb1e5e38 + 435 + 1 + + + 1.0 + Subject is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 354cea4f6e164a48ad12122c28a5b30d + 436 + 1 + + + 1.0 + Object is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 1ee2380c1eda4ebb8c9304820750ac88 + 437 + 1 + + + 1.0 + Communities of entities help manage variations in a noisy graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 57e00d4d4e0e4679a150f048deb80af3 + 438 + 1 + + + 1.0 + Common entities are described using rich descriptive text for homogeneous nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + c1e4a9dbe55c4fb89f0d927c9fb067a4 + 439 + 1 + + + 1.0 + LLMs are used to generate metrics for evaluating natural language generation + 973164fa90bf2b4ee267f4fd795916bf + 1474a72a5cff4b72ae6f99e804ceaa95 + 440 + 1 + + + 1.0 + Wang et al. (2023) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + 738fda68df7a49a0bae96673a8711afc + 441 + 1 + + + 1.0 + Zheng et al. (2024) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + 89dd5a0943c64247adae624abbc95afb + 442 + 1 + + + 1.0 + Relationship edges connect homogeneous nodes in a graph + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 405e9907440d4deab71f3960ae36f47b + 443 + 1 + + + 1.0 + Edge weights represent the normalized counts of detected relationship instances on relationship edges + 7fb7d9ce2da9c940a32afdd87d1d9e56 + f91e7c9600ca4623a8cc4a56d2dccd07 + 444 + 1 + + + 1.0 + Each level of the hierarchical community structure provides a community partition + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 3af2a8619c394be6adf06e4bc742b7ec + 445 + 1 + + + 2.0 + The hierarchical community structure is a framework used to organize and understand the relationships and dynamics within specialized communities. Root communities are an integral part of this structure, serving as the top-level communities. These root communities form the foundational layer in the hierarchical community structure, providing a basis for further subdivision and organization of more specific sub-communities. This hierarchical approach allows for a systematic analysis of complex networks, facilitating a deeper understanding of the interconnections and dependencies within the domain of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + c10ffc51dcb54708a1dc757693010bfe + 446 + 1 + + + 2.0 + The hierarchical community structure is a framework that organizes communities into different levels, with sub-communities representing the lower-level communities within this structure. Sub-communities are integral components of the hierarchical community structure, indicating that they are nested within larger communities and contribute to the overall organization and dynamics of the community. This hierarchical arrangement allows for a more detailed and nuanced understanding of the relationships and interactions within the community, facilitating more effective analysis and mapping of complex text data, particularly in specialized domains such as Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + e67ce34d48364422973ccf3a6b57af83 + 447 + 1 + + + 1.0 + Community levels are part of the hierarchical community structure + 843fc5421e086120ffa1c75856ecf6cd + 98773a34c9bb474d8a789ea08f57250e + 448 + 1 + + + 1.0 + The Leiden algorithm is used to detect communities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + ae260498423e4d55aa413423cd0eb20b + 449 + 1 + + + 1.0 + OpenORD is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 4aeecb9d885743ca9373337a43957dd8 + 450 + 1 + + + 1.0 + Force Atlas 2 is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 1121b50f7858427fa679d81861238825 + 451 + 1 + + + 1.0 + Nodes represent entities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 6e3c8aa3abab475bb0148faa9112f0bf + 452 + 1 + + + 1.0 + Edges represent connections between nodes in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 948a00e8ee1246cc90c47b292d03ddff + 453 + 1 + + + 1.0 + Covariates are variables linked to nodes and edges in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 45c42e619f5e488f914608780dcf0579 + 454 + 1 + + + 2.0 + Tang and Yang are the authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 2b3bea0d9ede41f193828526bcb8e02c + 455 + 1 + + + 1.0 + Questions are generated based on the target datasets + 1d07b4248c2655081c7af0e373bd70c9 + 6b2586cc1f8e4dc8af64913af63d9837 + 456 + 1 + + + 1.0 + N represents the number of test questions per dataset + 973164fa90bf2b4ee267f4fd795916bf + 7983bfa8d173414685272b3844d6612e + 457 + 1 + + + 1.0 + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + 09294e8220a445e288ea8841f234a440 + 458 + 1 + + + 1.0 + Root communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + d4e043cf972c4d129b6b855f1731caae + 459 + 1 + + + 1.0 + Level 0 represents the root-level communities in the hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + e0d63137270c426dbbfe7fcf78c474de + 460 + 1 + + + 1.0 + Reports provide detailed information about specific subtopics within sub-communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + c50bca18bc454a98b935df012b7fd6f9 + 461 + 1 + + + 1.0 + Sub-communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + 434b133c64bd46219e67c6eb296ad0ff + 462 + 1 + + + 1.0 + Level 1 represents sub-communities within the root-level communities + 843fc5421e086120ffa1c75856ecf6cd + cb895bf7e7c147e6b5d923b6c8f67d63 + 463 + 1 + + + 1.0 + Partitions can be organized into a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 87776e869a01402499a317cb9cf09453 + 464 + 1 + + + 1.0 + Level 0 is the root level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + d1e5359d2e344260bf1b83823df839b7 + 465 + 1 + + + 1.0 + Level 1 is a sub-level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 0522f6580b824bc39792b695fc8be66b + 466 + 1 + + + 1.0 + The token limit defines the maximum number of tokens in the LLM context window + 843fc5421e086120ffa1c75856ecf6cd + 580fd6d19460460fa40613f66b3ee200 + 467 + 1 + + + 1.0 + Prominence is used to prioritize community edges + 843fc5421e086120ffa1c75856ecf6cd + 84f4684a7a5241c18bb087ccb00550d3 + 468 + 1 + + + 1.0 + Combined source and target node degree is used to measure prominence + 843fc5421e086120ffa1c75856ecf6cd + 9607ba4a796f46be8d4f79bc7065d60b + 469 + 1 + + + 1.0 + Chunks are divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + 236dd7dce9ee4cf5918fddd44b4863e5 + 470 + 1 + + + 1.0 + Helpfulness scores are assigned to intermediate answers + 1d07b4248c2655081c7af0e373bd70c9 + 9e92fed814a64d9d88bfab9a227859d3 + 471 + 1 + + + 1.0 + Tech journalist is interested in episodes dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + 7dccecb29d3a419093b279b22e207539 + 472 + 1 + + + 1.0 + Tech journalist is interested in how guests perceive the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + 89857eb61e63461cbad7c5014f5098f9 + 473 + 1 + + + 1.0 + Tech journalist is interested in discussions about the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + 7b2e7a0d910c4988a7b64489f4159a65 + 474 + 1 + + + 1.0 + Tech journalist is interested in suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + 38630cf0996f4cff8d32b2dbdaa5ba85 + 475 + 1 + + + 1.0 + Tech journalist is interested in discussions about collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + bd0fb68ac7014b91a314c93ec55897f5 + 476 + 1 + + + 1.0 + Educator is interested in current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + f24dcb3cd6d644f8af2b6c47983e280b + 477 + 1 + + + 1.0 + Educator is interested in how news articles address the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + b1cad695afbc4ec3bbcd46ea34bd26ca + 478 + 1 + + + 1.0 + Educator is interested in examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + 72f7974758d74e5d89ddb64ad739abb8 + 479 + 1 + + + 1.0 + Educator is interested in insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + e6ee83249adf4e14b98d1676b1c6b05f + 480 + 1 + + + 1.0 + Educator is interested in highlighting the importance of health literacy through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + f805fd9fe42947a38b92a3db6e8cc986 + 481 + 1 + + + 1.0 + The size of the context window and the prompts used for answer generation are the same across all conditions + 973164fa90bf2b4ee267f4fd795916bf + e8b956218d5c4e5d9d390abcf527a514 + 482 + 1 + + + 1.0 + The task is an activity or goal that the user aims to achieve + 1d07b4248c2655081c7af0e373bd70c9 + 9525aa223d774e62ad856c2201cfab1b + 483 + 1 + + + 1.0 + Questions are generated based on the user's task + 1d07b4248c2655081c7af0e373bd70c9 + 1087596b06d1400a8f863d0ac1af64a4 + 484 + 1 + + + 1.0 + Datasets were used in combination with questions for the analysis + 4c855404ee3d3c94aa2136f1513c666f + 39058965295643c8a7738350cc18ceac + 485 + 1 + + + 1.0 + Questions were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + 9a8a2e5e3f2645619a0403532d935afe + 486 + 1 + + + 2.0 + Zheng et al. are the authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + f0c21c67baac47f097f74f5055b89877 + 487 + 1 + + + 1.0 + Zheng, L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 323a4c7407ac401db79a6023c3a5a17d + 488 + 1 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 686bc2bd59644e398dde88ffd37bf49b + 489 + 1 + + + 1.0 + Sheng, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + e368f8e9c9864acc880fdb5113631f3f + 490 + 1 + + + 1.0 + Zhuang, S. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 05063c19ddb847a89ae1746588464288 + 491 + 1 + + + 1.0 + Wu, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 019b34e800414f7b87f38a14adf2eb67 + 492 + 1 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 1064a663ca4742a78e743128546f6d87 + 493 + 1 + + + 1.0 + Lin, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 9a5e0a4ae34f46b39a5a028cbc135264 + 494 + 1 + + + 1.0 + Li, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 5a224002ecbc4725abeb5a424aaca6a6 + 495 + 1 + + + 1.0 + Li, D. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 8826a17bbda34012b3ea84d58ae531eb + 496 + 1 + + + 1.0 + Xing, E. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + bab69d76defb402da2a2a358739f1497 + 497 + 1 + + + 1.0 + MT-Bench and Chatbot Arena are both tools used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + ea465e5cd92247829f52ff0c8591d1bb + 498 + 1 + + + 2.0 + Koesten et al. authored a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + 2dbac25b512c4f21965169a95a910a94 + 499 + 1 + + + 2.0 + Xu and Lapata authored a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + 97958ed004f645b1b331fa0e66faa313 + 500 + 1 + + + 1.0 + Text summarization method applies a map-reduce approach directly to source texts (TS) + 973164fa90bf2b4ee267f4fd795916bf + 48129b4ee99f4e30843fd4395d4815c0 + 501 + 1 + + + 1.0 + Text summarization is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + 6de4c00e48b3480883e696e24df9fda4 + 502 + 1 + + + 1.0 + Semantic search RAG is a na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached (SS) + 973164fa90bf2b4ee267f4fd795916bf + 4b3d236101de4904ab348e3e3b11b4be + 503 + 1 + + + 1.0 + Semantic search RAG is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + 5be2ce9957ba404f939b6c8175015619 + 504 + 1 + + + 1.0 + C0 uses root-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + fe77344850214c1cac923094de81098c + 505 + 1 + + + 1.0 + C0 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 798f739abfc14a13bf3911d0a9cfb63b + 506 + 1 + + + 1.0 + C0 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 3105de8188fd41d88d0dbf0a5d48e443 + 507 + 1 + + + 1.0 + C0 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + 8108dde0e62a48008a270138a690a0b9 + 508 + 1 + + + 1.0 + C1 uses high-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + d07207b853c14504a44eea1d4778f902 + 509 + 1 + + + 1.0 + C1 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 406822a1a01a4140baf9bbf1d479f07e + 510 + 1 + + + 1.0 + C1 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + fe47ba3762ae4feda39904d59cbb4160 + 511 + 1 + + + 1.0 + C1 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 53032c2afcb5474a88446ad7c5506980 + 512 + 1 + + + 1.0 + C1 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + 5c66a88612a245cb91fbba9c094f12fc + 513 + 1 + + + 1.0 + C2 uses intermediate-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + b4c54fb7ce0b4b77afd5fbe5a8a2527f + 514 + 1 + + + 1.0 + C2 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + c8b60cdb74104667b5d2b4b70d74d039 + 515 + 1 + + + 1.0 + C2 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + 333e294d7cc34df4abc47ad9ced3d186 + 516 + 1 + + + 1.0 + C2 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 15e66e10d12f4520abca20985d2cb39c + 517 + 1 + + + 1.0 + C2 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + 2a271d9b5d7b46fea4046d5590eed1d7 + 518 + 1 + + + 1.0 + C3 uses low-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + 99e372089bed4a0394af57175679f8e4 + 519 + 1 + + + 1.0 + C3 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 62afe93767684ea38f861d20fb05ff71 + 520 + 1 + + + 1.0 + C3 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + 8fc1fbff7e6c459c93ce2c2f5a62226e + 521 + 1 + + + 1.0 + C3 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 04b3ae04020349a9bc568f26d17eab14 + 522 + 1 + + + 1.0 + C3 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + bbc4d367c60f41ad8a279c12e5cc7da6 + 523 + 1 + + + 1.0 + TS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 9a1aff251eda416ea6270e6158e663fc + 524 + 1 + + + 1.0 + TS is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 7a9e50846c274338ab09e7313b540edb + 525 + 1 + + + 1.0 + TS is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + b268cc3ef860434ba663dd46af633cc5 + 526 + 1 + + + 1.0 + SS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 1c9f67904a4c4fcc8cdac6a605900248 + 527 + 1 + + + 1.0 + The graph indexing process used 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + e00c403d1dc84ba6a37ee193596e320f + 528 + 1 + + + 1.0 + A graph was created for the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 3f2e726c3b624fe7bf11de9be2c0457e + 529 + 1 + + + 1.0 + Units are used to measure the context in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + f71dc0c394f04771af7e2ed37f85647e + 530 + 1 + + + 1.0 + Tokens are used to measure the word count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 2fea9c1856e54a91b79a9ce85755fbf5 + 531 + 1 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 7823b4c5b3364c5f890d05f33a46bdde + 532 + 1 + + + 1.0 + Intermediate-level summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 183f3a0b73ff41c5bb4a19fd7adf0c1d + 533 + 1 + + + 1.0 + The graph indexing process used 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + 392e06f17d724484a9cfb85fe69aac50 + 534 + 1 + + + 1.0 + A graph was created for the News dataset + 36db32c37e1987e2c5863898ad882190 + 6f49e00cdac04a358173ecd40351ee00 + 535 + 1 + + + 1.0 + Units are used to measure the context in the News dataset + 36db32c37e1987e2c5863898ad882190 + 3fef96af4ec343da8c34f8b09518de8a + 536 + 1 + + + 1.0 + Tokens are used to measure the word count in the News dataset + 36db32c37e1987e2c5863898ad882190 + bd403eff654e42c997e5656a2b1c1a20 + 537 + 1 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the News dataset + 36db32c37e1987e2c5863898ad882190 + 5763d829837144f199fac2b490b38110 + 538 + 1 + + + 1.0 + Datasets were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + 234c6f1859f0405ab607f0be53e7b06c + 539 + 1 + + + 1.0 + Wang et al., 2023a discusses the state-of-the-art results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + 21800eab85b94d4880bcada7a60763e5 + 540 + 1 + + + 1.0 + Zheng et al., 2024 discusses the competitive results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + b8bb28a7a9624b6d805be89adfe29eb5 + 541 + 1 + + + 1.0 + Zheng et al., 2024 discusses the LLM-as-a-judge method + 322e02986c8724eedbcf3ebfa20b989c + 61f26f8850504d56a6b7cd764c33299d + 542 + 1 + + + 1.0 + Embedding-based matching is used to match user queries + e4d9b12cf2b4c691c74019eefff4fb39 + d4456fac0ada4b6fbe3cfee873403d00 + 543 + 1 + + + 1.0 + Query-time LLM use was evaluated with different context window sizes + 4c855404ee3d3c94aa2136f1513c666f + f8fd3fcf650b47b2b1692506ebe77762 + 544 + 1 + + + 2.0 + The **CONTEXT WINDOW SIZE** and **FINAL EVALUATION** are closely related in the given data. A fixed context window size of 8k tokens was used for the final evaluation. This indicates that during the final evaluation phase, the system or model was configured to process and analyze text data within a predefined window of 8,000 tokens, ensuring consistency and standardization in the evaluation process. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + d95acc24180c47caa34114627d501592 + 545 + 1 + + + 1.0 + Natural Language Generation achieves state-of-the-art results + 322e02986c8724eedbcf3ebfa20b989c + f4753ab09adc42a9a52754e95440d4b9 + 546 + 1 + + + 1.0 + Natural Language Generation achieves competitive results + 322e02986c8724eedbcf3ebfa20b989c + 12f5a7c56b454a3d8aae97f65908f96b + 547 + 1 + + + 1.0 + Natural Language Generation is compared against human judgements + 322e02986c8724eedbcf3ebfa20b989c + 95f79ff0b8a34080ae2ac8448ce561f1 + 548 + 1 + + + 1.0 + Natural Language Generation can generate reference-based metrics + 322e02986c8724eedbcf3ebfa20b989c + 8733d4602c084e1cab1384dde0306abf + 549 + 1 + + + 1.0 + Natural Language Generation can measure qualities in a reference-free style + 322e02986c8724eedbcf3ebfa20b989c + ded3a49efdf6479a991cad53d0758cf4 + 550 + 1 + + + 1.0 + Es et al., 2023 discusses the RAGAS method + 322e02986c8724eedbcf3ebfa20b989c + 816fceb7e1ca4b5d9277368f78e6ed80 + 551 + 1 + + + 1.0 + RAGAS evaluates context relevance + 322e02986c8724eedbcf3ebfa20b989c + 50539d4503a4495097f49a8ed83e2462 + 552 + 1 + + + 1.0 + RAGAS evaluates faithfulness + 322e02986c8724eedbcf3ebfa20b989c + d6f67aa7ef0e4a19bf5830e777aafea5 + 553 + 1 + + + 1.0 + RAGAS evaluates answer relevance + 322e02986c8724eedbcf3ebfa20b989c + bbf61f9cd3e14f46a010d704e86be008 + 554 + 1 + + + 1.0 + The LLM evaluator assesses answers based on the directness metric + 322e02986c8724eedbcf3ebfa20b989c + 5d34e587bd2f41dba285e9178f179577 + 555 + 1 + + + 1.0 + Table 2 shows an example of LLM-generated assessment + 322e02986c8724eedbcf3ebfa20b989c + 901b491be7344401b4544ff05e591a0e + 556 + 1 + + + 1.0 + The LLM evaluator uses a head-to-head comparison approach + 322e02986c8724eedbcf3ebfa20b989c + ecacbf62b81d485396a56e1730e75a04 + 557 + 1 + + + 1.0 + The LLM evaluator assesses answers based on target metrics + 322e02986c8724eedbcf3ebfa20b989c + ba0ad1bcf02b4928a1b7ff7b23acdd6f + 558 + 1 + + + 1.0 + The LLM evaluator uses a control metric for validity + 322e02986c8724eedbcf3ebfa20b989c + 0e3c66c25d7e43a7960c37d28315e5d8 + 559 + 1 + + + 1.0 + The LLM evaluator accounts for stochasticity + 322e02986c8724eedbcf3ebfa20b989c + a0e0d5b7db9f4efcb5277856db799775 + 560 + 1 + + + 1.0 + The LLM evaluator uses mean scores from multiple comparisons + 322e02986c8724eedbcf3ebfa20b989c + 3f85dab93736440f9776020b6410aa9b + 561 + 1 + + + 1.0 + Directness is used to evaluate the straightforwardness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + 710ed70c346342ff81ccf205e30271bb + 562 + 1 + + + 1.0 + The question asks about public figures mentioned in entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + b3d3e8ba2ede4574a0498f082f0c15ae + 563 + 1 + + + 1.0 + Public figures are repeatedly mentioned across various entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + 8686013390614eca9116ccbab27431d7 + 564 + 1 + + + 1.0 + Answer 1 covers a wide range of public figures from different sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + fd8c8b7e3b9248abb1d8cb8958ab86d3 + 565 + 1 + + + 1.0 + Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports. + 718017a4871c909420f84b85b8ba969d + 039594428123415f95deb246f5097169 + 566 + 1 + + + 1.0 + Controversies involve public figures and impact public discourse. + 718017a4871c909420f84b85b8ba969d + d78ce7696ff14234a544de945ffe40d6 + 567 + 1 + + + 1.0 + Entertainment articles cover topics related to the entertainment industry + 322e02986c8724eedbcf3ebfa20b989c + 59b21508be904875af22b5c1cfdcd211 + 568 + 1 + + + 1.0 + Taylor Swift is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + e9c7a1d505b14229afbbef7c0d04751e + 569 + 1 + + + 1.0 + Travis Kelce is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 4b0efcd54efc40e8a884ac6c31deada2 + 570 + 1 + + + 1.0 + Britney Spears is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 0970f08f3d1a4d638d44e2ccb9237382 + 571 + 1 + + + 1.0 + Justin Timberlake is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 8f10c11ecb5142029869025521c73431 + 572 + 1 + + + 1.0 + Taylor Swift is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + e36a0e3901864a7eaa5f5ad4280a6471 + 573 + 1 + + + 1.0 + Travis Kelce is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6fce354faa104fe58ba8a565eb3c43f2 + 574 + 1 + + + 1.0 + Britney Spears is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 20585e9a43c04375aa334e946e2dd144 + 575 + 1 + + + 1.0 + Justin Timberlake is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 32e343c0ae454660bdfcd1d3133baf0a + 576 + 1 + + + 1.0 + Actors and Directors are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 505ab840f6cc4fa6a839ebfe82d255ed + 577 + 1 + + + 1.0 + Musicians and Executives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + e38eb1698900424bb7392a74ff0f3351 + 578 + 1 + + + 1.0 + Athletes and Coaches are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 855c57eecf2a45c7aab02ff1ac36938d + 579 + 1 + + + 1.0 + Influencers and Entrepreneurs are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6ee77949c94d4906bd98c24341fdfa03 + 580 + 1 + + + 1.0 + Public Figures in Controversy are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + d06f506604b249feb423915db282ed75 + 581 + 1 + + + 1.0 + Film is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 0f642f63d4af4fc38298822bfc952719 + 582 + 1 + + + 1.0 + Television is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + e017ad1f09b049a7ad41d5a11dc1e3d9 + 583 + 1 + + + 1.0 + Music is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 5cbced0ba7044b7490f520a436261c57 + 584 + 1 + + + 1.0 + Sports is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + d45dea925f8d4e7e93d0e17317001eec + 585 + 1 + + + 1.0 + Digital Media is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 8123eee04a3a4c779f03bdb85de99f9f + 586 + 1 + + + 1.0 + Cultural Narratives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6129d90c83194bcfaede9ff00a011297 + 587 + 1 + + + 1.0 + Trends are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6ef76e963a564dbe9c9feff4f8ce1683 + 588 + 1 + + + 1.0 + Social Discussions are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 1c8bad73fda646f8b3f413e432f0e351 + 589 + 1 + + + 1.0 + Public Discourse is a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 7e75749d13d24321b8b10c5be0138805 + 590 + 1 + + + 1.0 + Reference-based metrics require gold standard answers + 322e02986c8724eedbcf3ebfa20b989c + 05bfaf60aa304a288e6789443bd6fd6c + 591 + 1 + + + 1.0 + Gold standard answers are lacking for sensemaking questions + 322e02986c8724eedbcf3ebfa20b989c + 6097e047a74d41ca996a0b7949ef6f0e + 592 + 1 + + + 3.0 + End users play a crucial role in the validation process of sensemaking questions and target metrics. Sensemaking questions are specifically validated with end users to ensure their relevance and accuracy. This collaborative approach ensures that the questions and metrics are aligned with the needs and expectations of the end users, thereby enhancing the overall effectiveness and applicability of the sensemaking process. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + e257439ce5be47a88faaeb0fe01bc4a1 + 593 + 1 + + + 1.0 + Target metrics are validated with end users + e4d9b12cf2b4c691c74019eefff4fb39 + 067b9486d59f45d2963235220f723a41 + 594 + 1 + + + 1.0 + The control metric is used as an indicator of validity + 322e02986c8724eedbcf3ebfa20b989c + 87c46c7ead5447bc8309ab116a316959 + 595 + 1 + + + 1.0 + Taylor Swift is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + f607d795f00347109cab3b2370c414f7 + 596 + 1 + + + 1.0 + Taylor Swift is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + 8f0610c89e9f42e9b8c3d8a947fa2852 + 597 + 1 + + + 1.0 + Travis Kelce is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + 75ef3591790a49748154ddbba20e9cdf + 598 + 1 + + + 1.0 + Travis Kelce is a public figure in the sports sector. + 718017a4871c909420f84b85b8ba969d + 58b7f26cb17b4b2283d3cacbaed15cfc + 599 + 1 + + + 1.0 + Britney Spears is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + 277cdf13617e47ca883b949f495bc243 + 600 + 1 + + + 1.0 + Britney Spears is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + 26080c121c9645b2bb258e4d61d47672 + 601 + 1 + + + 1.0 + Justin Timberlake is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + ee91a06f13b4495f95c800a0c7329ef7 + 602 + 1 + + + 1.0 + Justin Timberlake is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + 6ed8b67be79242e98aa1b9283431d5df + 603 + 1 + + + 1.0 + Answer 1 includes public figures from the film sector. + 718017a4871c909420f84b85b8ba969d + 40c2425cb1c34c1591f7cb89f9f5e0bf + 604 + 1 + + + 1.0 + Answer 1 includes public figures from the television sector. + 718017a4871c909420f84b85b8ba969d + 7cf59650687a435ba26a7c5ffc6c4f4c + 605 + 1 + + + 1.0 + Answer 1 includes public figures from the music sector. + 718017a4871c909420f84b85b8ba969d + 53c2882604b74192a649a4eaa0536c5e + 606 + 1 + + + 1.0 + Answer 2 focuses on public figures primarily from the music sector. + 718017a4871c909420f84b85b8ba969d + 3fbb8aeacea54ca9a957118fba613ccf + 607 + 1 + + + 1.0 + Answer 1 includes public figures from the sports sector. + 718017a4871c909420f84b85b8ba969d + 496ae6a894584a6cb12e50b516341788 + 608 + 1 + + + 1.0 + Answer 2 focuses on public figures primarily from the sports sector. + 718017a4871c909420f84b85b8ba969d + dd1a82c597794ba3a490cb70d488d9dd + 609 + 1 + + + 1.0 + Answer 1 includes public figures from the digital media sector. + 718017a4871c909420f84b85b8ba969d + bbd206ae4c1a4794813fd239fcfef313 + 610 + 1 + + + 1.0 + Answer 1 cites specific data sources from the News article dataset for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + 708ac105e8bb4beeade0472c899f214d + 611 + 1 + + + 1.0 + Answer 1 provides insights into controversies involving public figures and their impact on public discourse. + 718017a4871c909420f84b85b8ba969d + b4fe3c6aea95472db73a5e8bf575895a + 612 + 1 + + + 1.0 + Answer 1 includes public figures from the gaming sector. + 718017a4871c909420f84b85b8ba969d + a861f44aa7dd414790ee82b3f651c609 + 613 + 1 + + + 1.0 + Answer 1 cites specific data sources for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + add9948a221a4aabafbaaed650b1db26 + 614 + 1 + + + 1.0 + Answer 2 was generated using the Naïve RAG method, which directly lists specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d + 6c04f2ab7c9843ea900c3444b014bed8 + 615 + 1 + + + 2.0 + ANSWER 2 is a generated answer for a question in the NEWS ARTICLE DATASET. It relies heavily on a single source from the NEWS ARTICLE DATASET for data. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + 88356435ca9d43ebaf93134b3af8a53e + 616 + 1 + + + 1.0 + Answer 2 relies heavily on a single data source. + 718017a4871c909420f84b85b8ba969d + 233edf428a04436a8d32849af584f9d8 + 617 + 1 + + + 1.0 + Naïve RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + 28a6317db3d141db82a4a22525265fef + 618 + 1 + + + 1.0 + The global approach to Graph RAG shows improvements over naïve RAG + e4d9b12cf2b4c691c74019eefff4fb39 + 90051a1b69cd40f696e440d54085887e + 619 + 1 + + + 1.0 + LLM-generated assessments are used to evaluate the answers produced for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + 973fe6e8a1314a269748f40a98786115 + 620 + 1 + + + 1.0 + Example question is part of the News article dataset used for analysis + ebf5249c888e07fedce6572a4c03f88c + 9a442a8c054d48339aff04923bafe47f + 621 + 1 + + + 1.0 + Head-to-head win rate percentages were used to compare different conditions + 4c855404ee3d3c94aa2136f1513c666f + ffdacb33c3a94b7f9d890d7cc03a1f40 + 622 + 1 + + + 1.0 + Win rate percentages were used to measure the performance of different conditions + 4c855404ee3d3c94aa2136f1513c666f + 8792fc245cc94235a7764481ebad4828 + 623 + 1 + + + 1.0 + The overall winner per dataset and metric was determined for each condition + 4c855404ee3d3c94aa2136f1513c666f + b5982d09c32e4e7387e88f9160b4dd78 + 624 + 1 + + + 1.0 + Self-win rates were shown as the expected 50% for each condition + 4c855404ee3d3c94aa2136f1513c666f + 04ed223f57e44cf18284ba42ba760423 + 625 + 1 + + + 1.0 + The indexing process resulted in the creation of graphs + 36db32c37e1987e2c5863898ad882190 + 0debfb49a28d480db1b7d5ef713cac8f + 626 + 1 + + + 1.0 + Map-reduce summarization requires the highest number of context tokens + 36db32c37e1987e2c5863898ad882190 + 1f9abc7d006f4afa86200385acc3d1ae + 627 + 1 + + + 1.0 + Root-level community summaries require dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + cac3f76fbc11413e92cdfd3064d56ece + 628 + 1 + + + 2.0 + Queries are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + f120d98b793a4276a6f1a0a8e51a589a + 629 + 1 + + + 2.0 + Self-memory is related to generation-augmented retrieval + f35de4d9fb65f1d5a392064b20545c19 + bfda4c94278b49ab98cd3f407980d4d8 + 630 + 1 + + + 2.0 + CAiRE-COVID is a system for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + 7e5e5b80b84749d98cb36f56dbfcb47b + 631 + 1 + + + 2.0 + ITRG is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + 6330604339ca4113b94624bc9bed5ede + 632 + 1 + + + 2.0 + IR-CoT is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + 9638492c4f034be6b3bf88f8abd82edc + 633 + 1 + + + 2.0 + DSP is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + a76322b81f644f3c8733d04fa046b4e4 + 634 + 1 + + + 2.0 + RAPTOR is a method for generating a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + 653ee6584dbc46d1b8e97a05a3eac81e + 635 + 1 + + + 2.0 + The paper by Baek et al. discusses the KAPING method + 92e93fc6449756c0a60200636b297f65 + 9f0d58a479ec404d8e8f493f9269b08d + 636 + 1 + + + 2.0 + The paper by He et al. discusses the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + c5ae09d00a3f417981fc4177ef333eff + 637 + 1 + + + 2.0 + The paper by Zhang discusses the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + 4dd086fcba704d26b976c08a81c1465c + 638 + 1 + + + 2.0 + The paper by Kang et al. discusses the SURGE method + 92e93fc6449756c0a60200636b297f65 + f1ef6375ea84496eaed13c03318d80c6 + 639 + 1 + + + 2.0 + The paper by Ranade and Joshi discusses the FABULA method + 92e93fc6449756c0a60200636b297f65 + ba6829116d114532b99530f101ff0c72 + 640 + 1 + + + 2.0 + Both LangChain and LlamaIndex support a variety of graph databases + 92e93fc6449756c0a60200636b297f65 + 1ab2048463174873883061373d480ac4 + 641 + 1 + + + 2.0 + LangChain supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + 97038fe907af4710859c3daeb13972e9 + 642 + 1 + + + 2.0 + LangChain supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + e0595082eb9f41a4ac2afd9e614b363c + 643 + 1 + + + 1.0 + LangChain developed Langchain graphs + 71f6daf11e64e5273a3847d46bf228e1 + 5bd2ef268d4f4ba18925c17242370e21 + 644 + 1 + + + 2.0 + LlamaIndex supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + 96a21e764d1143fc90de0b2cc7751983 + 645 + 1 + + + 2.0 + LlamaIndex supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + 801b7142ab5b4c5eac41dade999a7c1f + 646 + 1 + + + 2.0 + NaLLM is a method that can create and reason over knowledge graphs in Neo4J format + 92e93fc6449756c0a60200636b297f65 + aac39de4e7e74d1c83f0eb835e635c88 + 647 + 1 + + + 2.0 + Neo4J developed Project NaLLM + 833e7d67dcd30790b26b71c9b5306f6b + c2e801c8221c4806a4f59ba5b793c784 + 648 + 1 + + + 2.0 + GraphRAG is a method that can create and reason over knowledge graphs in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + 351fc21986564103b324540289e2e608 + 649 + 1 + + + 2.0 + The paper by Manakul et al. discusses the SelfCheckGPT method + 92e93fc6449756c0a60200636b297f65 + 1c8a90b0aed7439286bbf85903d423d4 + 650 + 1 + + + 1.0 + SelfCheckGPT is an approach mentioned in the work by Manakul et al., 2023 + e4d9b12cf2b4c691c74019eefff4fb39 + 6c98609312154f118c04d8781663b16a + 651 + 1 + + + 1.0 + SelfCheckGPT is used to compare fabrication rates + e4d9b12cf2b4c691c74019eefff4fb39 + b91a6bf16e334b3ab7ec57665e980ceb + 652 + 1 + + + 1.0 + Embedding-based matching is used to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + c067e41adf9840df89052b111e6c0a6a + 653 + 1 + + + 1.0 + Hybrid RAG schemes combine embedding-based matching against community reports + e4d9b12cf2b4c691c74019eefff4fb39 + 76d7feb8140b4064b5492d3055736ee0 + 654 + 1 + + + 1.0 + The roll-up operation can be extended using map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + 53273797509d45178c49045830ec9fc2 + 655 + 1 + + + 1.0 + The drill down mechanism follows the information scent in the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + 2383fd3c3b4a4249a5a96550c494edb2 + 656 + 1 + + + 1.0 + The global approach to Graph RAG achieves competitive performance at a fraction of the token cost + e4d9b12cf2b4c691c74019eefff4fb39 + 29f172df150042e0a6db5481d5d91cfc + 657 + 1 + + + 1.0 + The open-source implementation of Graph RAG approaches is Python-based + e4d9b12cf2b4c691c74019eefff4fb39 + a243935f440241a281fbabb20422c641 + 658 + 1 + + + 1.0 + The drill down mechanism follows the information scent + e4d9b12cf2b4c691c74019eefff4fb39 + 34b704124fe94c2f933a344c11165f2e + 659 + 1 + + + 1.0 + Alonso Guevara Fernández and Amber Hoak both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + e429a497b01c40f3aef7e2205eaf01d8 + 660 + 1 + + + 1.0 + Alonso Guevara Fernández and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + f521933b91564693b07bd838160083ac + 661 + 1 + + + 1.0 + Alonso Guevara Fernández and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + e0d361b6991b40debf5599e86f2638ca + 662 + 1 + + + 1.0 + Alonso Guevara Fernández and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + cf16005cfadf4e48832ffd0e43f57be1 + 663 + 1 + + + 1.0 + Alonso Guevara Fernández and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + db10b0e690754748b0d75639f3e8d2b8 + 664 + 1 + + + 1.0 + Alonso Guevara Fernández and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + ec13b80209e246588bb5486d516f85eb + 665 + 1 + + + 1.0 + Alonso Guevara Fernández and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 617a76d54ed546e29428a31dea955b96 + 666 + 1 + + + 1.0 + Alonso Guevara Fernández and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + c88ffd2aa7284ac38eb4351c5fad6f44 + 667 + 1 + + + 1.0 + Alonso Guevara Fernández and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0dace3b24df14aae909a2815653e9db1 + 668 + 1 + + + 1.0 + Alonso Guevara Fernández and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + b3cfde857302479aa59b91d6648a40df + 669 + 1 + + + 1.0 + Alonso Guevara Fernández and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 01a52a986b6a444badc83fb11aa7a160 + 670 + 1 + + + 1.0 + Alonso Guevara Fernández and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2516b3485b754bdabf6820863c918e3d + 671 + 1 + + + 1.0 + Alonso Guevara Fernández and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0523ed6330f4429f8468f5b49169c940 + 672 + 1 + + + 1.0 + Alonso Guevara Fernández and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0b7ac591dfd34971b24d38e344b40c37 + 673 + 1 + + + 1.0 + Alonso Guevara Fernández and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2864f68297e94d7e84213833e22da077 + 674 + 1 + + + 1.0 + Alonso Guevara Fernández and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 629cd969d05c4c329bbe24f5d86e0089 + 675 + 1 + + + 1.0 + Alonso Guevara Fernández and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + f7bc5d1fb1404acdb77d50a6b9129141 + 676 + 1 + + + 1.0 + Alonso Guevara Fernández and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + d47535a52c2b40a3bacb3d520b8f0f1c + 677 + 1 + + + 1.0 + Alonso Guevara Fernández and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 557ed8720c2845cabcce0287f7284b3e + 678 + 1 + + + 1.0 + Alonso Guevara Fernández and Sarah Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + cef8ff96a0e842fdae4751933bcb1a28 + 679 + 1 + + + 1.0 + Alonso Guevara Fernández and Shane Solomon both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 93c8356953da486e9630d7e7304a6ff3 + 680 + 1 + + + 1.0 + Amber Hoak and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + dc52f3641c1548bba5b3cf8c65a5c072 + 681 + 1 + + + 1.0 + Amber Hoak and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + d0fdcc6945d84b20aa1de4afe2786592 + 682 + 1 + + + 1.0 + Amber Hoak and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + af6e03260c5946be96737b148b5edd9d + 683 + 1 + + + 1.0 + Amber Hoak and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1949eb874c544c58a71bbd04d6241a22 + 684 + 1 + + + 1.0 + Amber Hoak and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 09384ed4453846cb8c4d0076ecbf928a + 685 + 1 + + + 1.0 + Amber Hoak and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 181833ae880a4d0ab24ba0ccb158138d + 686 + 1 + + + 1.0 + Amber Hoak and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 3426a7ea02f740aeabcb552feee11bcc + 687 + 1 + + + 1.0 + Amber Hoak and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + e47ae0eea85b4f6e86b77fe56396460e + 688 + 1 + + + 1.0 + Amber Hoak and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1ad7e0ad19334488b5d3b008f93a4ef4 + 689 + 1 + + + 1.0 + Amber Hoak and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 8c76a8cb5951422ba3b3cc6fcb66a391 + 690 + 1 + + + 1.0 + Amber Hoak and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 096dee591c1f4141a73fd628a59ffbe9 + 691 + 1 + + + 1.0 + Amber Hoak and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 9886a385806e46a69d92a726017b99b6 + 692 + 1 + + + 1.0 + Amber Hoak and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 4ecf891e9a5b4daf9e02d5b2ec963079 + 693 + 1 + + + 1.0 + Amber Hoak and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 59207227178147e39296a4059ac1055d + 694 + 1 + + + 1.0 + Amber Hoak and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0682e47ed49146c0bc5e2b77fb924b6c + 695 + 1 + + + 1.0 + Amber Hoak and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 6022e4aa784f42b88dbcb27a5d9d2614 + 696 + 1 + + + 1.0 + Amber Hoak and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 386047cff59549ea83158b69bbac1870 + 697 + 1 + + + 1.0 + Amber Hoak and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2ead31e49ce643ebae4d5f047bb7a37b + 698 + 1 + + + 1.0 + J. Achiam and S. Adler co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + a74ee5b02e1e41b0ac4cf5449f7cdf2c + 699 + 1 + + + 1.0 + J. Achiam and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0f73fcdab31348a880a468124099071c + 700 + 1 + + + 1.0 + J. Achiam and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 24d2dcb7f28144cbad714b0a8b6c9e70 + 701 + 1 + + + 1.0 + J. Achiam and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 093447e0342e490aa6a55bd70ce7c2f2 + 702 + 1 + + + 1.0 + J. Achiam and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0439db7ac7d2484596e02246bd340424 + 703 + 1 + + + 1.0 + J. Achiam and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 81ffef86ebb341bebf145c742fb33dbd + 704 + 1 + + + 1.0 + J. Achiam and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + a0da2872126f43769f75c8533fca5e26 + 705 + 1 + + + 1.0 + J. Achiam and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 7facdc1f91014f42a67e34bac31a95ce + 706 + 1 + + + 1.0 + J. Achiam and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + ca7a635373294067b5f3050c82d38983 + 707 + 1 + + + 1.0 + S. Adler and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 2833c46f05984f729c7ec15e071f0c8e + 708 + 1 + + + 1.0 + S. Adler and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + ada7cf1171b74ad793f7856febc9c6fe + 709 + 1 + + + 1.0 + S. Adler and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 33b355c0a8044ef2b2b8be81bea0d431 + 710 + 1 + + + 1.0 + S. Adler and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 85698526e09a47878e3255a251d95406 + 711 + 1 + + + 1.0 + S. Adler and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 1a625c51e7ad497b86041757d1cde642 + 712 + 1 + + + 1.0 + S. Adler and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + df91c0e5657a4bafa849c8a3079ca582 + 713 + 1 + + + 1.0 + S. Adler and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 091a9788b29443509feda24aa5f5c241 + 714 + 1 + + + 1.0 + S. Adler and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 6f750deaedcb4612b419c3d8dd7e5cb2 + 715 + 1 + + + 1.0 + S. Agarwal and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + d5ea3f061e9c419fb1c07b680bfb287a + 716 + 1 + + + 1.0 + S. Agarwal and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 1f4fd98283df43c69d5537c002b98f58 + 717 + 1 + + + 1.0 + S. Agarwal and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + f3bb299bf6454785a8a406dce9776789 + 718 + 1 + + + 1.0 + S. Agarwal and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 218d3d5a4a544df99caed612e48add5b + 719 + 1 + + + 1.0 + S. Agarwal and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 02393af06d3649549b3e9290b4e46c0a + 720 + 1 + + + 1.0 + S. Agarwal and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + ffa9208f15744978a4ea45c1cff18a86 + 721 + 1 + + + 1.0 + S. Agarwal and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 23ea2a1d78984eb38721adeadee662e1 + 722 + 1 + + + 1.0 + L. Ahmad and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0d0a729e30634e1fb198609ce10c69bf + 723 + 1 + + + 1.0 + L. Ahmad and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 896e7d03cad7450e8044fcb0fd9f6e92 + 724 + 1 + + + 1.0 + L. Ahmad and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + a777a0c3a34b4990899f2e1e1f1c2074 + 725 + 1 + + + 1.0 + L. Ahmad and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 6cd46299702049bcbd39407fa97f0dc0 + 726 + 1 + + + 1.0 + L. Ahmad and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 7af01185ebd648e49bf9a57481e0dc7c + 727 + 1 + + + 1.0 + L. Ahmad and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + d6aad3495b4d4c7ab2a03c44600584ba + 728 + 1 + + + 1.0 + I. Akkaya and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 482d5ecf4ce949e9a5d81f1b368769ee + 729 + 1 + + + 1.0 + I. Akkaya and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 64f7a02f89bd4a37844c482f00d00643 + 730 + 1 + + + 1.0 + I. Akkaya and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 26609be86d614d85ae97deeae4a4be1e + 731 + 1 + + + 1.0 + I. Akkaya and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 5dffe69693734eaeb360de4582d489b0 + 732 + 1 + + + 1.0 + I. Akkaya and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + e4fe96090a7641c68d0b1995d1f238b4 + 733 + 1 + + + 1.0 + F. L. Aleman and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 3fa1055cd26840678d546570e8b423d9 + 734 + 1 + + + 1.0 + F. L. Aleman and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9802dae4757b42269e93c66b5214a396 + 735 + 1 + + + 1.0 + F. L. Aleman and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + c39e66acec274a5980ce275709a847ba + 736 + 1 + + + 1.0 + F. L. Aleman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 667794d1397a40bb904d406205960864 + 737 + 1 + + + 1.0 + D. Almeida and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 891f50162c0140e4b9c0e4ba33f69a1b + 738 + 1 + + + 1.0 + D. Almeida and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 922773e5841a475d89d5904fe7a324f8 + 739 + 1 + + + 1.0 + D. Almeida and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0338bebae19c41c196ee6c09ccba36e3 + 740 + 1 + + + 1.0 + J. Altenschmidt and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0b407647077c4288b2324f06ac355985 + 741 + 1 + + + 1.0 + J. Altenschmidt and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 4e9254fd4b234106843cf8ff91fd3b6f + 742 + 1 + + + 1.0 + S. Altman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + e6a7f4ccd6f54136b784572db0d5cb88 + 743 + 1 + + + 1.0 + R. Anil and S. Borgeaud co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + ccf54360ef954353b71c1c8175cd7f4e + 744 + 1 + + + 1.0 + R. Anil and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + bd1c72f46b81427892b1f415fecce77e + 745 + 1 + + + 1.0 + R. Anil and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 148e7caecdf740e58ee09a9ff549d19c + 746 + 1 + + + 1.0 + R. Anil and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + df78f3e3415a4d47b6dffdd3890f3eee + 747 + 1 + + + 1.0 + R. Anil and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 003abb3c5743482aa63022cf20cf5ccc + 748 + 1 + + + 1.0 + R. Anil and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 4788433078b843079ccd9a64e5430169 + 749 + 1 + + + 1.0 + R. Anil and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + cbecfdbc04c9405aa139566d727d3a33 + 750 + 1 + + + 1.0 + R. Anil and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + f514a867efb948868009b435fecbe372 + 751 + 1 + + + 1.0 + S. Borgeaud and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 6e4c8a7f1da147f5b38103c51c999502 + 752 + 1 + + + 1.0 + S. Borgeaud and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + ac0e52b6b3ae4cc485f9eef2f2dea7e7 + 753 + 1 + + + 1.0 + S. Borgeaud and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + b150af2f3df24f17a7fd836ba663680a + 754 + 1 + + + 1.0 + S. Borgeaud and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + f9740be4adc946149b5941f355d45c74 + 755 + 1 + + + 1.0 + S. Borgeaud and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + c6afe51b28f94c3ba21640387edd2ee8 + 756 + 1 + + + 1.0 + S. Borgeaud and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 040b86f496dc4930a895f2c21cb0731c + 757 + 1 + + + 1.0 + S. Borgeaud and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + e13ed464574f483c9f1db5f569e91445 + 758 + 1 + + + 1.0 + Y. Wu and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + e1131985dc53451fa7543912b2e7db07 + 759 + 1 + + + 1.0 + Y. Wu and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + d6c00e0a975e4adc979afd25d4037d4d + 760 + 1 + + + 1.0 + Y. Wu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 9cac1113be2148ce8abaa957620f9d59 + 761 + 1 + + + 1.0 + Y. Wu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + c277297e3e7b417892e986c8767f58ad + 762 + 1 + + + 1.0 + Y. Wu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 744f492f46d841c0b0fee5f4a9b40b6c + 763 + 1 + + + 1.0 + Y. Wu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + d59386dbfa0349b49f7b904e288b21ad + 764 + 1 + + + 1.0 + J.-B. Alayrac and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 1005971b7e764bffa0a4610ad403976b + 765 + 1 + + + 1.0 + J.-B. Alayrac and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 1eaf3527e2804c75bbd9e3ccac9d760e + 766 + 1 + + + 1.0 + J.-B. Alayrac and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 301ab7b5f81d40788e46dacb09579b50 + 767 + 1 + + + 1.0 + J.-B. Alayrac and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + d6f25f03a08b41b4a2eaa9df3db9dceb + 768 + 1 + + + 1.0 + J.-B. Alayrac and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 6529cf63872440a98aeab73beee3762a + 769 + 1 + + + 1.0 + J. Yu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 65faae6c13f5444b8d71b4b2be38eba3 + 770 + 1 + + + 1.0 + J. Yu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 9187af05545a4c8d92e38c2b46254092 + 771 + 1 + + + 1.0 + J. Yu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + f12578d1ff7b46f5ae84c7672fac8deb + 772 + 1 + + + 1.0 + J. Yu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 6b5c42bba0ec48c1a5de177a7f1b9bfc + 773 + 1 + + + 1.0 + R. Soricut and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 444ab529d10c47f19ef33e931489b8b8 + 774 + 1 + + + 1.0 + R. Soricut and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 0ffea4c3c86849ab828036b67b58acdc + 775 + 1 + + + 1.0 + R. Soricut and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 65b488142164407a81b496b4820ef556 + 776 + 1 + + + 1.0 + J. Schalkwyk and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + b3b006f8629b44df81a266c1e4d81d3f + 777 + 1 + + + 1.0 + J. Schalkwyk and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 2a48f66b6a424b9ebf38562836fe1c82 + 778 + 1 + + + 1.0 + A. M. Dai and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 54e486668fa94feda285f377fb05d14d + 779 + 1 + + + 1.0 + J. Baek and A. F. Aji co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + d7c4899260084560905ac54dba81f0e6 + 780 + 1 + + + 1.0 + J. Baek and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 880db899ce864932843fe230e3d364ad + 781 + 1 + + + 1.0 + A. F. Aji and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 59463c48d2fb48308cd38ee8dd869f59 + 782 + 1 + + + 1.0 + T. Ban and L. Chen co-authored the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + ff69f1aae7404c38b8bde6abc5a79b57 + 783 + 1 + + + 1.0 + Baumel, T. and Eyal, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 8db22f709edb4ae98f0fef060ccd24b8 + 784 + 1 + + + 1.0 + Baumel, T. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 0e412e834e62475a9fe1920438f7b75b + 785 + 1 + + + 1.0 + Baumel, T. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + 2e3f6dbac98742ddb213037ae77f0a82 + 786 + 1 + + + 1.0 + Eyal, M. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 59ced15f5a1d485ebf0eac7fa85c1cdf + 787 + 1 + + + 1.0 + Eyal, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + 7cef69e2a71c4379b0816844799fc71e + 788 + 1 + + + 1.0 + Elhadad, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + 0f565c275f8148d885ae53c315ddc568 + 789 + 1 + + + 1.0 + Blondel, V. D. and Guillaume, J.-L. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 29c05af68cc541b79fdf499eac42b9c6 + 790 + 1 + + + 1.0 + Blondel, V. D. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 8d5d7b8fb7d14d0ba46ce7f0be6de661 + 791 + 1 + + + 1.0 + Blondel, V. D. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 607a66de21cf42e497c23013327b751f + 792 + 1 + + + 1.0 + Guillaume, J.-L. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + f4f85b6086384211a25248f614bfb786 + 793 + 1 + + + 1.0 + Guillaume, J.-L. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + c1a4bcd4e7874e699f06bc795e291150 + 794 + 1 + + + 1.0 + Lambiotte, R. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + ca585d891433495aa70a3a01b252e50c + 795 + 1 + + + 1.0 + Brown, T. and Mann, B. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + f26b5e1c52e445998b6a63738d203b38 + 796 + 1 + + + 1.0 + Brown, T. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 0945030309e14518a16df16fbb25c76f + 797 + 1 + + + 1.0 + Brown, T. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 35939cc34a734b5f867f8d75df419f37 + 798 + 1 + + + 1.0 + Brown, T. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 807561f61906451b880e04ac6a33687f + 799 + 1 + + + 1.0 + Brown, T. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 0538e2dc04174140a43bc0359fed2d23 + 800 + 1 + + + 1.0 + Brown, T. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 6e87d210775b45e4a09e518492329bce + 801 + 1 + + + 1.0 + Brown, T. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + bde6223f81884473a1acc3b75dd056aa + 802 + 1 + + + 1.0 + Brown, T. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 7beb3a2ecfd5419b950a20a155e06169 + 803 + 1 + + + 1.0 + Brown, T. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 961dda09b0df497a974c38c28eb90686 + 804 + 1 + + + 1.0 + Mann, B. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 1fb946075dd54b218b8dfad20647d33e + 805 + 1 + + + 1.0 + Mann, B. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + bf90efb1858e49b19987cbd280d0e911 + 806 + 1 + + + 1.0 + Mann, B. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + f9949e43ea004014abec1b59f2155b5a + 807 + 1 + + + 1.0 + Mann, B. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 3d89c7fe0b6448e0a0d27bceccc09f09 + 808 + 1 + + + 1.0 + Mann, B. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 185f98d04d9f484ab3d626fd459a23a2 + 809 + 1 + + + 1.0 + Mann, B. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 688e9b443bc44782855aea4afd8a9d16 + 810 + 1 + + + 1.0 + Mann, B. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + cd56c4963a0e49d7bab0e25f0e068779 + 811 + 1 + + + 1.0 + Mann, B. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + c4d4216677af42f5a29a0f4dcb442220 + 812 + 1 + + + 1.0 + Ryder, N. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + d99684f88f2d43eaacd62ba9082b64a5 + 813 + 1 + + + 1.0 + Ryder, N. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9ca07c62b7e146298882e33f3c6cb653 + 814 + 1 + + + 1.0 + Ryder, N. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + eef3aeb29aba43da93b433a816e77203 + 815 + 1 + + + 1.0 + Ryder, N. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 1c2a05515b9f49e1966a4ceb4bb0a3a5 + 816 + 1 + + + 1.0 + Ryder, N. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 86db00646d264b0a922c6b639dc9d16b + 817 + 1 + + + 1.0 + Ryder, N. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 7892ab98e1b0475c97a798aa8b2d7f6c + 818 + 1 + + + 1.0 + Ryder, N. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9732636cdd50433bb146a241cd72dbc5 + 819 + 1 + + + 1.0 + Subbiah, M. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + a5a8a63d5baf4946b7d7d1696f0e4e0e + 820 + 1 + + + 1.0 + Subbiah, M. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + a69a82dd8773426096c58ddc56832770 + 821 + 1 + + + 1.0 + Subbiah, M. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 72a7215e3e4a4b0db851351dfe5afd37 + 822 + 1 + + + 1.0 + Subbiah, M. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 5f8224530d424618acb32b74a3afe2c9 + 823 + 1 + + + 1.0 + Subbiah, M. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 99d4510732d843299514461aebd5f176 + 824 + 1 + + + 1.0 + Zhao, D. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 1730cbfab07747508d5b5ea421b97953 + 825 + 1 + + + 1.0 + Es, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + 7dd3aadc8f424988a72f8ba3ccf17155 + 826 + 1 + + + 1.0 + James, J. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + a2e7cbaf38c24564b2abe61680cacd72 + 827 + 1 + + + 1.0 + Espinosa-Anke, L. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + 162f1ccf8dfc46cea4d54a36ed9ec823 + 828 + 1 + + + 1.0 + Schockaert, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + 17127080df794121830177e93631aa3b + 829 + 1 + + + 1.0 + Feng, Z. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 374a8f81e6304b6d90b44cdceb90ecb4 + 830 + 1 + + + 1.0 + Feng, X. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 8c6aafd2a5da496385bea2c69be03a5a + 831 + 1 + + + 1.0 + Yang, M. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 06b9da0d4d9b4d6bb762bd2eeca7028a + 832 + 1 + + + 1.0 + Qin, B. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + a5f6de474fb24ec9af7403231c616831 + 833 + 1 + + + 1.0 + Gao, Y. and Xiong, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + a0ef87eb823b400594300f5c47e5c9c3 + 834 + 1 + + + 1.0 + Gao, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + ed076834490640acbb5d837aaac9fed5 + 835 + 1 + + + 1.0 + Gao, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + f935684600f34a27906def1902627ff2 + 836 + 1 + + + 1.0 + Gao, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 72fc2f604fc644e39f7d70e25094e347 + 837 + 1 + + + 1.0 + Gao, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 3ddb886b220c4bb2ab3d68f7f29ce5c5 + 838 + 1 + + + 1.0 + Gao, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 2f482bb08d564072a5ff4f2509dfdda6 + 839 + 1 + + + 1.0 + Gao, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + fc86507186da4c6c94fe3b788d77c471 + 840 + 1 + + + 1.0 + Gao, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + c235c2c649484c83967e2a42523028bb + 841 + 1 + + + 1.0 + Xiong, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 7e9d748907ea4b74925a32999a2b40d9 + 842 + 1 + + + 1.0 + Xiong, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + eefaef8a5c7149d18d304f39bf41f280 + 843 + 1 + + + 1.0 + Xiong, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 327114716cea49a79d33ba609158cd87 + 844 + 1 + + + 1.0 + Xiong, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + b45920e405af45f787ab167f54cfd2e9 + 845 + 1 + + + 1.0 + Xiong, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 00db8f6e99254c99be6c6f5c14a79500 + 846 + 1 + + + 1.0 + Xiong, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + aadd82f0e70c4fc49b1bdee3f60c1890 + 847 + 1 + + + 1.0 + Xiong, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 9528d92ccc10454793c4df59e24586db + 848 + 1 + + + 1.0 + Gao, X. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 80618f4e809e4af1bcdb59342c375377 + 849 + 1 + + + 1.0 + Gao, X. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 32d785e275be458fb7178ad2021ecdfc + 850 + 1 + + + 1.0 + Gao, X. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 0757f97d1fbf49748169ba696a364e4c + 851 + 1 + + + 1.0 + Gao, X. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + ca9a355bf38b452cbde62dba747ec65f + 852 + 1 + + + 1.0 + Gao, X. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + ba297c67512447e4b86f0cbc39fbc301 + 853 + 1 + + + 1.0 + Gao, X. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 00a9c8745b404b659c76a694dba9851c + 854 + 1 + + + 1.0 + Jia, K. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + eb338f2214414f0f9fa396f06ca12860 + 855 + 1 + + + 1.0 + Jia, K. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + d119680bac3744e58d2ed3273b1208b6 + 856 + 1 + + + 1.0 + Jia, K. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + d60eefaddf1e4b1db125d8f9ac49bacb + 857 + 1 + + + 1.0 + Jia, K. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 6058343c0824402e9843c92b2991f778 + 858 + 1 + + + 1.0 + Jia, K. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 383a003edc5a4f2387c7dd7865a984c9 + 859 + 1 + + + 1.0 + Pan, J. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 6bf9623c44824e48b7451bdfa1b47816 + 860 + 1 + + + 1.0 + Pan, J. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + cc62f333666e427eb1c66ec3f12a7a55 + 861 + 1 + + + 1.0 + Pan, J. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + c6d99481f5f545278ca8a73650b66e87 + 862 + 1 + + + 1.0 + Pan, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 5a0887b99d8b4bd89286962cd6f07037 + 863 + 1 + + + 1.0 + Bi, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 75174e7af26f434c9154b182087b58dc + 864 + 1 + + + 1.0 + Bi, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 0c43dd117fe6495991d4b4d8c2f5d70e + 865 + 1 + + + 1.0 + Bi, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 65e5d27c4f8a4dfa8ad92f227964b9cf + 866 + 1 + + + 1.0 + Dai, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 63c4595187884af29aa46d03319acded + 867 + 1 + + + 1.0 + Dai, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + b32482039edd4d50bc43514570500345 + 868 + 1 + + + 1.0 + Sun, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + f3e6e2c82bab4430a33987a19e3d1835 + 869 + 1 + + + 1.0 + Goodwin, T. R. and Savery, M. E. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 6768cc24da5d4a2492ff936dd4b35661 + 870 + 1 + + + 1.0 + Goodwin, T. R. and Demner-Fushman, D. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 223e1e3e7c4f4282b086e940f8c935c2 + 871 + 1 + + + 2.0 + Khattab, O. and Santhanam, K. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + f5bb365c9a814b909df0351498d79bb5 + 872 + 1 + + + 2.0 + Khattab, O. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and highlights their collaborative work in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + ab4ad26863b44497a1e48aa7c17a096c + 873 + 1 + + + 2.0 + Khattab, O. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 145bc384290c49228a231ac124ce88a8 + 874 + 1 + + + 2.0 + Khattab, O. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + f64c99537adf489ea58940e417cb5924 + 875 + 1 + + + 2.0 + Khattab, O. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + fe98ea566cf6486b85f8ed14aabb2618 + 876 + 1 + + + 2.0 + Khattab, O. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 6bfb83cb716745fcb591c8d2fb54f8f4 + 877 + 1 + + + 1.0 + Khattab, O. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 9866640f891944c7bb0a08748aa8b91f + 878 + 1 + + + 2.0 + Santhanam, K. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This collaboration is mentioned in the text, highlighting their joint contribution to the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + b0d513adad104e14a89a767a66f30848 + 879 + 1 + + + 2.0 + Santhanam, K. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + db37a25770a34437b472fa0038837868 + 880 + 1 + + + 2.0 + Santhanam, K. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 67746ba67d80491da102aab7704dfd30 + 881 + 1 + + + 2.0 + Santhanam, K. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 873a1ca522e6461090d5cdebc2c9ae98 + 882 + 1 + + + 2.0 + Santhanam, K. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + d919ccd28e2248b5ab1dcdd7af8b00cf + 883 + 1 + + + 1.0 + Santhanam, K. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 1883a3ca4d6a4bfd984e7053e2553e16 + 884 + 1 + + + 2.0 + Li, X. L. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 2a0c93cf781a4020aceef7230b286bbf + 885 + 1 + + + 2.0 + Li, X. L. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 06282cc8998f4b2ea43e0a9522383639 + 886 + 1 + + + 2.0 + Li, X. L. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 182efa2240c14212bb021746a18936bd + 887 + 1 + + + 2.0 + Li, X. L. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 727bc610ea9a4393bfa5de453b84340f + 888 + 1 + + + 1.0 + Li, X. L. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 960c3b508a294332ba7c05ffd897db31 + 889 + 1 + + + 2.0 + Hall, D. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + bd0c1d79ef1144a49f3ce09d4cdf099b + 890 + 1 + + + 2.0 + Hall, D. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + b8ae79ed2d6d43f98e0808b5bea884dd + 891 + 1 + + + 2.0 + Hall, D. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + a03b33a4ee97467c808946679e240ddf + 892 + 1 + + + 1.0 + Hall, D. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 00b685bcb7a54c4493cd78da1f4752ab + 893 + 1 + + + 2.0 + Liang, P. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 9a54e0361b684d24aefdc05fc340cf41 + 894 + 1 + + + 2.0 + LIANG, P. and ZAHARIA, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + d423b97f085947bd89529bc1ed2c41a7 + 895 + 1 + + + 1.0 + Liang, P. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 6fd2b0d5156c424a89cb1c068cf1e076 + 896 + 1 + + + 2.0 + Potts, C. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + a8c8dd8ddbc44363ac2102b9b8989c6d + 897 + 1 + + + 1.0 + Potts, C. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 00cb0db6e46749f7af97701ad26e23be + 898 + 1 + + + 1.0 + Zaharia, M. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 4baa53e4336d4807964fa8d186b32bc5 + 899 + 1 + + + 1.0 + Kim, G. and Kim, S. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + f325a83e0c854a7ba5d46663ddff1a29 + 900 + 1 + + + 1.0 + Kim, G. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + c2b7e5d9761e423a81149a94537f6def + 901 + 1 + + + 1.0 + Kim, G. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 3d1ccd312d3a4e7387e888aaa137c7c2 + 902 + 1 + + + 1.0 + Kim, G. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 840966e7ac4a4b14ac912e75102d50b7 + 903 + 1 + + + 1.0 + Kim, G. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + f0ede764bdb1437b8cfcc20ca9598712 + 904 + 1 + + + 1.0 + Kim, S. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + dd38d562c21f444190768c8a154280da + 905 + 1 + + + 1.0 + Kim, S. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 1e66c1cbb56b41269555d27e1505ec92 + 906 + 1 + + + 1.0 + Kim, S. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + a653bd47ad3d4009ab6a5b8e6ff18679 + 907 + 1 + + + 1.0 + Kim, S. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 37b7cf055e604ec6927a9f0b15b2698d + 908 + 1 + + + 1.0 + Jeon, B. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 182e0f3d0abd4181820acdd2bf8e5eaf + 909 + 1 + + + 1.0 + Jeon, B. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 6fd6398a9bfd496f9a0505d9f3190362 + 910 + 1 + + + 1.0 + Jeon, B. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 2192d6436ca840a1bce77dbf9fd354af + 911 + 1 + + + 1.0 + Park, J. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + ef5e0bbdb3774a22900cf45e9b8863ad + 912 + 1 + + + 1.0 + Park, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 307de250d35e43a8b122c4232fa8fb7c + 913 + 1 + + + 1.0 + Kang, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 0e2f68c8ff734b279b7aad333bcf2fda + 914 + 1 + + + 1.0 + Klein, G. and Moon, B. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 172b0d0fa0794494a3c50b135c1f2cd6 + 915 + 1 + + + 1.0 + Klein, G. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 96f016e343b34ac894b0b7153f474ab0 + 916 + 1 + + + 1.0 + Klein, G. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + d356499ae9a345b6bbfb33b5fa01f47b + 917 + 1 + + + 1.0 + Moon, B. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + d7ebead985b34576821f30d83a416cd2 + 918 + 1 + + + 1.0 + Moon, B. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + 02c1447586fc46dfa65b793e0105a878 + 919 + 1 + + + 1.0 + Hoffman, R. R. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + 1b1560bb4b0447e5860f8ba351af112e + 920 + 1 + + + 1.0 + Koesten, L. and Gregory, K. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 23b929895979486cba3bf6a13f4ce655 + 921 + 1 + + + 1.0 + Koesten, L. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + f1ebe367253a4a4088b363a6cc4601a1 + 922 + 1 + + + 1.0 + Koesten, L. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + cf575adc3eb140f9aec33757ec040eb8 + 923 + 1 + + + 1.0 + Koesten, L. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + 4e581bc7d4424c2fb0023e5b11687e02 + 924 + 1 + + + 1.0 + Gregory, K. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + ae9f8a02ac0f43d4ba67ccce412989d6 + 925 + 1 + + + 1.0 + Gregory, K. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + b972541545604529a30cabc262d83dae + 926 + 1 + + + 1.0 + Gregory, K. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + ec3f73ffbb9742e090b65893d040434b + 927 + 1 + + + 1.0 + Groth, P. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 20be7b3222174d31839fac6a278f8b61 + 928 + 1 + + + 1.0 + Groth, P. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + 35688e258b0e4cc78c8b92ef8a13d3e3 + 929 + 1 + + + 1.0 + Simperl, E. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + 3a9503f2d69343c396c9b1d842d1aa74 + 930 + 1 + + + 1.0 + Kuratov, Y. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 39516d28d39e49a2b80e6cfac32e2609 + 931 + 1 + + + 1.0 + Bulatov, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 43cf2e01543540789eb8781fdb5f287d + 932 + 1 + + + 1.0 + Anokhin, P. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 54beb6d012d844058715f8ef8a91c5da + 933 + 1 + + + 1.0 + Sorokin, D. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 5a291bf96ac141b98730ac27c96e829e + 934 + 1 + + + 1.0 + Sorokin, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 5e572d8b0a614ce1839ec9a568078cdc + 935 + 1 + + + 1.0 + Burtsev, M. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 3f7b36b371da40568ce15510a35b58e7 + 936 + 1 + + + 1.0 + Laskar, M. T. R. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + 6db7771bcc674e4ead899fbdd417930f + 937 + 1 + + + 2.0 + Laskar, M. T. R. and Hoque, E. co-authored two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning with transformer models to improve the effectiveness of query-focused abstractive summarization. Both works contribute to advancing the application of transformer models in specialized summarization tasks. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 176a96262ad64ccbacb1efdfb36bd88a + 938 + 1 + + + 1.0 + Laskar, M. T. R. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + bd120225f7b84bbdb0567048ca803e3c + 939 + 1 + + + 1.0 + Laskar, M. T. R. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + 8ecda003a3d044279b1f0bdc1c96c25e + 940 + 1 + + + 1.0 + Laskar, M. T. R. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + 8e02a61bda6a4470b693e7e234abfc94 + 941 + 1 + + + 1.0 + Hoque, E. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + b62e3858d801445facc3a501c5100723 + 942 + 1 + + + 1.0 + Hoque, E. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + 64e8971970e94ea79d10e46c55b3e761 + 943 + 1 + + + 1.0 + Hoque, E. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + b10fd3628e7a45d29a2814771f53ad60 + 944 + 1 + + + 1.0 + Hoque, E. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + 826bb16575a141d683fb871ec94517e0 + 945 + 1 + + + 1.0 + Huang, J. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + 52384316108d433397224cb36486407c + 946 + 1 + + + 1.0 + The 33rd Canadian Conference on Artificial Intelligence is also known as Canadian AI 2020 + 6cd82819982879bd164547d2773ba5c7 + 9a27717e1a1b499981031fd69c58aff1 + 947 + 1 + + + 1.0 + Springer published the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + 4efbe8fc23a64506b36d6cf29f968baa + 948 + 1 + + + 1.0 + Huang, J. X. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + d3b80330258d412f9ac6a7670fe79044 + 949 + 1 + + + 1.0 + Lewis, P. and Perez, E. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 52703e888bf4493b866186b889d85783 + 950 + 1 + + + 1.0 + Lewis, P. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 5e25a814a9a04bcda6017c9cc99880a7 + 951 + 1 + + + 1.0 + Lewis, P. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + bb2070d133f74049b88c96510fc807ba + 952 + 1 + + + 1.0 + Lewis, P. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + d0734be4aaab40eb9f2be6229f4a803c + 953 + 1 + + + 1.0 + Lewis, P. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + a5839bfcc6c0471c9337257ed05b361b + 954 + 1 + + + 1.0 + Lewis, P. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + bee441f5c41e41ff8220254bbf714eb4 + 955 + 1 + + + 1.0 + Lewis, P. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 8d93d45adbe547f78460a9ef3eb40ab2 + 956 + 1 + + + 1.0 + Lewis, P. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + c6f21de31a6f4fbda2eed1780ffed5b1 + 957 + 1 + + + 1.0 + Lewis, P. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 07c3f03764874b7680710ca030cdb60c + 958 + 1 + + + 1.0 + Perez, E. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 7a7990b6045c440ba606d142bd8ddc02 + 959 + 1 + + + 1.0 + Perez, E. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 803cb895664c40319ca40cc9abb6a03d + 960 + 1 + + + 1.0 + Perez, E. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4c9e4e36560946699b6cb1e67b1437ae + 961 + 1 + + + 1.0 + Perez, E. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 8a7a2d6266424b9f9006502e82fcd778 + 962 + 1 + + + 1.0 + Perez, E. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 6b38285eedc544b08b444ee781db9f0c + 963 + 1 + + + 1.0 + Perez, E. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9d59b69c7c984abb9d3e281c04e73510 + 964 + 1 + + + 1.0 + Perez, E. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 83ee1d8db753419f8b240f419a139815 + 965 + 1 + + + 1.0 + Perez, E. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9211b015bb074bcd89ae6c75ec10e6da + 966 + 1 + + + 1.0 + Piktus, A. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + cbc280dac03a4c19bb6737e3789c928f + 967 + 1 + + + 1.0 + Piktus, A. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 166cfa435aac4465943f59c2d04a0da1 + 968 + 1 + + + 1.0 + Piktus, A. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 44bced1d9b184aa29376cf3b0cdac625 + 969 + 1 + + + 1.0 + Piktus, A. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + ca58c4e2fae84899a780ff379e1927eb + 970 + 1 + + + 1.0 + Piktus, A. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 2060ce64f028490798a3ed69832e048d + 971 + 1 + + + 1.0 + Piktus, A. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + fa8fa48e2a7542fc8ff2c43c35e1b32b + 972 + 1 + + + 1.0 + Piktus, A. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + a37bd07ff1694b6c90572399f084e1ec + 973 + 1 + + + 1.0 + Petroni, F. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 826e32d5ea1d455f8ae2d3b77cd2b41e + 974 + 1 + + + 1.0 + Petroni, F. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 1a3bd511e04d4929a45a36fb80127353 + 975 + 1 + + + 1.0 + Petroni, F. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + d0cd49577d6a49f4a21fdc389aa84805 + 976 + 1 + + + 1.0 + Petroni, F. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + aa667f462aae45a8a700d83a68c1982f + 977 + 1 + + + 1.0 + Petroni, F. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 37f40795969a42b3b50e61b76a96fa07 + 978 + 1 + + + 1.0 + Petroni, F. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 6db48bd5e4ce4337aaac4648376ed07d + 979 + 1 + + + 1.0 + Karpukhin, V. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 422e6a449e7e4ce69182113a6493a4e5 + 980 + 1 + + + 1.0 + Xu, Y. and Lapata, M. co-authored the paper "Text summarization with latent queries" + fc4b27d64f055b7fc30176ba110dd02e + 68511afc6e204c0b996d76cb75de081c + 981 + 1 + + + 1.0 + Huang, M. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + 050f02fbf9e64d08b108c5b921581335 + 982 + 1 + + + 1.0 + Duan, N. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + 97367a34fd6b4451b6be397496d646ea + 983 + 1 + + + 3.0 + Martin, S. and Brown, W. M. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with large graph structures, making it a valuable resource for researchers and practitioners in the domain of graph theory and network analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 80da3caff5344d56b6ca12660594949a + 984 + 1 + + + 3.0 + Martin, S. and Klavans, R. co-authored the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + e7efb7b459ae4ed4aa412cd20d808970 + 985 + 1 + + + 3.0 + Martin, S. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the capabilities and applications of the Openord toolbox, emphasizing its utility in handling extensive graph data efficiently. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + f0cc756b32314a1aae3e3cbb507850a2 + 986 + 1 + + + 3.0 + Brown, W. M. and Klavans, R. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + f3082b58c8a54c538cf3a0110296955b + 987 + 1 + + + 3.0 + Brown, W. M. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + c5f2254b67c04ad4add88875e5623e5a + 988 + 1 + + + 3.0 + KLAVANS, R. and BOYACK, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 6e151c66c5574df39a7f243858e2ad3f + 989 + 1 + + + 2.0 + Newman, M. E. published the paper "Modularity and community structure in networks" in the Proceedings of the National Academy of Sciences + 833e7d67dcd30790b26b71c9b5306f6b + 166366ae9ec842ec9a1deeb13c94026e + 990 + 1 + + + 2.0 + Ram, O. and Levine, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + cdecc293edb847ae92c3bf8ff39d1e9a + 991 + 1 + + + 2.0 + Ram, O. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 3a615d980a124616a659136b4fd277b7 + 992 + 1 + + + 2.0 + Ram, O. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + a9b46d1f9ef747b69d6211386b5aaa20 + 993 + 1 + + + 2.0 + Ram, O. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 891a6dbec2ef4a039efaca78040b00c1 + 994 + 1 + + + 2.0 + Ram, O. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 981d45442a11448097acebc6080da414 + 995 + 1 + + + 2.0 + Ram, O. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + fad1c8144b504954bea46ede106d93ec + 996 + 1 + + + 2.0 + Levine, Y. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 53edb7e587204ed48e523c6f1f8f4056 + 997 + 1 + + + 2.0 + Levine, Y. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + f386b02e36884167a5db1a12ee6fcb1a + 998 + 1 + + + 2.0 + Levine, Y. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 2809d8a73b71495ca4220571dd54ba1e + 999 + 1 + + + 2.0 + Levine, Y. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 49f82fa775fb466bb9ae3db14db5b29a + 1000 + 1 + + + 2.0 + Levine, Y. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + f9312ee0bac24ea1b497e16e0958d621 + 1001 + 1 + + + 2.0 + Dalmedigos, I. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + c84bc84ffea84df9ad25ae9f972b4ec0 + 1002 + 1 + + + 2.0 + Dalmedigos, I. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + be1068c6efb24cde96e5a523eb04aee8 + 1003 + 1 + + + 2.0 + Dalmedigos, I. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + ecc4f28a7d574a5886f4c80a0b7cddd4 + 1004 + 1 + + + 2.0 + Dalmedigos, I. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + b506a4813da44600b136e949db4f2832 + 1005 + 1 + + + 2.0 + Muhlgay, D. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 77735209cabb415289c8ae4e102ff6df + 1006 + 1 + + + 2.0 + Muhlgay, D. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 1f1d88a6f6ce46bab94a4b50693c89ff + 1007 + 1 + + + 2.0 + Muhlgay, D. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 03454aaf00c54112a09ea4e52185b195 + 1008 + 1 + + + 2.0 + Shashua, A. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 3c5cd9cbad0d456cab4c76f1dfcde25b + 1009 + 1 + + + 2.0 + Shashua, A. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 8503eae9f6c746afae0caa58070f25e6 + 1010 + 1 + + + 2.0 + Leyton-Brown, K. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 6e3c655e5b544277a62832a0974aa0ed + 1011 + 1 + + + 2.0 + Ranade, P. and Joshi, A. co-authored the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + bd0363bace6b42b0b3879bed5a064274 + 1012 + 1 + + + 2.0 + Sarthi, P. and Abdullah, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 99969eec8bf8441eaf9cb004cb61a13e + 1013 + 1 + + + 2.0 + Sarthi, P. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 0bcceb946a94486faf935f58dabea978 + 1014 + 1 + + + 2.0 + Sarthi, P. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 4d8421b4a6c74627afaa45aefa08c43a + 1015 + 1 + + + 2.0 + Sarthi, P. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + f879674860784f9eb4289aeb91728351 + 1016 + 1 + + + 2.0 + Sarthi, P. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + dea50b1765e54936b3d0b1e499ab2053 + 1017 + 1 + + + 2.0 + Abdullah, S. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 26634913d18f4629b39dffa19c1df734 + 1018 + 1 + + + 2.0 + Abdullah, S. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + a69bde74fc9d41cfa669f148c7c43dd8 + 1019 + 1 + + + 2.0 + Abdullah, S. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + e65667ec99e145fea2055d6b583cb05b + 1020 + 1 + + + 2.0 + Abdullah, S. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + cdfcbba5664d42508cd34df9af42b0dc + 1021 + 1 + + + 2.0 + Tuli, A. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 570de818eba04615a6afb3a573e82ff1 + 1022 + 1 + + + 2.0 + Tuli, A. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 473533c454d34975a17a0193e39e0bac + 1023 + 1 + + + 2.0 + Tuli, A. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + e22af264c702440f93070465f45e630e + 1024 + 1 + + + 1.0 + Yang, Z. and Manning, C. D. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + ad95fbd957ae4c22859f58446dd8c9cc + 1025 + 1 + + + 1.0 + Huang, M. and Duan, N. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + 37d42129ca4d49dea240f66d1fdd4b78 + 1026 + 1 + + + 1.0 + Su, D. and Xu, Y. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1f40481f4ee342d4be51d33ffafc17d1 + 1027 + 1 + + + 1.0 + Su, D. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + e892d46e07e44bd5a2d1626875cc024f + 1028 + 1 + + + 1.0 + Su, D. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + fcb033a54d734ce5a87e0d8ad555867a + 1029 + 1 + + + 1.0 + Su, D. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 964fc01bfd9a400eb668761539dc9d9f + 1030 + 1 + + + 1.0 + Su, D. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 86b0d920fa504eba81c26cfc3f4d2b9f + 1031 + 1 + + + 1.0 + Xu, Y. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + e0221df7b7e44dd7956c8d0348d46b6d + 1032 + 1 + + + 1.0 + Xu, Y. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + a20d7f4fee104273b9628d648c05a5ac + 1033 + 1 + + + 1.0 + Xu, Y. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + b352398c5b1742d8a61acd8534ef0f53 + 1034 + 1 + + + 1.0 + Xu, Y. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 38b778af3c3f4be2a23e3932c94390c3 + 1035 + 1 + + + 1.0 + Yu, T. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 35d528e52a6441a58e58385d85bfae4b + 1036 + 1 + + + 1.0 + Yu, T. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 06d0d34ab3d043c689044a0fbfc65e10 + 1037 + 1 + + + 1.0 + Yu, T. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9e0ec036c91e44daa8e1a2af50df2081 + 1038 + 1 + + + 1.0 + Siddique, F. B. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 580b2395e68442539a606d37ddba691d + 1039 + 1 + + + 1.0 + Siddique, F. B. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 370fd1c6563045499e1d711fcd9ef9d5 + 1040 + 1 + + + 1.0 + Barezi, E. J. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 3b6c4319026844ecb645f650e30b7d1a + 1041 + 1 + + + 1.0 + Tang, Y. and Yang, Y. co-authored the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + e044334b7d2e426ca2cab7eb763d8bc9 + 1042 + 1 + + + 1.0 + Touvron, H. and Martin, L. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 4a2fa382e77946d2be8e95edc04c6a64 + 1043 + 1 + + + 1.0 + Touvron, H. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + cd015281a069460e844faeb327b7d65f + 1044 + 1 + + + 1.0 + Touvron, H. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + c7a9fbf22a054056bf4f4562eaecfc08 + 1045 + 1 + + + 1.0 + Touvron, H. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 44bf341f78d74c4bb15ae209649d0ca9 + 1046 + 1 + + + 1.0 + Touvron, H. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 045ff6885d424b4caeabc76c50468c7c + 1047 + 1 + + + 1.0 + Touvron, H. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 543a52396f0b4f7f99ea755fba11d290 + 1048 + 1 + + + 1.0 + Touvron, H. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 33e487870db646e5b3d9c1f2962a7c6a + 1049 + 1 + + + 1.0 + Touvron, H. and Bhargava, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + b5b628d809514bfe9bbb3bd362815e79 + 1050 + 1 + + + 1.0 + Touvron, H. and Bhosale, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9d9ae51d2af44ebe8324dd2dd1dcd83b + 1051 + 1 + + + 1.0 + Martin, L. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + b7c606aa6ad1416e9f934628acce5f24 + 1052 + 1 + + + 1.0 + Martin, L. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 4521342f29774fab85e6acb0490d46e5 + 1053 + 1 + + + 1.0 + Martin, L. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + e7ac741e4aa4433ca5f2379726f90b33 + 1054 + 1 + + + 1.0 + Martin, L. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 43645eb9258244a8bd334ce77216b1c0 + 1055 + 1 + + + 1.0 + Martin, L. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0c9c52488ad647abbaf2b4589c976957 + 1056 + 1 + + + 1.0 + Martin, L. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0e8fb49a531e4ea48fece73957bd8a54 + 1057 + 1 + + + 1.0 + Wang, J. and Liang, Y. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 6187217c38ca4225b97d04d9644dcdf0 + 1058 + 1 + + + 1.0 + Wang, J. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8c8330abe276487294eba3a341ee9e0c + 1059 + 1 + + + 1.0 + Wang, J. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 200df54d93964e81ae2dcf727bffb23c + 1060 + 1 + + + 1.0 + Wang, J. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 4d0478c05f614675b336a76a0c088b3e + 1061 + 1 + + + 1.0 + Wang, J. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + c0dc4d618b5e49f7a18efa34dbf450ac + 1062 + 1 + + + 1.0 + Wang, J. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + dec05f79120940b78cd921a0a67f1540 + 1063 + 1 + + + 1.0 + Wang, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8edbf3e4f0d94f6ab78127c61bf87b76 + 1064 + 1 + + + 1.0 + Wang, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1d722426930a42eeadfa624a6eb2408f + 1065 + 1 + + + 1.0 + Liang, Y. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 6d22f2009f6a4df9a242f03e2642981e + 1066 + 1 + + + 1.0 + Liang, Y. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + db5035c544214c72987eed4d4d9e327f + 1067 + 1 + + + 1.0 + Liang, Y. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + fd00e337b5c4465cbcbdf07bc294a3a8 + 1068 + 1 + + + 1.0 + Liang, Y. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + ea1e51b558c149649711a29157f4e604 + 1069 + 1 + + + 1.0 + Liang, Y. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + b888ad14e84347f8831a7dd2cea294fd + 1070 + 1 + + + 1.0 + Liang, Y. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0e2323c0256d40579e7526dbdd019a8d + 1071 + 1 + + + 1.0 + Liang, Y. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + deb3bc5537a14352b22a0a473a59d8c7 + 1072 + 1 + + + 1.0 + Meng, F. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + ba445c400c8e405bb646387eab98a62b + 1073 + 1 + + + 1.0 + Meng, F. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 95a90d05e82d44ada6f8577ca49dd491 + 1074 + 1 + + + 1.0 + Meng, F. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + fa15140701a54689835604665d187c54 + 1075 + 1 + + + 1.0 + Meng, F. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8346468f7f4c46bebe1eaafd9753d55f + 1076 + 1 + + + 1.0 + Meng, F. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 46ddbfe32d444912b423dd1769fbaa43 + 1077 + 1 + + + 1.0 + Meng, F. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + ffce5a64e9394d1399319588d7fd4e3e + 1078 + 1 + + + 1.0 + Sun, Z. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + e934202aa3b344ba9fef89ecb42530b4 + 1079 + 1 + + + 1.0 + Sun, Z. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 22d5ff62439047ccaeaa63fd8a30f3e5 + 1080 + 1 + + + 1.0 + Sun, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9b2f77b408ec4147bd5dd67a01d9f439 + 1081 + 1 + + + 1.0 + Sun, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 96b1264e89394adfaf026471e3b6ad47 + 1082 + 1 + + + 1.0 + Sun, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 00741dfb8f6d477f913d20406dfcd65d + 1083 + 1 + + + 1.0 + Shi, H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 3eb344aa8b05448984dacac93482ebc4 + 1084 + 1 + + + 1.0 + Shi, H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8db8a8680d534161b0772d7a771df6bd + 1085 + 1 + + + 1.0 + Shi, H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 7f2c628b4fa54c0b9254049602ed20d2 + 1086 + 1 + + + 1.0 + Shi, H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2f8f5d33916d4824bec6773bacd37d87 + 1087 + 1 + + + 2.0 + Li, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + b8e1f95f9d3e497393d86e6bd137fe10 + 1088 + 1 + + + 2.0 + Li, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 13a41e2ff8b847ee8073e1e23b0bffc6 + 1089 + 1 + + + 2.0 + Li, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 9c61fb5ee44744a48bc5638bd42f654b + 1090 + 1 + + + 1.0 + H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 946213d345b64cbaa6becb8723b01d87 + 1091 + 1 + + + 1.0 + Zheng, L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e75c65762f064dfc95787fa331c95392 + 1092 + 1 + + + 1.0 + Chiang, W.-L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 9a586c1629464133920fb19d8bd1e690 + 1093 + 1 + + + 1.0 + Sheng, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e7a82e12e4f84f3e82c1ec74d3088235 + 1094 + 1 + + + 1.0 + Zhuang, S. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 51828127e71d40829039e033add265c4 + 1095 + 1 + + + 1.0 + Wu, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 04918b80bc714753b00af559d439a4ec + 1096 + 1 + + + 1.0 + Zhuang, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 3b4dbcb1c7c24bf8b6d55485c0304f7e + 1097 + 1 + + + 1.0 + Lin, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + fdd2a43d9b9f450c899adfb60b05e711 + 1098 + 1 + + + 1.0 + Li, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + bced585ff9d54fb7acd03f54f5729391 + 1099 + 1 + + + 1.0 + Li, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 50aec048280a4cdb8572993faab794dd + 1100 + 1 + + + 1.0 + Li, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + e128a7179b6e476c98d6bbfecf2a3f9a + 1101 + 1 + + + 2.0 + Xu, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + a3fa60c3370e4d5e8147250e2a18104a + 1102 + 1 + + + 2.0 + Xu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 1a7ef91522514b9f8b1ddaf68424351d + 1103 + 1 + + + 1.0 + H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + e5f094bf02d84a0889cd041199156ad7 + 1104 + 1 + + + 2.0 + Qu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 5c64fc0a74044110906120ca1d5c7919 + 1105 + 1 + + + 1.0 + H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 524d1b6a01d34b0098a0da8af056bfc8 + 1106 + 1 + + + 1.0 + H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 73af37cadd3c4d3dbfb8bfd697aeef58 + 1107 + 1 + + + 1.0 + Wang, S. and Khramtsova, E. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + c6b26fcc94044c368b2fe0db4b9b72f2 + 1108 + 1 + + + 1.0 + Wang, S. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 36b9f94e17c6481fb83670b70b192eb7 + 1109 + 1 + + + 1.0 + Wang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 24c63641b4e241589336236d5f916e34 + 1110 + 1 + + + 1.0 + Khramtsova, E. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 718e03207dcd44a080806880d08268ea + 1111 + 1 + + + 1.0 + Khramtsova, E. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 6d0efadfba5046eb86869827544c2703 + 1112 + 1 + + + 1.0 + Zhuang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 2c33b12183ad4722ab1ab2cbd75f8312 + 1113 + 1 + + + 1.0 + Zheng, L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + f334bc9701204b1b943f9ece317ca68a + 1114 + 1 + + + 1.0 + Chiang, W.-L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 4ea3ecb74c91452da866f4c9163386e2 + 1115 + 1 + + + 1.0 + Sheng, Y. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + a12fe8fd9bc34db69d8de6944283d3c9 + 1116 + 1 + + + 1.0 + Zhuang, S. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 02b877f830cc4bb58dfad02f13a6d6ce + 1117 + 1 + + + 1.0 + Zhuang, S. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + abffcf93dc114332a181990ad56b7863 + 1118 + 1 + + + 1.0 + Zhuang, S. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + f894b0ae91eb412d93b6b06d4a73f350 + 1119 + 1 + + + 1.0 + Zhuang, S. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 75e06eb1f93c4ee38b782965ea905b5b + 1120 + 1 + + + 1.0 + Zhuang, S. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 70fd5c73bbe34d918b3dca3fc7294d28 + 1121 + 1 + + + 1.0 + Zhuang, S. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 8f1edc1d00764d6fb23859242c659deb + 1122 + 1 + + + 1.0 + Wang, Y. and Lipka, N. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + c8554314efb44679a898bbce08372abe + 1123 + 1 + + + 1.0 + Wang, Y. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + a41dbc4319f74bc995a93dbe0f4d9aee + 1124 + 1 + + + 1.0 + Wang, Y. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + ef76a41bf9bf45c893c475a7bd5a2938 + 1125 + 1 + + + 1.0 + Wang, Y. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 88f486cbb1904425a5fd5dfa268cf85d + 1126 + 1 + + + 1.0 + Wang, Y. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + ec10a8695b1a4e8787d9d29114e9d5ce + 1127 + 1 + + + 1.0 + Lipka, N. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 9397185bb4d7492b88eaa20fa10c0ae5 + 1128 + 1 + + + 1.0 + Lipka, N. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 2b1b9b1ed49c4ace91ff099752b8c0a5 + 1129 + 1 + + + 1.0 + Lipka, N. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 62c66c25992d4974829678313ed60b1d + 1130 + 1 + + + 1.0 + Lipka, N. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 26a889667b614ab890d863c4b8762e69 + 1131 + 1 + + + 1.0 + Rossi, R. A. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 30461855b0604128a4f10d0b348ce60f + 1132 + 1 + + + 1.0 + Rossi, R. A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + edef06de541f493f98d9281a704d785d + 1133 + 1 + + + 1.0 + Rossi, R. A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 0016a9dec22543e9b203f540860bf2e7 + 1134 + 1 + + + 1.0 + Siu, A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4f79558a259f4de58df5b022b68a459e + 1135 + 1 + + + 1.0 + Siu, A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + ca07919df74f4e5abfbd370c50eacc00 + 1136 + 1 + + + 1.0 + Zhang, R. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 94b8715919cd49d08ac0ce99b930ea53 + 1137 + 1 + + + 1.0 + Yang, Z. and Qi, P. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4a10c341918b4d888b8b0466bd101b1d + 1138 + 1 + + + 1.0 + Yang, Z. and Zhang, S. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 5173ce4188034717b9c90eef40b94932 + 1139 + 1 + + + 1.0 + Yang, Z. and Bengio, Y. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + cff3415902bf4745992473697570aef0 + 1140 + 1 + + + 1.0 + Yang, Z. and Cohen, W. W. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4c0cf727ec2843a288aa00b43f25b2de + 1141 + 1 + + + 1.0 + Yang, Z. and Salakhutdinov, R. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 87ece3a4dcc84c98a291c1138ae56544 + 1142 + 1 + + + 1.0 + Zheng, L. and Chiang, W.-L. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 143dc5f4cb4b4596900ee5158594b1b0 + 1143 + 1 + + + 1.0 + Zheng, L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 3c222c5fdfab46e1ac1352a0f85a5fdd + 1144 + 1 + + + 1.0 + Zheng, L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 8c13a465b483417691c9b8d40b913da3 + 1145 + 1 + + + 1.0 + Zheng, L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 51633d2f0adf4123a23eeb292d95e649 + 1146 + 1 + + + 1.0 + Zheng, L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 849136ae8c9f4f9589a989bfe4c4155d + 1147 + 1 + + + 1.0 + Zheng, L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 84b11b37d1dd4e75b4c453669fbd4df9 + 1148 + 1 + + + 1.0 + Zheng, L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 373c198a0ed2402cb885b8d9f9de92f3 + 1149 + 1 + + + 1.0 + Zheng, L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 64ec8c4eb0734d60a5287e3df62652bd + 1150 + 1 + + + 1.0 + Chiang, W.-L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 7636104f26794a4e9e74b2d6943c879d + 1151 + 1 + + + 1.0 + Chiang, W.-L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + fc4b406a34ea4b2d9f305600aab14ea3 + 1152 + 1 + + + 1.0 + Chiang, W.-L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 5400473bab9d4105a1517fdc55c58f17 + 1153 + 1 + + + 1.0 + Chiang, W.-L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 624a1e6ea1d248f8b5126527e82e76c0 + 1154 + 1 + + + 1.0 + Chiang, W.-L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 66e53a4f6fc740aaaa379aa63d15f0e9 + 1155 + 1 + + + 1.0 + Chiang, W.-L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e676d0167a3f43478a209ec9526c90df + 1156 + 1 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 44fa3d2247904198b1c776e060d35eb2 + 1157 + 1 + + + 1.0 + Sheng, Y. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + a135859c7d3d4d3596f1e4ab218eff8a + 1158 + 1 + + + 1.0 + Sheng, Y. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 5e6fd98384a24a34b80311842661e00a + 1159 + 1 + + + 1.0 + Sheng, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 3a10d5261d4240c7b05b6cdb7838ff24 + 1160 + 1 + + + 1.0 + Sheng, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + ec595c7b07e148dba900040a68ef0fdb + 1161 + 1 + + + 1.0 + Sheng, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 4b43619c5c6a4ea3826bfd3c06aa6e66 + 1162 + 1 + + + 1.0 + Sheng, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 3173671571f14c75bfb9141754424efa + 1163 + 1 + + + 1.0 + Wu, Z. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 1675e75d7b524d4ab236eeaefd2dc992 + 1164 + 1 + + + 1.0 + Wu, Z. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + bb906c8e869141aa9be12118dcd3d3b5 + 1165 + 1 + + + 1.0 + Wu, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 27c95f7d6c3d4732897ae7bffd7c5dc8 + 1166 + 1 + + + 1.0 + Wu, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + b63c467faf714acd8a006431faf7a141 + 1167 + 1 + + + 1.0 + Wu, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 19a5840a67e14c468f9f3d6851eaee5c + 1168 + 1 + + + 1.0 + Zhuang, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 7a58673144d849e7a784caee9d9d4e99 + 1169 + 1 + + + 1.0 + Zhuang, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 8076db94b7214fdf9e006ce5a7e1cbe2 + 1170 + 1 + + + 1.0 + Zhuang, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e733886404db4d46862bdddb2aee5211 + 1171 + 1 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + b75594a9d9c8404688a5cfe02272cdfc + 1172 + 1 + + + 1.0 + Lin, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + f253ff2311024729a758bb77b14bf72d + 1173 + 1 + + + 1.0 + Lin, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 1f0cb1e7fae64c238efb659d254d6221 + 1174 + 1 + + + 1.0 + Lin, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 6740be36e0e14774a5551a17db648a13 + 1175 + 1 + + + 1.0 + Li, D. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 0f926e9dfaae4615b16a794e984b85ae + 1176 + 1 + + + 1.0 + Li, D. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 47851446a4df4f5aa4505c999daaaaf7 + 1177 + 1 + + + 1.0 + Xing, E. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + c3e51a2782ba4f86b49b4038a316d9fb + 1178 + 1 + + + \ No newline at end of file diff --git a/graphfleet/output/graphindex/artifacts/clustered_graph.2.graphml b/graphfleet/output/graphindex/artifacts/clustered_graph.2.graphml new file mode 100644 index 000000000..ecd13be4d --- /dev/null +++ b/graphfleet/output/graphindex/artifacts/clustered_graph.2.graphml @@ -0,0 +1,15824 @@ + + + + + + + + + + + + + + + + + + + PERSON + Darren Edge is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 0 + b45241d70f0e43fca764df95b2b81f77 + + + PERSON + Ha Trinh is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 1 + 4119fd06010c494caa07f439b333f4c5 + + + PERSON + Newman Cheng is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 2 + d3835bf3dda84ead99deadbeac5d0d7d + + + PERSON + Joshua Bradley is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 3 + 077d2820ae1845bcbb1803379a3d1eae + + + PERSON + Alex Chao is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 4 + 3671ea0dd4e84c1a9b02c5ab2c8f4bac + + + PERSON + Apurva Mody is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 5 + 19a7f254a5d64566ab5cc15472df02de + + + PERSON + Steven Truitt is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 6 + e7ffaee9d31d4d3c96e04f911d0a8f9e + + + PERSON + Jonathan Larson is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8 + 7 + f7e11b0e297a44a896dc67928368f600 + + + ORGANIZATION + Microsoft Research is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + 3 + 8 + 1fd3fa8bb5a2408790042ab9573779ee + + + ORGANIZATION + Microsoft Strategic Missions and Technologies is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + 3 + 9 + 27f9fbe6ad8c4a8b9acee0d3596ed57c + + + ORGANIZATION + Microsoft Office of the CTO is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + 2 + 10 + e1fd0e904a53409aada44442f23a51cb + + + METHOD + RAG (Retrieval-Augmented Generation) is a developing research area with multiple established directions, including knowledge graph creation, completion, and extraction of causal graphs. It is a method used for generating responses in text generation tasks by retrieving relevant information from an external knowledge source to enable large language models to answer questions. This approach incorporates the retrieval of relevant data to augment text generation, producing direct responses in various text generation tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 31 + 11 + de988724cfdf45cebfba3b13c43ceede + + + TECHNOLOGY + LLM (Large Language Model) is a type of artificial intelligence model used for a variety of tasks in the field of Natural Language Processing and Information Retrieval. These tasks include generating and assessing text, entity extraction, summarization, understanding relationships in text, and automating human-like sensemaking and reasoning over large collections of documents. LLMs are also employed to generate intermediate answers and scores for text chunks, process these chunks to extract elements of a graph index, and automate the generation of questions for dataset evaluation. Additionally, LLMs can analyze and generate text based on retrieved information and queries, and they possess a context window that can be exceeded by external datasets. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,2c6ed90897310eea2f28e33fff1c32b0,6f33a085ff3304e5994f7fbb86c881a4,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 77 + 2 + 32 + 12 + 96aad7cb4b7d40e9b7e13b94a67af206 + + + METHOD + Graph RAG (Retrieval-Augmented Generation) is a sophisticated method that leverages the natural modularity of graphs to partition data for global summarization tasks. This approach integrates multiple stages and concepts, including knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS), to support human sensemaking over text corpora. It is particularly effective in providing comprehensive and structured overviews of public figures across various sectors of the entertainment industry, as well as generating answers for questions in the News article dataset. + +Graph RAG employs a high-level data flow and pipeline for processing and summarizing text, combining both global and local approaches to optimize token usage in text generation tasks. It uses community summaries to improve answer comprehensiveness and diversity while requiring fewer tokens compared to traditional source text summarization methods. This method has been shown to outperform naive RAG in terms of comprehensiveness and diversity in text generation tasks. + +A specific implementation of Graph RAG involves using four levels of graph communities, incorporating concepts from other systems such as self-memory and parallel generation of community answers. This multi-stage mechanism allows for the comparison of multiple conditions, enhancing the overall effectiveness of the summarization process. + +Graph RAG, launched by NebulaGraph, is a retrieval-augmented generation technology based on knowledge graphs. It combines global summarization with graph-based text indexing to answer questions over private text corpora, making it a versatile tool for various text analysis and summarization applications. + 086021a89900a39bcb62036981737bfa,21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,718017a4871c909420f84b85b8ba969d,833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19,fb3c48579608fa28be585ceb6cd2f0fe + 87 + 2 + 90 + 13 + c9632a35146940c2a86167c7726d35e9 + + + METHOD + QFS (Query-Focused Summarization) is a method used to generate summaries based on specific user queries + e8d83e6e7a7c0f57b218cef24976b745 + 87 + 2 + 1 + 14 + 9646481f66ce4fd2b08c2eddda42fc82 + + + CONCEPT + Community summaries are generated summaries of data clusters or communities, used to answer queries and tailored to the domain. These summaries are pre-generated for groups of closely-related entities, particularly in the Graph RAG approach, and are derived from community-generated content to compare with source texts. They are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks. Community summaries provide report-like insights into each community within a hierarchical structure, which is useful for understanding the dataset. They are generated from modular communities in the knowledge graph and cover different levels of each graph community hierarchy, including root-level communities in an entity-based graph index. Additionally, these summaries act as a kind of self-memory for generation-augmented retrieval. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + 99 + 2 + 26 + 15 + d91a266f766b4737a06b0fda588ba40b + + + CONCEPT + Global sensemaking questions are questions that require understanding and summarizing large datasets, often beyond the explicit content of the source texts + e8d83e6e7a7c0f57b218cef24976b745 + 3 + 16 + bc0e3f075a4c4ebbb7c7b152b65a5625 + + + METRIC + 1 million token range refers to the scale of datasets used in the evaluation of the Graph RAG approach + e8d83e6e7a7c0f57b218cef24976b745 + 1 + 17 + 254770028d7a4fa9877da4ba0ad5ad21 + + + TECHNOLOGY + Python is a programming language used for implementing both global and local Graph RAG approaches. Additionally, Python is utilized to implement the open-source version of the Graph RAG approach. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + 87 + 2 + 1 + 18 + 4a67211867e5464ba45126315a122a8a + + + URL + The URL "HTTPS://AKA.MS/GRAPHRAG" is the location where the open-source, Python-based implementation of Graph RAG approaches will be available. This URL serves as the repository for accessing the open-source implementation of the Graph RAG approach. + e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745 + 2 + 19 + 04dbbb2283b845baaeac0eaf0c34c9da + + + METHOD + Query-Focused Summarization (QFS) is a method used to generate summaries that are relevant to specific user queries. This summarization technique focuses on answering specific queries by utilizing the entire corpus of information available. It is designed to provide concise and relevant information based on the specific needs of the user, ensuring that the generated summaries are directly aligned with the queries posed. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + 5 + 20 + 1943f245ee4243bdbfbd2fd619ae824a + + + CONCEPT + An external knowledge source is a repository of information that can be accessed to retrieve relevant data for answering questions + e8d83e6e7a7c0f57b218cef24976b745 + 1 + 21 + 273daeec8cad41e6b3e450447db58ee7 + + + CONCEPT + A text corpus is a large collection of written texts used for analysis and research + e8d83e6e7a7c0f57b218cef24976b745 + 1 + 22 + e69dc259edb944ea9ea41264b9fcfe59 + + + CONCEPT + An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + 96 + 2 + 2 + 23 + e2f5735c7d714423a2c4f61ca2644626 + + + CONCEPT + Source documents are the original texts from which information is extracted, retrieved, or summarized. These documents serve as the foundational input texts for various processing tasks, including the Graph RAG (Retrieval-Augmented Generation) approach. In this context, source documents are critical for extracting and analyzing information, ensuring that the data used in computational linguistics and information retrieval tasks is accurate and comprehensive. + bc9e2c9e369c4108cf4f6dd5f60960f4,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + 96 + 2 + 5 + 24 + deece7e64b2a4628850d4bb6e394a9c3 + + + CONCEPT + A partial response is an intermediate answer generated from community summaries before being combined into a final response + e8d83e6e7a7c0f57b218cef24976b745 + 101 + 2 + 2 + 25 + e657b5121ff8456b9a610cfaead8e0cb + + + CONCEPT + A final response is the comprehensive answer generated after combining all partial responses + e8d83e6e7a7c0f57b218cef24976b745 + 101 + 2 + 1 + 26 + bf4e255cdac94ccc83a56435a5e4b075 + + + METRIC + COMPREHENSIVENESS is a metric used to evaluate the quality of generated responses by measuring how much detail an answer provides to cover all aspects and details of a question. It assesses the completeness and thoroughness of answers, ensuring that they encompass all relevant information. This metric is particularly important in evaluating the summarization approach, focusing on the completeness of the summary. In practical applications, such as evaluating Podcast transcripts and News articles, comprehensiveness has shown win rates between 72-83% and 72-80%, respectively. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + 91 + 2 + 9 + 27 + 3b040bcc19f14e04880ae52881a89c1c + + + METRIC + DIVERSITY is a metric used to evaluate the variety and richness of answers generated in response to a question. It measures how varied and rich an answer is in providing different perspectives and insights. This metric is particularly important in assessing the quality of summarization approaches, focusing on the variety of information included in the summary. DIVERSITY is applied to various types of content, including Podcast transcripts, where win rates range from 75-82%, and News articles, with win rates ranging from 62-71%. It is a crucial target quality for evaluating the effectiveness of different methods in generating diverse and informative responses. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + 87 + 2 + 7 + 28 + 3d6b216c14354332b1bf1927ba168986 + + + ACTIVITY + Human endeavors refer to activities and efforts across various domains that rely on reading and reasoning about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + 2 + 29 + 1c109cfdc370463eb6d537e5b7b382fb + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models designed to understand and generate human-like text, playing a crucial role in automating sensemaking in complex domains. Modern language models, such as GPT, Llama, and Gemini, leverage in-context learning to effectively summarize content. These models are integral to the field of Natural Language Processing and Information Retrieval, enabling sophisticated text analysis and generation capabilities. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + 11 + 30 + 3d0dcbc8971b415ea18065edc4d8c8ef + + + DOMAIN + Scientific discovery is a complex domain where sensemaking is applied to understand and generate new knowledge from scientific texts + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 31 + 68105770b523412388424d984e711917 + + + DOMAIN + Intelligence analysis is a complex domain where sensemaking is applied to understand and generate insights from intelligence data + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 32 + 85c79fd84f5e4f918471c386852204c5 + + + PROCESS + SENSEMAKING is the process of understanding and making sense of complex information. It involves understanding connections among people, places, and events to anticipate their trajectories and act effectively. This process is crucial for navigating and interpreting intricate data landscapes, enabling individuals and organizations to make informed decisions based on the relationships and patterns identified within the information. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 7 + 33 + eae4259b19a741ab9f9f6af18c4a0470 + + + DATA + TEXT CHUNKS are segments of text that are embedded into a vector space for analysis. These segments are extracted from source documents and are used for processing in the Graph RAG (Retrieval-Augmented Generation) approach. By embedding these text chunks into a vector space, they can be analyzed more effectively, facilitating various natural language processing and information retrieval tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + 95 + 2 + 5 + 34 + 3138f39f2bcd43a69e0697cd3b05bc4d + + + DATA + Element instances are identified and extracted instances of graph nodes and edges from text chunks. They represent individual occurrences of entities, relationships, and claims extracted from source texts. These specific pieces of information are tailored to the domain, providing a structured representation of the underlying data. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 4 + 35 + dde131ab575d44dbb55289a6972be18f + + + DATA + Element summaries are concise representations of element instances, tailored to the domain. They are descriptive texts created by the LLM to summarize entities, relationships, and claims extracted from source texts. These summaries provide detailed descriptions of nodes, edges, and covariates within a community, and are used to understand the structure and semantics of the dataset. In essence, element summaries serve as a tool to encapsulate and convey the intricate details of elements within a graph, facilitating a deeper comprehension of the dataset's structural dynamics and semantic relationships. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 10 + 36 + de9e343f2e334d88a8ac7f8813a915e5 + + + DATA + Graph communities are groups of elements, including nodes, edges, and covariates, detected within a graph index, primarily used for summarization. These communities consist of groups of nodes that exhibit stronger connections to each other than to nodes outside the group. This structural characteristic allows for the identification and analysis of densely connected subgraphs, which can be crucial for understanding the underlying relationships and dynamics within complex networks. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 3 + 37 + e2bf260115514fb3b252fd879fb3e7be + + + DATA + COMMUNITY ANSWERS are query-focused summaries derived from community summaries. These answers are generated in parallel from the community summaries and are designed to respond to user queries effectively. Essentially, COMMUNITY ANSWERS serve as responses that synthesize information from community summaries to provide concise and relevant answers to specific questions posed by users. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + 99 + 2 + 3 + 38 + b462b94ce47a4b8c8fffa33f7242acec + + + DATA + GLOBAL ANSWER is a comprehensive response generated from multiple community summaries to answer a user query. It is the final query-focused summary produced from all relevant community summaries. The final answer is generated by combining intermediate community answers based on their helpfulness scores. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 7 + 39 + 17ed1d92075643579a712cc6c29e8ddb + + + TIME + Indexing time refers to the time when the graph index is created and elements are summarized + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 40 + 3ce7c210a21b4deebad7cc9308148d86 + + + TIME + Query time refers to the time when a query is made and the relevant summaries are generated + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 41 + d64ed762ea924caa95c8d06f072a9a96 + + + PROCESS + Graph RAG pipeline is a process using an LLM-derived graph index to detect, extract, and summarize nodes, edges, and covariates in source documents + f0306814bf64f5c9e79603fc6a52f4ea + 7 + 42 + adf4ee3fbe9b4d0381044838c4f889c8 + + + DATA + NODES are entities detected in the graph index of source documents. They represent the individual elements or points in a graph. For instance, in the Podcast dataset, there are 8,564 nodes, while the News dataset contains 15,754 nodes. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + 74 + 2 + 3 + 43 + 32ee140946e5461f9275db664dc541a5 + + + DATA + EDGES are relationships detected in the graph index of source documents. They represent the connections or links between nodes in a graph. For instance, in the Podcast dataset, there are 20,691 edges, while the News dataset contains 19,520 edges. These edges are crucial for understanding the structural dynamics and relationships within the datasets, providing insights into how different nodes (such as topics, entities, or documents) are interconnected. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + 74 + 2 + 3 + 44 + c160b9cb27d6408ba6ab20214a2f3f81 + + + DATA + Covariates are additional attributes associated with extracted node instances in the graph index. They represent claims or additional information detected in the graph index of source documents. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 2 + 45 + 23527cd679ff4d5a988d52e7cd056078 + + + METHOD + LEIDEN is a community detection algorithm renowned for its efficiency in recovering hierarchical community structures. It is widely used to partition graphs into modular communities, effectively grouping elements within a graph index. The algorithm's ability to identify and organize these communities makes it a valuable tool in the analysis of complex networks, particularly within the domains of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 7 + 46 + f1c6eed066f24cbdb376b910fce29ed4 + + + METHOD + Retrieval-Augmented Generation (RAG) is an established approach in the field of Natural Language Processing and Information Retrieval, designed to answer user questions over entire datasets. This method involves retrieving relevant text regions to provide grounding for the generation task, thereby enhancing the accuracy and relevance of the generated responses. By combining retrieval and generation processes, RAG effectively synthesizes and presents pertinent information, making it a powerful tool for handling complex queries and large datasets. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + 4 + 47 + 83a6cb03df6b41d8ad6ee5f6fef5f024 + + + ORGANIZATION + Microsoft is a technology company whose Chief Technology Officer, Kevin Scott, actively participates in podcast conversations. The organization is deeply involved in automating sensemaking in scientific discovery through the use of large language models (LLMs). Notably, Microsoft conducted a study examining the impact of large language models, specifically GPT-4, on scientific discovery. + 1d07b4248c2655081c7af0e373bd70c9,833e7d67dcd30790b26b71c9b5306f6b,f0306814bf64f5c9e79603fc6a52f4ea + 3 + 48 + 147c038aef3e4422acbbc5f7938c4ab8 + + + PERSON + Ranade is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 49 + b7702b90c7f24190b864e8c6e64612a5 + + + PERSON + Joshi is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 50 + de6fa24480894518ab3cbcb66f739266 + + + PERSON + Klein is an author who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 51 + 6fae5ee1a831468aa585a1ea09095998 + + + PERSON + Lewis is an author who contributed to the development of the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 52 + ef32c4b208d041cc856f6837915dc1b0 + + + PERSON + Traag is an author who contributed to the development of the Leiden community detection method + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 53 + 07b2425216bd4f0aa4e079827cb48ef5 + + + PUBLICATION + arXiv is a preprint repository where several significant papers in the field of Natural Language Processing and Information Retrieval have been published. It serves as a platform for electronic preprints (known as e-prints) that are approved for publication after moderation, but not full peer review. Notable papers published on arXiv include "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models," "Lost in the middle: How language models use long contexts," "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," "Llama 2: Open foundation and fine-tuned chat models," "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy," "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries," "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions," "Enhancing knowledge graph construction using large language models," "Is chatgpt a good nlg evaluator? a preliminary study," "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt," "Causal graph discovery with retrieval-augmented generation based large language models," "Knowledge graph prompting for multi-document question answering," "Text summarization with latent queries," "Retrieval-augmented generation for large language models: A survey," and "Knowledge graph-augmented language models for knowledge-grounded dialogue generation." This repository is a crucial resource for researchers to disseminate their findings rapidly and access the latest advancements in their fields. + 00e8e4e881bd0862022f4dfc913b900b,086021a89900a39bcb62036981737bfa,58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035,f0306814bf64f5c9e79603fc6a52f4ea,fc4b27d64f055b7fc30176ba110dd02e + 81 + 2 + 39 + 54 + 2670deebfa3f4d69bb82c28ab250a209 + + + PUBLICATION + Preprint refers to the version of the research paper that is under review and available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 4 + 55 + 404309e89a5241d6bff42c05a45df206 + + + CATEGORY + cs.CL is the category under which the research paper is classified on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 56 + b785a9025069417f94950ad231bb1441 + + + DATE + 24 Apr 2024 is the date when the research paper was submitted to arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 57 + 3b6cd96a27304614850709aba1c9598b + + + IDENTIFIER + 2404.16130v1 is the identifier for the research paper on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 58 + d54956b79dd147f894b67a8b97dcbef0 + + + DATA + Document collections refer to large sets of documents that are analyzed for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 59 + 958beecdb5bb4060948415ffd75d2b03 + + + TECHNOLOGY + LLM PROMPTS are specific instructions given to large language models (LLMs) to tailor their responses to the domain of the dataset. These prompts are also used to extract elements from text chunks, ensuring that the LLMs provide relevant and precise information based on the given context. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 2 + 60 + b999ed77e19e4f85b7f1ae79af5c002a + + + METHOD + Community detection is a method used to identify groups of related elements within a graph. It involves the process of identifying communities within a graph, which are clusters of nodes that are more densely connected internally than with the rest of the network. This technique is crucial in understanding the structural dynamics and relationships within complex networks, such as those found in social networks, biological systems, and information retrieval systems. By uncovering these communities, researchers can gain insights into the underlying structure and function of the network, facilitating more effective analysis and interpretation of the data. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 3 + 61 + 48c0c4d72da74ff5bb926fa0c856d1a7 + + + METHOD + Domain-tailored summarization is a method used to create summaries that are specific to the domain of the dataset + f0306814bf64f5c9e79603fc6a52f4ea + 2 + 62 + 4f3c97517f794ebfb49c4c6315f9cf23 + + + PERSON + Klein et al. are authors who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 63 + 1745a2485a9443bab76587ad650e9be0 + + + PERSON + Ranade and Joshi are authors who discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 64 + 32e6ccab20d94029811127dbbe424c64 + + + PERSON + Lewis et al. are authors who developed the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + 1 + 65 + 94a964c6992945ebb3833dfdfdc8d655 + + + PERSON + Traag et al. are the authors who developed the Leiden algorithm, a method renowned for its efficiency in recovering hierarchical community structures. This algorithm is widely recognized in the field of Natural Language Processing and Information Retrieval for its ability to accurately detect and map out complex community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 1 + 66 + 1eb829d0ace042089f0746f78729696c + + + METHOD + QFS is a task framing that focuses on generating summaries based on specific queries, rather than just concatenating excerpts + fb3c48579608fa28be585ceb6cd2f0fe + 1 + 67 + 015e7b58d1a14b44beab3bbc9f912c18 + + + METHOD + A type of summarization that generates natural language summaries based on specific queries, rather than just extracting text + fb3c48579608fa28be585ceb6cd2f0fe + 3 + 68 + 26f88ab3e2e04c33a459ad6270ade565 + + + TECHNOLOGY + A neural network architecture that has shown substantial improvements in various summarization tasks + fb3c48579608fa28be585ceb6cd2f0fe + 3 + 69 + babe97e1d9784cffa1c85abc1e588126 + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + 3 + 70 + 1033a18c45aa4584b2aef6ab96890351 + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + 2 + 71 + c9b8ce91fc2945b4907fe35519339cac + + + TECHNOLOGY + GEMINI is a family of highly capable multimodal models, as described in an arXiv preprint. These models are known for their ability to perform in-context learning and summarization, making them a significant advancement in the field of Natural Language Processing and Information Retrieval. + 086021a89900a39bcb62036981737bfa,fb3c48579608fa28be585ceb6cd2f0fe + 2 + 72 + fa3c4204421c48609e52c8de2da4c654 + + + TECHNOLOGY + A knowledge graph is a structured representation of information, utilized in the Graph RAG approach for summarization. This structured representation of knowledge is specifically employed in the Graph RAG approach for global summarization, highlighting its role in organizing and integrating information to facilitate comprehensive and coherent summaries. + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + 88 + 2 + 2 + 73 + 53af055f068244d0ac861b2e89376495 + + + REFERENCE + Authors of a paper on Retrieval-augmented generation (RAG) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 74 + c03ab3ce8cb74ad2a03b94723bfab3c7 + + + REFERENCE + Author of a paper on query-focused summarization (QFS) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 75 + ed6d2eee9d7b4f5db466b1f6404d31cc + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 76 + fc01e9baa80e417c9206f941bb279407 + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 77 + 56d0e5ebe79e4814bd1463cf6ca21394 + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 78 + 7c49f2710e8b4d3b8dc9310834406ea5 + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 79 + c6d1e4f56c2843e89cf0b91c10bb6de2 + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 80 + 0adb2d9941f34ef7b2f7743cc6225844 + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 81 + 6b02373137fd438ba96af28f735cdbdb + + + REFERENCE + Authors of a paper on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 82 + 36a4fcd8efc144e6b8af9a1c7ab8b2ce + + + REFERENCE + "BROWN ET AL., 2020" refers to a publication by Brown et al. in 2020, which discusses in-context learning with few-shot examples. The authors of this paper are also known for their work on the GPT series of large language models. + bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 2 + 83 + fbeef791d19b413a9c93c6608286ab63 + + + REFERENCE + Authors of a paper on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 84 + d2b629c0396f4180a03e16ddf3818589 + + + REFERENCE + Authors of a paper on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 1 + 85 + 6102fc6619ed422ebc42588bfa97355d + + + REFERENCE + "KURATOV ET AL., 2024" refers to a publication by Kuratov and colleagues in 2024. The study discusses the recall degradation and potential for information loss in longer context windows of Large Language Models (LLMs). The authors explore the limitations of these extended context windows, providing insights into how the performance of LLMs can be affected when dealing with longer sequences of text. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 3 + 86 + 8d141c0b80f74b79a05eed7fe161fe49 + + + REFERENCE + "LIU ET AL., 2023" refers to a publication by Liu et al. in 2023, which discusses the recall degradation and potential for information loss in longer context windows of large language models (LLMs). The authors explore the limitations of LLM context windows, highlighting how extended contexts can lead to decreased recall accuracy and information retention. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + 3 + 87 + e22d1d1cd8d14f12b81828d940f40d70 + + + TECHNOLOGY + COMMUNITY DETECTION ALGORITHMS are algorithms used to partition a graph into communities of nodes with stronger connections to one another. These algorithms are designed to identify modular communities of closely-related nodes within a graph, thereby revealing the underlying structure and relationships within the network. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + 6 + 88 + 9ab48505fb1b487babd0d1f6d3a3f980 + + + ALGORITHM + Louvain is a community detection algorithm used to partition graphs into modular communities + 21e52bc06a82796b1f4bcd73edda1f2a + 1 + 89 + 148fffeb994541b2b4b6dcefda7001a8 + + + DATASET + HOTPOTQA is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical resource for evaluating entity extraction prompts, particularly with advanced models like GPT-4-turbo. Additionally, HotPotQA is utilized to observe the behavior of text chunk extraction within the Graph RAG (Retrieval-Augmented Generation) approach, making it a versatile tool in the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4 + 3 + 90 + 89c08e793298442686292454a1abff31 + + + TECHNOLOGY + GPT-4-Turbo is a version of the GPT-4 model characterized by its large context size of 128k tokens, which is utilized in various analytical tasks. Specifically, GPT-4-Turbo is employed for entity extraction in evaluations, leveraging its extensive context capacity to enhance the accuracy and comprehensiveness of the analysis. This model is particularly suited for tasks within the Natural Language Processing and Information Retrieval domain, where handling large volumes of text and extracting relevant entities are critical. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + 2 + 91 + 0467928aa65e4a4fba62bdb1467e3a54 + + + DATASET + The "PODCAST TRANSCRIPTS" dataset is a comprehensive collection of compiled transcripts from podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders. This dataset is used for analysis and consists of 1669 text chunks, each containing 600 tokens with 100-token overlaps between chunks, amounting to approximately 1 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620,ebf5249c888e07fedce6572a4c03f88c + 72 + 2 + 19 + 92 + 43c3390303c6476cb65f584e37c3e81c + + + DATASET + The "NEWS ARTICLES" dataset is a comprehensive collection of news articles used for analysis. It serves as a benchmark dataset comprising news articles published from September 2013 to December 2023. The dataset spans a range of categories, including entertainment, business, sports, technology, health, and science. It consists of 3197 text chunks, each containing 600 tokens, with a 100-token overlap between chunks, amounting to approximately 1.7 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620 + 71 + 2 + 13 + 93 + fa14b16c17e3417dba5a4b473ea5b18d + + + METHOD + MAP-REDUCE is a method employed for text summarization by applying a map-reduce approach directly to source texts. It is particularly utilized for query-focused summarization of an entire corpus, enabling efficient processing and extraction of relevant information from large datasets. This technique leverages the map-reduce paradigm to distribute the computational workload, making it suitable for handling extensive text collections in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,973164fa90bf2b4ee267f4fd795916bf + 2 + 94 + 7cc3356d38de4328a51a5cbcb187dac3 + + + METRIC + "EMPOWERMENT" is a concept and metric used in the evaluation of various methods, with an average win rate of 51.3%. It measures how well an answer helps the reader understand and make informed judgments about a topic. Specifically, it evaluates the effectiveness of generated answers in empowering users by developing their understanding of broad issues and themes. Empowerment is a target quality in summarization approaches, focusing on the ability to help users reach an informed understanding. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,ebf5249c888e07fedce6572a4c03f88c + 6 + 95 + bef16fb5fd7344cca5e295b13ef3e0cd + + + METHOD + Naive RAG is a basic retrieval-augmented generation (RAG) method used as a baseline for comparison in text generation tasks. It converts documents to text, splits them into chunks, and embeds these chunks into a vector space for query matching. While it produces the most direct responses, it is outperformed by global approaches in terms of comprehensiveness and diversity. Naive RAG is also noted for listing public figures mentioned in entertainment articles, focusing on their professional achievements and personal lives. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c + 91 + 2 + 13 + 96 + bb9e01bc171d4326a29afda59ece8d17 + + + METHOD + A method for summarizing source texts using a map-reduce approach + 21e52bc06a82796b1f4bcd73edda1f2a + 87 + 2 + 1 + 97 + 3c063eea52e94164b70c99431ea30bae + + + OUTPUT + Questions generated to evaluate the summarization approach, focusing on understanding activities + 21e52bc06a82796b1f4bcd73edda1f2a + 87 + 2 + 1 + 98 + 252cc8452bfc4c2aa58cab68d8b61879 + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + 87 + 2 + 1 + 99 + 7e2c84548fb94ee395ba8588d8f2a006 + + + METRIC + TOKEN COSTS refer to the computational cost measured in tokens used in the summarization process. Specifically, in the context of the Graph RAG (Retrieval-Augmented Generation) approach, token costs denote the number of tokens required for processing text. This metric is crucial for evaluating the efficiency and scalability of text processing methods within the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 87 + 2 + 1 + 100 + f034618dde7948beb6dab30176d0fc87 + + + PROCESS + The high-level process of the Graph RAG approach and pipeline + 21e52bc06a82796b1f4bcd73edda1f2a + 87 + 2 + 1 + 101 + 5c41f96be13e49dba649454297834546 + + + PARAMETER + Design parameters are key settings and configurations in the Graph RAG approach. These parameters are crucial as they influence the design of the Graph RAG approach and pipeline, determining the effectiveness and efficiency of the overall system. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + PARAMETER + 87 + 2 + 1 + 102 + 7ea4afbf8a264f29af29950ce98105ba + + + METHOD + GLOBAL SUMMARIZATION is a method for summarizing information on a global scale. It aims to encapsulate the overall structure and semantics of a dataset, providing a comprehensive overview of information from large datasets or corpora. This technique is particularly useful in the field of Natural Language Processing and Information Retrieval, where it helps in distilling vast amounts of data into coherent and concise summaries, facilitating better understanding and analysis of the underlying information. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e4d9b12cf2b4c691c74019eefff4fb39 + 5 + 103 + 91ff849d12b24574b0691dbddf44968b + + + ATTRIBUTE + Modularity is an inherent quality of graphs that allows them to be partitioned into communities of closely-related nodes + 21e52bc06a82796b1f4bcd73edda1f2a + 88 + 2 + 1 + 104 + d73c1f2fb3094d8dace42ad2a76e9a52 + + + OUTPUT + Descriptions generated from modular communities in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + 99 + 2 + 1 + 105 + cdc8901e668749889bd49bebdc4ff1f6 + + + INPUT + A specific question or request for information that the summarization methods aim to answer + 21e52bc06a82796b1f4bcd73edda1f2a + 87 + 2 + 1 + 106 + 36084a9fab53433493f079e97e68bf65 + + + DATASET + A large collection of texts or documents used for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + 87 + 2 + 1 + 107 + eebcc7ec8e3e4df7aea83659bbdc2199 + + + OUTPUT + Intermediate answers generated from community summaries before being combined into a final global answer + 21e52bc06a82796b1f4bcd73edda1f2a + 100 + 2 + 2 + 108 + ceadf262ef834e9ab146b20650912cae + + + OUTPUT + The comprehensive answer generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + 100 + 2 + 1 + 109 + 7f65feab75424b53b24470d305ba331a + + + METHOD + A method that focuses on generating questions to understand activities from datasets + 21e52bc06a82796b1f4bcd73edda1f2a + 2 + 110 + fd9cb733b28d420cb5cef01e545a132c + + + INPUT + Brief descriptions of datasets used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + 1 + 111 + 0fbcca3f17c649a08aea64b5a7d9ef36 + + + DATASET + Datasets that represent real-world information, such as podcast transcripts and news articles + 21e52bc06a82796b1f4bcd73edda1f2a + 87 + 2 + 1 + 112 + 482027a59f32484c9c44fd700615c1b6 + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + 87 + 2 + 1 + 113 + de837ff3d626451282ff6ac77a82216d + + + METHOD + A method that summarizes the original source texts directly + 21e52bc06a82796b1f4bcd73edda1f2a + 87 + 2 + 1 + 114 + 460295fed3ae4cd39f9f274cec9c2506 + + + OUTPUT + LOW-LEVEL COMMUNITY SUMMARIES are a type of community summary used in the News dataset for analysis. These summaries provide a detailed overview of the source text and are generated from lower hierarchical levels of the community in the knowledge graph. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,bc9e2c9e369c4108cf4f6dd5f60960f4 + 96 + 2 + 3 + 115 + 553b285bba60460ab1ed8341ae61282b + + + OUTPUT + INTERMEDIATE-LEVEL COMMUNITY SUMMARIES are summaries that provide a mid-level overview of the source text. These summaries are generated from intermediate hierarchical levels of the community in the knowledge graph, offering a balanced perspective that captures essential details without overwhelming the reader with excessive information. This approach ensures that the summaries are both informative and concise, making them valuable for understanding the structural dynamics and relationships within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 96 + 2 + 2 + 116 + cec95bf17e7e4c939b56c9c6f402a29f + + + OUTPUT + Summaries generated from higher hierarchical levels of the community in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + 87 + 2 + 1 + 117 + 599164aead034bc19446efacc77554d2 + + + PROCESS, SYSTEM + The entity "PIPELINE" refers to a series of processes or steps used to analyze and summarize a dataset. Specifically, in the context of the Graph RAG approach, the pipeline denotes the sequence of steps and processes involved. This structured sequence is essential for systematically handling data, ensuring that each stage of the analysis is methodically executed to achieve accurate and comprehensive results. + 7fb7d9ce2da9c940a32afdd87d1d9e56,bc9e2c9e369c4108cf4f6dd5f60960f4 + 2 + 118 + bbf148ae4d48422f8fdef754cfa2b9e4 + + + DATA STRUCTURE, OUTPUT + The "GRAPH INDEX" is a data structure used in Retrieval-Augmented Generation (RAG) systems to organize and retrieve information. It is a self-generated index that enables Graph RAG by utilizing a graph structure to organize and retrieve data. This index is created from a graph structure and is employed for tasks such as query-focused summarization. The graph index includes various elements extracted from text chunks using Large Language Model (LLM) prompts. Additionally, it supports conditions C0-C3 and is created using generic prompts for entity and relationship extraction. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + 18 + 119 + de61b2670999433f807a6a1dc2b81e43 + + + DATA, UNIT + Entity references are mentions of entities within text chunks, extracted during the processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + 97 + 2 + 3 + 120 + 3e95dacfe57b4d57b5da4310ef2e157f + + + METRIC + Recall is a metric used to measure the completeness of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 97 + 2 + 1 + 121 + 1f1545308e9347af91fd03b94aadc21f + + + METRIC + Precision is a metric used to measure the accuracy of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 97 + 2 + 1 + 122 + 6ea81acaf232485e94fff638e03336e1 + + + TECHNIQUE, METHOD + FEW-SHOT EXAMPLES are specialized instances provided to the Large Language Model (LLM) to improve its performance in domains with specialized knowledge such as science, medicine, and law. These examples are tailored to the domain of the data used in the graph indexing process and serve as sample inputs for in-context learning. By tailoring the extraction prompt to the document corpus domain, few-shot examples enhance the LLM's ability to understand and process domain-specific information effectively. + 2c6ed90897310eea2f28e33fff1c32b0,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4 + 78 + 2 + 8 + 123 + d136b08d586d488f9e4188b524c85a29 + + + DATA, UNIT + Named entities are specific types of entities such as people, places, and organizations, extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 77 + 2 + 1 + 124 + cccfa151fedc4b218a8d96adc7dceabe + + + REFERENCE, PUBLICATION + A reference to a publication by Yang et al. in 2018, introducing the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + 1 + 125 + ce54725672a74ebcabe6127577dacb2b + + + METHOD, APPROACH + Techniques refer to the specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 87 + 2 + 1 + 126 + ea2b28ca1a974ffab4517811dc1d1e5c + + + ATTRIBUTE, CONFIGURATION + Implementation details are specific configurations and settings used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 87 + 2 + 1 + 127 + aff21f1da1654e7babdcf3fb0e4a75fc + + + PROCESS, METHOD + A single extraction round refers to one complete cycle of extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + 79 + 2 + 1 + 128 + dc2cc9016e3f49dbac7232f05cce794d + + + ATTRIBUTE, CONFIGURATION + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + 95 + 2 + 1 + 129 + 6ea0cef05f694dcea455478f40674e45 + + + METRIC, ISSUE + Recall degradation refers to the decrease in recall performance when using longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + 77 + 2 + 1 + 130 + 7ab5d53a872f4dfc98f3d386879f3c75 + + + PROCESS, METHOD + The extraction process involves identifying and extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + 79 + 2 + 2 + 131 + af1d0fec22114a3398b8016f5225f9ed + + + ATTRIBUTE, CONFIGURATION + Domain refers to the specific area of knowledge or field to which the document corpus belongs + bc9e2c9e369c4108cf4f6dd5f60960f4 + 98 + 2 + 1 + 132 + b07a7f088364459098cd8511ff27a4c8 + + + DATA, INPUT + Document corpus refers to the collection of documents being processed in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 98 + 2 + 2 + 133 + 8870cf2b5df64d2cab5820f67e29b9f1 + + + TECHNIQUE, METHOD + Default prompt is the standard set of instructions given to the LLM for extracting named entities + bc9e2c9e369c4108cf4f6dd5f60960f4 + 78 + 2 + 2 + 134 + cd130938a2844050be991af70baf5ee0 + + + TECHNIQUE, METHOD + Secondary extraction prompt is an additional set of instructions given to the LLM for extracting covariates + bc9e2c9e369c4108cf4f6dd5f60960f4 + 78 + 2 + 2 + 135 + 43544b99c3b04b059546198a0ae6366d + + + METHOD + A covariate prompt is used to extract additional attributes associated with detected entities, including claims linked to the entities + 2c6ed90897310eea2f28e33fff1c32b0 + 2 + 136 + a671bf7fea2f4514b6e96ba99127fafd + + + CONCEPT + Claims are statements or assertions linked to detected entities, including details such as subject, object, type, description, source text span, and start and end dates + 2c6ed90897310eea2f28e33fff1c32b0 + 7 + 137 + 525f41ea20274a05af4e52b625b473f3 + + + METHOD + Gleanings refer to multiple rounds of entity extraction to ensure that no entities are missed in the process + 2c6ed90897310eea2f28e33fff1c32b0 + 77 + 2 + 1 + 138 + 071a416efbec4f0886c19ac68f6d43cb + + + TECHNIQUE + Logit bias is a technique used to force a yes/no decision from the LLM during the entity extraction process + 2c6ed90897310eea2f28e33fff1c32b0 + 77 + 2 + 1 + 139 + 6d8473ef3b1042bf87178a611e3dbcc6 + + + CONCEPT + An entity node is a representation of an entity in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 140 + 30c9641543c24773938bd8ec57ea98ab + + + CONCEPT + A relationship edge is a representation of a relationship between entities in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 141 + 18b839da898e4026b81727d759d95c6a + + + CONCEPT + A claim covariate is an additional attribute or variable associated with a claim in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 142 + eeef6ae5c464400c8755900b4f1ac37a + + + CONCEPT + Communities of entities are groups of closely-related entities detected and summarized by the LLM + 2c6ed90897310eea2f28e33fff1c32b0 + 80 + 2 + 2 + 143 + 422433aa45804c7ebb973b2fafce5da6 + + + CONCEPT + The "NOISY GRAPH STRUCTURE" refers to a graph structure that may contain inconsistencies or errors, making it challenging to analyze. This type of graph often includes duplicate or inconsistent entity elements due to variations in text format. These inconsistencies can arise from various sources, such as data entry errors, differing data formats, or incomplete information, which complicate the process of extracting meaningful insights and relationships from the graph. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56 + 80 + 2 + 1 + 144 + 86505bca739d4bccaaa1a8e0f3baffdc + + + DOMAIN + Science is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + 78 + 2 + 1 + 145 + 1af9faf341e14a5bbf4ddc9080e8dc0b + + + DOMAIN + Medicine is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + 78 + 2 + 1 + 146 + 353d91abc68648639d65a549e59b5cf3 + + + DOMAIN + Law is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + 78 + 2 + 1 + 147 + 7ce637e4f35b42e3a9f8272cab69cd22 + + + ATTRIBUTE + Source text span is an attribute of a claim that indicates the specific portion of text from which the claim was extracted + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 148 + 4d999d7744b04a998475f8f8531589f0 + + + ATTRIBUTE + Start date is an attribute of a claim that indicates when the event or fact described in the claim began + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 149 + 9a6f414210e14841a5b0e661aedc898d + + + ATTRIBUTE + End date is an attribute of a claim that indicates when the event or fact described in the claim ended + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 150 + db541b7260974db8bac94e953009f60e + + + ATTRIBUTE + Description is an attribute of a claim that provides a detailed explanation of the claim + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 151 + f2ff8044718648e18acef16dd9a65436 + + + ATTRIBUTE + Subject is an attribute of a claim that indicates the main entity involved in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 152 + 00d785e7d76b47ec81b508e768d40584 + + + ATTRIBUTE + Object is an attribute of a claim that indicates the entity that is affected by the subject in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + 1 + 153 + 87915637da3e474c9349bd0ae604bd95 + + + CONCEPT + A concept referring to an entity that has multiple name variations but is resilient to such variations due to sufficient connectivity to closely-related entities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1 + 154 + 8f1eba29f39e411188200bf0d14628ec + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models capable of understanding and generating human-like text. They are utilized for a variety of tasks, including the creation and completion of knowledge graphs, which are essential for structuring and interlinking information in a meaningful way. Additionally, LLMs serve as evaluators of natural language generation, assessing the quality and coherence of text produced by other AI systems. These models play a crucial role in the field of Natural Language Processing and Information Retrieval, contributing significantly to advancements in how machines comprehend and interact with human language. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf + 5 + 155 + 7282c73622b8408e97289d959faff483 + + + TECHNOLOGY + Structured representations of knowledge in the form of triples (subject, predicate, object) used for reasoning tasks + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1 + 156 + 3deb220d31f74103aa44870a36a63220 + + + CONCEPT + Nodes in a graph that are of the same type and are described using rich descriptive text + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 2 + 157 + af7a1584dd15492cb9a4940e285f57fc + + + CONCEPT + Edges in a graph that represent relationships between entity nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 2 + 158 + 6e8d9029ce4e4ea182367173ab2c7bbf + + + METRIC + Weights assigned to edges in a graph, representing the normalized counts of detected relationship instances + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1 + 159 + cbf232211e7d4eb6abdbe182f71c2cf0 + + + CONCEPT + The "HIERARCHICAL COMMUNITY STRUCTURE" is a framework in which communities are organized in a hierarchy, with each level providing a partition of the graph nodes. This structure organizes data into a hierarchy of communities, facilitating a multi-level clustering approach. Hierarchical community structure is utilized to generate community summaries, offering a comprehensive method for understanding the relationships and structural dynamics within specialized communities. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39 + 7 + 160 + bb0cff774a4440b289cc6f3b929fe13c + + + CONCEPT + A division of graph nodes into mutually-exclusive, collectively-exhaustive communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 2 + 161 + ce55841ebfdd47008bab8c258f10372e + + + TECHNOLOGY + MULTIHOP-RAG is a benchmark dataset comprising news articles published from September 2013 to December 2023, covering a range of categories including entertainment, business, sports, technology, health, and science. It is specifically designed for open-domain question answering, targeting explicit fact retrieval. Additionally, MULTIHOP-RAG represents a specific implementation or variant of Retrieval-Augmented Generation (RAG) used in the context of graph communities. This dataset is also utilized for community detection and analysis, making it a versatile tool in the field of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,922778ce1cb2fdd6dbab1746c8795620 + 8 + 162 + 6090e736374d45fd84f0e4610a314f8f + + + PERSON + An author who has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1 + 163 + 0e8d921ccd8d4a8594b65b7fd19f7120 + + + PERSON + Authors who have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1 + 164 + 59c726a8792d443e84ab052cb7942b4a + + + CONCEPT + The entity "DATASET" refers to a collection of data used for various purposes such as analysis, summarization, and evaluation. This can include diverse types of data like podcast transcripts and news articles. Specifically, the term encompasses datasets used for evaluation purposes, including notable examples like the Podcast and News datasets. + 1d07b4248c2655081c7af0e373bd70c9,7fb7d9ce2da9c940a32afdd87d1d9e56,973164fa90bf2b4ee267f4fd795916bf + 3 + 165 + 4f2c665decf242b0bfcaf7350b0e02ed + + + CONCEPT + GLOBAL QUERIES refer to questions or inquiries that require comprehensive answers derived from multiple sources or datasets. These queries aim to retrieve information from a global perspective, covering the entire dataset. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 2 + 166 + 66cdf168f36d4a57a505028c97dc06e0 + + + CONCEPT + ROOT COMMUNITIES are the top-level clusters in a hierarchical community structure. These communities represent the highest level of organization within the hierarchy, serving as the primary divisions from which more specific sub-communities branch out. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 4 + 167 + 38f51478f41f48db9bee570859b6f43e + + + CONCEPT + SUB-COMMUNITIES are lower-level clusters within root communities in a hierarchical community structure, providing more detailed information. These sub-communities play a crucial role in breaking down the larger, more general root communities into more specific and focused groups, thereby facilitating a deeper and more granular understanding of the overall community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 5 + 168 + 896d2a51e8de47de85ba8ced108c3d53 + + + TECHNOLOGY + Detailed documents that provide information about specific subtopics within a community + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1 + 169 + 14555b518e954637b83aa762dc03164e + + + CONCEPT + The division of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 2 + 170 + b1f6164116d44fe8b8f135d7f65b9e58 + + + CONCEPT + A system in which elements are ranked or organized in levels + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 3 + 171 + c8b2408617804483b620e1a6691ac90d + + + CONCEPT + LEVEL 0 represents the root-level communities in the hierarchical clustering with maximum modularity. It serves as the foundational layer in a hierarchical community structure, indicating the initial and most significant division of the dataset into distinct groups. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 2 + 172 + a5e0d1644eb547ba9a5c3211aac4631a + + + CONCEPT + LEVEL 1 is a sub-level in a hierarchical community structure, providing more detailed information about the internal organization. Specifically, Level 1 represents sub-communities within the root-level communities, thereby revealing the internal structure and dynamics of these larger groups. This level of granularity helps in understanding the intricate relationships and specialized interactions that occur within the broader community framework. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + 2 + 173 + 5a28b94bc63b44edb30c54748fd14f15 + + + CONCEPT + A visual representation of graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1 + 174 + f97011b2a99d44648e18d517e1eae15c + + + METHOD + The Leiden algorithm is a method used for detecting communities in large networks + 843fc5421e086120ffa1c75856ecf6cd + 1 + 175 + 35489ca6a63b47d6a8913cf333818bc1 + + + TOOL + OpenORD is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + 1 + 176 + 5d3344f45e654d2c808481672f2f08dd + + + TOOL + Force Atlas 2 is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + 1 + 177 + 6fb57f83baec45c9b30490ee991f433f + + + ELEMENT + Nodes represent entities in a graph, with size proportional to their degree + 843fc5421e086120ffa1c75856ecf6cd + 2 + 178 + 68762e6f0d1c41cd857c6b964a8e76c3 + + + ELEMENT + Edges represent connections between nodes in a graph + 843fc5421e086120ffa1c75856ecf6cd + 2 + 179 + 70634e10a5e845aa8c6a32fe7e8eb2b2 + + + ELEMENT + Covariates are variables that are linked to nodes and edges in a graph + 843fc5421e086120ffa1c75856ecf6cd + 2 + 180 + 04085f7cf46544b79597fc49286ff84d + + + CONCEPT + The LLM context window is the token limit within which summaries are added for processing by a language model + 843fc5421e086120ffa1c75856ecf6cd + 2 + 181 + d203efdbfb2f4b2a899abfb31cf72e82 + + + METHOD + Hierarchical clustering is a method of clustering data into a tree-like structure with multiple levels + 843fc5421e086120ffa1c75856ecf6cd + 2 + 182 + 6731a665561840c2898ce8c9788e4c88 + + + CONCEPT + The token limit is the maximum number of tokens that can be processed in a single context window by a language model + 843fc5421e086120ffa1c75856ecf6cd + 1 + 183 + 4026806fa92f4e849a59a7f5c9a45c79 + + + CONCEPT + Summary detail refers to the level of detail provided in a summary + 843fc5421e086120ffa1c75856ecf6cd + 99 + 2 + 1 + 184 + 68e0c60d2e8845d89d9d0ad397833648 + + + CONCEPT + Scope refers to the range or extent of information covered in a summary + 843fc5421e086120ffa1c75856ecf6cd + 99 + 2 + 1 + 185 + 101572f552b54e529fe7765c05168981 + + + CONCEPT + A "USER QUERY" is a question or inquiry posed by a user seeking information, which the system aims to answer. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd + CONCEPT + 2 + 186 + 60c58026b2764b40adffca6eaa31d6d9 + + + ELEMENT + Chunks are segments of community summaries divided into pre-specified token sizes + 843fc5421e086120ffa1c75856ecf6cd + ELEMENT + 99 + 2 + 1 + 187 + ad1595a78935472999444c9330e7730e + + + METRIC + Prominence is a metric used to prioritize community edges based on the combined degree of source and target nodes + 843fc5421e086120ffa1c75856ecf6cd + 2 + 188 + 735d19aea0744b2295556841c5c4c3fd + + + METRIC + Combined source and target node degree is a metric used to measure the overall prominence of community edges + 843fc5421e086120ffa1c75856ecf6cd + 1 + 189 + c725babdb14a485582f8fbdf95429030 + + + ELEMENT + Community edges are connections between nodes within a community, prioritized based on prominence + 843fc5421e086120ffa1c75856ecf6cd + 1 + 190 + a0047221896d418d849847d422fa4bb8 + + + CONCEPT + Sub-community summaries are shorter summaries of sub-communities used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + 1 + 191 + 98fc2ee593184c5a839454db4eec7013 + + + CONCEPT + Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + 99 + 2 + 1 + 192 + 80020a1da63042459e00266b2a605452 + + + CATEGORY + Community level refers to the different levels in the hierarchical community structure used to generate summaries + 843fc5421e086120ffa1c75856ecf6cd + 1 + 193 + 31a7e680c4d54101afe4c8d52d246913 + + + DATA + Chunks are segments of community summaries divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + 3 + 194 + 351abba16e5c448994c6daf48121b14d + + + METRIC + A score between 0-100 generated by the LLM to indicate how helpful an answer is in addressing the target question + 1d07b4248c2655081c7af0e373bd70c9 + 3 + 195 + 50ea7d3b69614bcdbfbff7ddbfbf3d34 + + + USER + A user looking for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + 6 + 196 + 004f40a5aeca48a1879db728eb12bcba + + + USER + A user incorporating current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + 6 + 197 + 4465efb7f6ed4dedad72a658184addd2 + + + TOPIC + A topic dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 198 + b0dd60e11dad4ff782623acf039b3948 + + + TOPIC + A topic discussing the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 199 + db8c43fa4df947b09e5754d3b1393ead + + + TOPIC + A topic discussing the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 200 + 5dabc4cd05da425cb194a04482bf0c29 + + + TOPIC + A topic discussing suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 201 + 9d08f285a7be4c79b8f359c51d51db37 + + + TOPIC + A topic discussing collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 202 + adffed660d154b519c1817e514e83096 + + + TOPIC + Current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 203 + b7e9c9ef572c445a9574ca571e41fb96 + + + TOPIC + A topic addressing the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 204 + dcb9f281cd6248c699e0ebb285a42a5e + + + TOPIC + Examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 205 + 072cdee531b74513984f49d99a8d64a0 + + + TOPIC + Insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 206 + 5ae335d9210a45fda3f92a9a028d6d9b + + + TOPIC + The importance of health literacy highlighted through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 207 + 5ac60a941a5b4934bdc43d2f87de601c + + + OUTPUT + Answers generated for each chunk of community summaries + 1d07b4248c2655081c7af0e373bd70c9 + 3 + 208 + d405c3154d0e48ce96fad4c28fe20590 + + + METRIC + The pre-specified size of tokens used to divide community summaries into chunks + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 209 + 7923d8521c744bd9aab131c1aea91ffd + + + TECHNOLOGY + The "CONTEXT WINDOW" refers to a window of text used to generate answers, constrained by token size. The size of the context window is consistent across all conditions, ensuring uniformity in answer generation processes. + 1d07b4248c2655081c7af0e373bd70c9,973164fa90bf2b4ee267f4fd795916bf + 2 + 210 + 5bd156c87ec44e19ae6f8f62e6e50b9d + + + PERSON + Kevin Scott is the Chief Technology Officer (CTO) of Microsoft and actively participates in podcast conversations. His involvement in these discussions is documented and compiled in the dataset, highlighting his contributions to the field of technology and his role in shaping Microsoft's strategic direction. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + 72 + 2 + 2 + 211 + c1a146d7fb16429ea6d0aa2a55ee597f + + + PERSON + Individuals who are leaders in the technology industry and participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + 72 + 2 + 1 + 212 + ede9350632084da5b0b577ff799ab14b + + + INPUT + A specific activity or goal that the user aims to achieve using the datasets + 1d07b4248c2655081c7af0e373bd70c9 + 2 + 213 + ed559fb4ebde45518849ec803b350fa3 + + + INPUT + QUESTIONS refer to specific inquiries generated by the Large Language Model (LLM) based on the user's task and the target datasets. These questions are utilized in the analysis to evaluate the performance of different methods within the domain of Natural Language Processing and Information Retrieval. The generation and subsequent use of these questions are crucial for assessing the effectiveness and accuracy of various computational techniques and models. + 1d07b4248c2655081c7af0e373bd70c9,4c855404ee3d3c94aa2136f1513c666f + 4 + 214 + f422035f8b78417f98e4d116971cf9f3 + + + + + 1d07b4248c2655081c7af0e373bd70c9 + 1 + 215 + c79d686eba044c5586c706cdc096817d + + + DATASET + MT-BENCH is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical tool in evaluating the performance of models in retrieving accurate and relevant information from a broad range of topics. MT-Bench is prominently featured in the academic paper titled "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," where it is utilized to assess the capabilities of large language models in a structured and rigorous manner. This dataset plays a significant role in advancing the field of Natural Language Processing and Information Retrieval by providing a standardized metric for model comparison and evaluation. + 922778ce1cb2fdd6dbab1746c8795620,b1bbda43309e8e0e2175ea034aa88e13 + DATASET + 84 + 2 + 12 + 216 + 0f70db1e598d463fbbcdd1e288bd9490 + + + PROCESS + The process through which people inspect, engage with, and contextualize data within the broader scope of real-world activities + 922778ce1cb2fdd6dbab1746c8795620 + PROCESS + 1 + 217 + b35c3d1a7daa4924b6bdb58bc69c354d + + + TECHNOLOGY + Retrieval-Augmented Generation systems used for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + TECHNOLOGY + 71 + 2 + 3 + 218 + a97e2ecd870944cfbe71c79bc0fcc752 + + + AUTHORS + Authors of a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 1 + 219 + 3e1b063bbfa9423d84e50311296d2f3c + + + AUTHORS + Authors of a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 1 + 220 + 9a8ce816ee954bdabd01ea2081538009 + + + AUTHORS + Authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 1 + 221 + 09f18f81442d4d6d93a90f0fac683f9b + + + AUTHORS + Authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 1 + 222 + e02be3e37ca0454883a4c1fd859c24bb + + + AUTHORS + Authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + 84 + 2 + 1 + 223 + 6e0c81bef5364c988b21bf9b709d9861 + + + + + 922778ce1cb2fdd6dbab1746c8795620 + 1 + 224 + 1dbc51475cb04dafa4a8833a8378635e + + + PODCAST + "BEHIND THE TECH" is a podcast series featuring conversations between Kevin Scott and other technology leaders. It serves as a media platform associated with Kevin Scott, providing insights and discussions on various technological advancements and industry trends. + 833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620 + 0 + 225 + c12b9ebd8b4e42b7896822a32e3fa6eb + + + PERSON + Kevin Scott, Microsoft CTO, who participates in the podcast conversations compiled in the dataset + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 226 + 27505f6ade4b4e5f9316ffe9c34821f7 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 227 + 0ee7db2c6bea4630ba9f0c25e8a967ad + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 228 + 5a6c1d15424149f69052cd8d91fbff75 + + + DATASET + A benchmark dataset for open-domain question answering, targeting explicit fact retrieval + 922778ce1cb2fdd6dbab1746c8795620 + 0 + 229 + d005bf75c31d4848ad7041f39651e59c + + + METRIC + N represents the number of test questions per dataset used in the evaluation + 973164fa90bf2b4ee267f4fd795916bf + 1 + 230 + 9b3eef8f3a3a45e6873838db95295b8a + + + METHOD + A method applying a map-reduce approach directly to source texts for summarization + 973164fa90bf2b4ee267f4fd795916bf + 3 + 231 + fdc954b454744820804d7798f3e0b5de + + + METHOD + A na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached + 973164fa90bf2b4ee267f4fd795916bf + 2 + 232 + 49c1383836934ec495c3b35769100a73 + + + CATEGORY + C0 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a root-level community summary, which is utilized to answer user queries by providing the fewest number of summaries. This category is essential for understanding the structural dynamics within the community, particularly in the domain of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 73 + 2 + 8 + 233 + 859dedcc3736439a8a563419f16cb3d8 + + + CATEGORY + C1 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a high-level community summary used to answer user queries, effectively representing sub-communities of C0. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 72 + 2 + 9 + 234 + 6078b9980a6c4dcd9198d151b833ead7 + + + CATEGORY + C2 is a category or condition used in the analysis, representing a specific subset of the data. It functions as an intermediate-level community summary used to answer user queries, representing sub-communities of C1. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 73 + 2 + 9 + 235 + f93cd6b8213e46dda67af7e5382e1bd2 + + + CATEGORY + C3 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a category or condition that encapsulates low-level community summaries, which are instrumental in answering user queries. These summaries represent sub-communities of C2, providing detailed insights into the structural dynamics and relationships within the broader community. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 72 + 2 + 9 + 236 + 496f17c2f74244c681db1b23c7a39c0c + + + METHOD + TS, or "Text Summarization," is a category or condition used in the analysis, representing a specific subset of the data. It is particularly focused on source text summarization within the analysis. TS employs a text summarization method that applies a map-reduce approach directly to source texts, facilitating efficient and scalable summarization processes. This category is integral to understanding and processing large volumes of text data, making it a crucial component in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 72 + 2 + 10 + 237 + da1684437ab04f23adac28ff70bd8429 + + + METHOD + "SS" is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a baseline condition and is associated with a na¨ıve RAG (Retrieval-Augmented Generation) approach. In this context, text chunks are retrieved and added to the context window until the token limit is reached. + 4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + 4 + 238 + 4517768fc4e24bd2a790be0e08a7856e + + + CONCEPT + The prompts used for answer generation, which are the same across all conditions with minor modifications + 973164fa90bf2b4ee267f4fd795916bf + 1 + 239 + 545edff337344e518f68d1301d745455 + + + DATASET + The "PODCAST DATASET" is a collection of podcast transcripts utilized for both analysis and evaluation purposes. This dataset is specifically designed to support various analytical tasks, providing a rich source of textual data for researchers and practitioners in the field of Natural Language Processing and Information Retrieval. The transcripts within the dataset offer valuable insights and serve as a critical resource for evaluating different computational models and techniques. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + 74 + 2 + 15 + 240 + 9376ce8940e647a99e5e087514b88fa4 + + + DATASET + The "NEWS DATASET" is a collection of news articles utilized for both analysis and evaluation purposes. This dataset serves as a valuable resource for examining and assessing various aspects of news content, making it an essential tool in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + 74 + 2 + 15 + 241 + b38a636e86984600bb4b57c2e2df9747 + + + CONCEPT + METRICS in the context of Natural Language Processing and Information Retrieval are essential tools used to evaluate the performance of natural language generation. These metrics include both reference-based metrics, which compare generated texts to a set of reference texts, and qualities of the generated texts themselves. They are crucial in the analysis to assess the effectiveness of different methods in generating natural language, ensuring that the outputs are both accurate and of high quality. + 4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + 3 + 242 + 4bc7440b8f4b4e4cae65a5c49defa923 + + + REFERENCE + "WANG ET AL., 2023A" refers to a study conducted by Wang and colleagues in 2023, which highlights the effectiveness of Large Language Models (LLMs) in evaluation. This study is a significant contribution to the field, providing insights into the capabilities and performance of LLMs in various evaluative tasks. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 2 + 243 + 5d1b038ce8be4533b54dd79d6496de9b + + + REFERENCE + "ZHENG ET AL., 2024" refers to a study conducted by Zheng and colleagues in 2024. This study highlights the effectiveness of Large Language Models (LLMs) in evaluation processes. The research, authored by Zheng et al., provides significant insights into the capabilities and applications of LLMs within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 3 + 244 + ac6e5a44e0c04a4fa93589376fde4c34 + + + REFERENCE + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + 1 + 245 + 40e4ef7dbc98473ba311bd837859a62a + + + CONCEPT + The entity "CONDITIONS" refers to the different scenarios or variables that are compared in an experiment. Specifically, in the context of the analysis, these conditions include Graph RAG, text summarization, and semantic search RAG. These conditions are used to evaluate and compare various aspects of performance and effectiveness within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 87 + 2 + 3 + 246 + 222f0ea8a5684123a7045986640ec844 + + + CONCEPT + USER QUERIES refer to the inquiries made by users to retrieve information. These queries are answered using different methods and conditions, depending on the context and the specific requirements of the information retrieval process. + 973164fa90bf2b4ee267f4fd795916bf,e4d9b12cf2b4c691c74019eefff4fb39 + 73 + 2 + 6 + 247 + 668cf1fdfd644d39acc6350b86117ea2 + + + CONCEPT + Types of entities extracted during the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + 1 + 248 + 478e4c72d8fb46dd8cc9f0691c9878fd + + + METRIC + The "CONTEXT WINDOW SIZE" refers to the fixed size of the context window used in various stages of natural language processing and information retrieval tasks. For the final evaluation, the context window size is set to 8k tokens. During the analysis phase, different context window sizes are tested, including 8k, 16k, 32k, and 64k tokens. Additionally, in the graph indexing process, the context window size is set to 600 tokens. This variability in context window sizes highlights the importance of adapting the window size to the specific requirements of different tasks within the domain. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + 9 + 249 + 82b0446e7c9d4fc793f7b97f890e9049 + + + CONCEPT + The process of extracting information, with 1 gleaning for the Podcast dataset and 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + 74 + 2 + 2 + 250 + 8169efeea3ce473d9fd2f1c688126a1c + + + TECHNOLOGY + Natural Language Generation (NLG) is a subfield of artificial intelligence that focuses on generating human-like text from data + 322e02986c8724eedbcf3ebfa20b989c + 7 + 251 + c2d48b75af6a4d7989ccf9eceabd934e + + + METHOD + A method where a Large Language Model (LLM) is used to compare and evaluate competing outputs + 322e02986c8724eedbcf3ebfa20b989c + 1 + 252 + 5f1fc373a8f34050a5f7dbd8ac852c1b + + + METHOD + A method for evaluating the performance of Retrieval-Augmented Generation (RAG) systems, focusing on context relevance, faithfulness, and answer relevance + 322e02986c8724eedbcf3ebfa20b989c + 4 + 253 + 0c010fa3aeac4b28b2fbb8c2339c2521 + + + PUBLICATION + A reference to a study or paper authored by Es and others in 2023 + 322e02986c8724eedbcf3ebfa20b989c + 1 + 254 + c2999bdca08a478b84b10219875b285e + + + TOOL + A Large Language Model (LLM) used to evaluate and compare generated texts based on specific metrics + 322e02986c8724eedbcf3ebfa20b989c + 11 + 255 + 263d07354a1b4336b462024288f9bcd3 + + + METRIC + DIRECTNESS is a metric that measures how specifically and clearly an answer addresses a question. It is used to evaluate the straightforwardness of the generated answers. Additionally, it serves as a validity test metric to measure the directness of responses, with naive RAG (Retrieval-Augmented Generation) producing the most direct responses. + 322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 3 + 256 + f9005e5c01b44bb489f7112322fd1162 + + + DATA + An example of LLM-generated assessment shown in a table format + 322e02986c8724eedbcf3ebfa20b989c + 1 + 257 + d9ef017549724f4fbc4ff4ba6701dac0 + + + DATA + The entity "QUESTION" refers to a specific query used in the evaluation process, particularly as a metric to evaluate the generated responses by asking specific questions. This approach is commonly employed in the domain of Natural Language Processing and Information Retrieval to assess the quality and relevance of responses generated by various models or systems. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + 94 + 2 + 2 + 258 + 33b9e826af3f43838c07c847b6349497 + + + ENTITY + Public figures are individuals who have gained fame or notoriety in various sectors such as entertainment, sports, and digital media. These individuals are well-known in the entertainment industry and are frequently mentioned across various articles. Their prominence in public discourse spans multiple domains, reflecting their influence and recognition in society. + 322e02986c8724eedbcf3ebfa20b989c,718017a4871c909420f84b85b8ba969d + 94 + 2 + 5 + 259 + dbe9063124d047dc8d6fcaeadcda038f + + + DATASET + ENTERTAINMENT ARTICLES is a collection of articles focused on the entertainment industry. This dataset consists of articles related to various aspects of the entertainment sector, providing a comprehensive resource for understanding trends, developments, and key topics within this field. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + 89 + 2 + 6 + 260 + c885166d0c454a748376b56279f96408 + + + DOMAIN + The **ENTERTAINMENT INDUSTRY** is a multifaceted sector that encompasses various forms of entertainment, including movies, music, television, sports, and digital media. This industry is characterized by its diverse range of content and mediums, which collectively contribute to its broad appeal and significant cultural impact. The entertainment industry plays a crucial role in shaping public opinion, trends, and cultural norms through its extensive reach and influence across different platforms and genres. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + 92 + 2 + 19 + 261 + 586bccefb1e344289c1ee984e165de9c + + + METRIC + A metric indicating the highest level of development or achievement in a particular field + 322e02986c8724eedbcf3ebfa20b989c + 1 + 262 + a2201b8753ba4847ab0b22054e27d2c0 + + + METRIC + A metric indicating results that are comparable to or better than those of others in the same field + 322e02986c8724eedbcf3ebfa20b989c + 1 + 263 + b5ecd0553dd742f5813c9b855d548a41 + + + METRIC + A metric based on evaluations made by humans + 322e02986c8724eedbcf3ebfa20b989c + 1 + 264 + 89b2003e97804961805ea1886d078ebd + + + METRIC + Metrics that require a gold standard or reference answers for evaluation + 322e02986c8724eedbcf3ebfa20b989c + 2 + 265 + 6dd7f5f6b4544271a97f6a136f82fc3d + + + METHOD + An evaluation method that does not require reference answers + 322e02986c8724eedbcf3ebfa20b989c + 1 + 266 + eb01db8435554f2cbafe39a50f62f20a + + + METRIC + A metric that measures how relevant the generated text is to the given context + 322e02986c8724eedbcf3ebfa20b989c + 1 + 267 + 3d175ad1f0014cd4871eff4e86db9f88 + + + METRIC + A metric that measures how accurately the generated text reflects the source information + 322e02986c8724eedbcf3ebfa20b989c + 1 + 268 + c8e706fbdc90420d952deed03c4f04b4 + + + METRIC + A metric that measures how relevant the generated answer is to the question + 322e02986c8724eedbcf3ebfa20b989c + 1 + 269 + cf6115e69d6649cc99ef2bd11854ccfb + + + METHOD + A method involving multiple stages or steps + 322e02986c8724eedbcf3ebfa20b989c + 87 + 2 + 1 + 270 + 9ed7e3d187b94ab0a90830b17d66615e + + + DATA + The correct or ideal answers used as a benchmark in evaluations + 322e02986c8724eedbcf3ebfa20b989c + 2 + 271 + b4c7432f712849d7aba9dccbb77471ef + + + DATA + "SENSEMAKING QUESTIONS" are a class of questions used to evaluate the performance of Retrieval-Augmented Generation (RAG) systems. These questions are specifically designed to help users understand and make sense of complex information, as well as to validate the understanding and interpretation of data. By employing sensemaking questions, researchers and practitioners can assess how effectively a RAG system can retrieve and generate relevant information, thereby ensuring that the system aids in the comprehension and accurate interpretation of intricate datasets. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + 3 + 272 + 434e752b992c4e6a812557529315c5b9 + + + METHOD + A method where two items are directly compared against each other + 322e02986c8724eedbcf3ebfa20b989c + 1 + 273 + df79a27b9a4f42fd839c90bb8a79ad91 + + + DATA + TARGET METRICS are specific measures used to evaluate the performance of RAG systems. These metrics are aimed to be achieved or measured in the analysis and are the focus of an evaluation. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + 3 + 274 + 8f140fd7126f47b6b00307b0181509f9 + + + DATA + A metric used as a baseline or standard for comparison + 322e02986c8724eedbcf3ebfa20b989c + 2 + 275 + 40450f2c91944a81944621b94f190b49 + + + METRIC + A metric that measures the accuracy and reliability of a method or result + 322e02986c8724eedbcf3ebfa20b989c + 1 + 276 + 5b9fa6a959294dc29c8420b2d7d3096f + + + METRIC + A metric that measures the randomness or variability in a process + 322e02986c8724eedbcf3ebfa20b989c + 1 + 277 + b84d71ed9c3b45819eb3205fd28e13a0 + + + DATA + The average scores obtained from multiple evaluations + 322e02986c8724eedbcf3ebfa20b989c + 1 + 278 + b0b464bc92a541e48547fe9738378dab + + + PERSON + Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 89 + 2 + 6 + 279 + 44c65dda6fb7472dae36f6eea720ab47 + + + PERSON + Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 90 + 2 + 6 + 280 + 5d97ff82691c4482973d73d1860e4757 + + + PERSON + Britney Spears is a public figure frequently mentioned in entertainment articles, known for her significant contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 89 + 2 + 6 + 281 + 2567445079794d1e84f17abc48776002 + + + PERSON + Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his significant contributions to the music industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 89 + 2 + 6 + 282 + 392be891f8b649fabdc20e7bf549f669 + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in film and television + e8c8f911135faf3ff35f24107eb3f99c + 92 + 2 + 1 + 283 + 0111777c4e9e4260ab2e5ddea7cbcf58 + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in music + e8c8f911135faf3ff35f24107eb3f99c + 92 + 2 + 1 + 284 + 785f7f32471c439e89601ab81c828d1d + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in sports + e8c8f911135faf3ff35f24107eb3f99c + 92 + 2 + 1 + 285 + 6768339b54084020aec27adcef8994ff + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in digital media and business + e8c8f911135faf3ff35f24107eb3f99c + 92 + 2 + 1 + 286 + f09f381c319f4251847d1a4bb8cdcac1 + + + CATEGORY + A category of public figures in the entertainment industry who are involved in controversies + e8c8f911135faf3ff35f24107eb3f99c + 92 + 2 + 1 + 287 + eec11f567e7f4943b157c3a657eb9a46 + + + METRIC + A metric used to determine the winner in the comparison of generated responses + e8c8f911135faf3ff35f24107eb3f99c + 91 + 2 + 3 + 288 + efef117839b64ce9adf614a461d41ba6 + + + METRIC + A metric used to evaluate the quality of LLM-generated responses + e8c8f911135faf3ff35f24107eb3f99c + 77 + 2 + 1 + 289 + 2171091ada0942d8ae7944df11659f6e + + + SECTOR + The entity "FILM" refers to a sector within the entertainment industry that encompasses movies and cinema. This sector includes public figures involved in the movie industry, such as actors, directors, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 93 + 2 + 2 + 290 + bcfdc48e5f044e1d84c5d217c1992d4b + + + SECTOR + The entity "TELEVISION" refers to a sector within the entertainment industry that encompasses TV shows and series. This sector includes public figures involved in TV shows, such as actors, hosts, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 93 + 2 + 2 + 291 + b232fb0f2ac14790b931d1e7fcddd8ad + + + SECTOR + MUSIC is a sector within the entertainment industry that encompasses musical performances and recordings. This sector includes public figures involved in the music industry, such as singers, musicians, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 89 + 2 + 6 + 292 + 1c16b22e18d3483b8d41b284754274e2 + + + SECTOR + The entity "SPORTS" refers to a sector within the entertainment industry that encompasses athletic events and competitions. This sector includes public figures involved in sports, such as athletes, coaches, and sports commentators. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 90 + 2 + 4 + 293 + 0080f96708cd4054a5f0986ca86889f4 + + + SECTOR + DIGITAL MEDIA is a sector within the entertainment industry that encompasses online content and social media. This sector includes public figures involved in online platforms, such as influencers, content creators, and digital marketers. These individuals play a significant role in shaping digital landscapes through their engagement with audiences and their ability to leverage various online tools and platforms for content dissemination and marketing purposes. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + 93 + 2 + 2 + 294 + e683130322ac47708a852a5e51abb7c5 + + + CATEGORY + A category within the entertainment industry that includes stories and themes that shape culture + e8c8f911135faf3ff35f24107eb3f99c + 92 + 2 + 1 + 295 + 71a0a8c1beb64da08124205e9a803d98 + + + CATEGORY + A category within the entertainment industry that includes popular movements and styles + e8c8f911135faf3ff35f24107eb3f99c + 92 + 2 + 1 + 296 + f84314943bee4c859c9a62f268c9c216 + + + CATEGORY + A category within the entertainment industry that includes public conversations and debates + e8c8f911135faf3ff35f24107eb3f99c + 92 + 2 + 1 + 297 + ba481175ee1d4329bf07757a30abd3a1 + + + CATEGORY + A category within the entertainment industry that includes formal discussions and communications + e8c8f911135faf3ff35f24107eb3f99c + 92 + 2 + 1 + 298 + 8d8da35190bf43c5878fa38f3eb4f3d2 + + + RESPONSE + Answer 1 provides a varied and rich response by covering a wide range of public figures from different sectors of the entertainment industry, including film, television, music, sports, gaming, and digital media. It offers insights into the contributions and influence of these figures, as well as controversies and their impact on public discourse. The answer also cites specific data sources for each mentioned figure, indicating a diverse range of evidence to support the claims. + 718017a4871c909420f84b85b8ba969d + 93 + 2 + 11 + 299 + 2fb7e14a3f124526bd7b24867fc18e81 + + + RESPONSE + "ANSWER 2" is a generated answer for the example question in the News article dataset. It focuses on a smaller group of public figures, primarily from the music industry and sports, and relies heavily on a single source for data. "ANSWER 2" provides concise explanations for the frequent mentions of specific public figures such as Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + 89 + 2 + 10 + 300 + 5c13c7d61e6c4bfe839f21e7ad3530a7 + + + METHOD + Naïve RAG is a baseline method for retrieval-augmented generation (RAG) that does not use advanced techniques. It is a basic form of RAG with certain drawbacks that advanced RAG systems aim to overcome. Naïve RAG is used to generate answers for questions in the News article dataset and to generate responses that directly list specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d,e4d9b12cf2b4c691c74019eefff4fb39,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19 + 4 + 301 + a621663edba64d99b7e50f1e53f32ee7 + + + DATASET + The "NEWS ARTICLE DATASET" is a collection of news articles utilized for various analytical purposes. This dataset is specifically employed for generating responses to questions about public figures in the entertainment industry, making it a valuable resource for both analysis and information retrieval tasks within this domain. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + 16 + 302 + 42be4e140061482ea509dd3e26189480 + + + TOPIC + Controversies are events or issues involving public figures that generate public debate and impact public discourse. + 718017a4871c909420f84b85b8ba969d + 94 + 2 + 2 + 303 + 4da4ef951ff340f1a3dd679de4be3341 + + + SECTOR + The gaming sector includes public figures involved in the gaming industry, including gamers, developers, and streamers. + 718017a4871c909420f84b85b8ba969d + 93 + 2 + 1 + 304 + 2f05fcce857e4a499ca4e89a3cefbcb3 + + + RESOURCE + Data sources are references or reports used to support claims about public figures and their influence. + 718017a4871c909420f84b85b8ba969d + 93 + 2 + 2 + 305 + b3aeb7ae009a4f52ae3ae4586e32fe11 + + + METHOD + Assessments generated by large language models (LLMs) to evaluate the answers produced by different methods + ebf5249c888e07fedce6572a4c03f88c + 1 + 306 + 089b9b9841714b8da043777e2cda3767 + + + DATASET + An example question used in the News article dataset for analysis + ebf5249c888e07fedce6572a4c03f88c + 1 + 307 + 38f1e44579d0437dac1203c34678d3c3 + + + DATA + The datasets used in the analysis, consisting of various text sources + 4c855404ee3d3c94aa2136f1513c666f + 2 + 308 + 1ca24718a96b47f3a8855550506c4b41 + + + METRIC + A metric used to compare the performance of different conditions in the analysis + 4c855404ee3d3c94aa2136f1513c666f + 1 + 309 + 9c980dfe3cab44b7a83408405edab0b6 + + + CATEGORY + A specific setup or scenario used in the analysis, such as C0, C1, C2, C3, TS, and SS + 4c855404ee3d3c94aa2136f1513c666f + 4 + 310 + f23484b1b45d44c3b7847e1906dddd37 + + + METRIC + WIN RATE is a measure used to evaluate the success rate of different methods in providing comprehensive and diverse answers. It represents the percentage of times a particular approach or method achieves a win in a given context. Specifically, it quantifies the percentage of times a condition outperformed another in the analysis. This metric is crucial in assessing the effectiveness of various strategies within the domain of Natural Language Processing and Information Retrieval, offering insights into the comparative performance of different techniques. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4 + 3 + 311 + 929f30875e1744b49e7b416eaf5a790c + + + METRIC + The condition that performed the best across all comparisons in the analysis + 4c855404ee3d3c94aa2136f1513c666f + 1 + 312 + 4920fda031804ce8a1073ace8e061ed6 + + + METRIC + The expected win rate of a condition when compared to itself, shown as 50% for reference + 4c855404ee3d3c94aa2136f1513c666f + 1 + 313 + 4b8aa4587c7344adac2cbfa69d5e40fa + + + METHOD + The use of large language models (LLMs) at the time of querying, evaluated in the analysis + 4c855404ee3d3c94aa2136f1513c666f + 1 + 314 + 52701d941dfb45359693baae8f267056 + + + METHOD + The "FINAL EVALUATION" is the last stage of the analysis where the best performing context window size was used. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + 3 + 315 + 31499ee6277a4d71b19cb5b6be554c69 + + + PROCESS + The process that resulted in the creation of graphs for the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + 75 + 2 + 1 + 316 + d99eabad5dfd47278692569d2a9395b1 + + + STRUCTURE + A data structure consisting of nodes and edges, used to represent the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + 75 + 2 + 4 + 317 + d53f15cb7f7845de91cc44ad44ff9f6e + + + METHOD + Approaches that consistently outperformed the naive RAG in comprehensiveness and diversity metrics + 36db32c37e1987e2c5863898ad882190 + 91 + 2 + 3 + 318 + 23becf8c6fca4f47a53ec4883d4bf63f + + + METRIC + The number of context units, such as community summaries or text chunks, used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + 71 + 2 + 4 + 319 + d0ffa3bcd1234258953ff4956d19f561 + + + METRIC + The term "TOKENS" refers to the number of individual words used in the analysis. The evaluation typically focuses on corpora in the region of 1 million tokens. This metric is crucial for understanding the scope and scale of the text data being analyzed, particularly in the fields of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,92e93fc6449756c0a60200636b297f65 + METRIC + 76 + 2 + 7 + 320 + ac41b77ba33c4c84877eb425aba03aa1 + + + METRIC + The percentage of the maximum token count used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + 74 + 2 + 4 + 321 + 5d3184dabfd647a5a7e565f72c60ff24 + + + METHOD + MAP-REDUCE SUMMARIZATION is a method for summarizing source texts using a map-reduce approach. This summarization technique is notably resource-intensive, necessitating the highest number of context tokens compared to other methods. The map-reduce framework, originally popularized for its efficiency in processing large-scale data, is adapted here to handle the complexities of text summarization, ensuring comprehensive and accurate extraction of key information from extensive source texts. + 36db32c37e1987e2c5863898ad882190,e4d9b12cf2b4c691c74019eefff4fb39 + 76 + 2 + 2 + 322 + 0ec262c2cfef4dd581f3655e5e496e31 + + + DATA + Summaries at the root level of the community hierarchy, requiring dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + 76 + 2 + 1 + 323 + 100c2fccd7f74d9281707082f062ba72 + + + DATASET + SOURCE TEXTS are the original texts from which summaries or analyses are derived. These texts serve as the foundational material used for comparison with community summaries in the analysis. + 6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 324 + 378fc7636eeb4aabbfd40995a6960c64 + + + REFERENCE + A reference to a paper by Ram et al. in 2023 discussing RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + 1 + 325 + 80a04aa18cd649d584292f23b10c0727 + + + REFERENCE + "GAO ET AL., 2023" is a paper published in 2023 by Gao et al. that delves into advanced Retrieval-Augmented Generation (RAG) techniques, specifically where the index is a knowledge graph. The publication also touches upon naive RAG approaches, providing a comprehensive examination of both advanced and basic methodologies within the domain of Natural Language Processing and Information Retrieval. + 6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + 91 + 2 + 3 + 326 + 4e9ca18ccc1d4527a3bc035d07f5e162 + + + CATEGORY + Intermediate-level summaries are a type of community summary used in the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 74 + 2 + 1 + 327 + 5564257e89f1428486a64fcf52f49490 + + + CATEGORY + Root-level summaries are a type of community summary used in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 87 + 2 + 1 + 328 + 83c76fbd2a004d90a5b0a6736ffed61d + + + METRIC + Answer comprehensiveness is a measure used to evaluate the completeness of answers provided by different methods + 6f33a085ff3304e5994f7fbb86c881a4 + 87 + 2 + 1 + 329 + d9779c41e3c74fe0b26e23822a4b995b + + + TECHNOLOGY + Ad-hoc LLM use refers to the spontaneous use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + 77 + 2 + 1 + 330 + 9d7a563b3b2d405092c31f1fe08cff77 + + + TECHNOLOGY + Element extraction prompts are used to extract specific details in the Graph RAG index + 6f33a085ff3304e5994f7fbb86c881a4 + 87 + 2 + 1 + 331 + bd43f3d439a54781bd4b721a9a269b92 + + + CONCEPT, TECHNOLOGY + A mathematical space in which text chunks and queries are embedded to represent similar semantics + f35de4d9fb65f1d5a392064b20545c19 + 1 + 332 + adc0f95733e74351a891c4dadf650a52 + + + CONCEPT, DATA + Search inputs that are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + 1 + 333 + 225105a7be14447cb03186bd40756059 + + + TECHNOLOGY, METHOD + A type of RAG system that includes patterns for iterative and dynamic cycles of interleaved retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 1 + 334 + efce8a9d61254447a26aee99e53f0398 + + + TECHNOLOGY, CONCEPT + A concept related to generation-augmented retrieval that facilitates future generation cycles + f35de4d9fb65f1d5a392064b20545c19 + 2 + 335 + 4a75a9f0b18a48bea9c0601c0fc395c4 + + + TECHNOLOGY, METHOD + A method that facilitates future generation cycles by using self-memory + f35de4d9fb65f1d5a392064b20545c19 + 1 + 336 + e19287afe00a431f9a593a4827d1b448 + + + TECHNOLOGY, METHOD + A strategy for iterative retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 87 + 2 + 1 + 337 + f2c06f3a0c704296bf3353b91ee8af47 + + + TECHNOLOGY, METHOD + A federated strategy for retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 87 + 2 + 1 + 338 + f512103ed4624accac6cbbf90d7d250a + + + TECHNOLOGY, METHOD + A method that combines multiple concepts for summarizing multiple documents + f35de4d9fb65f1d5a392064b20545c19 + 2 + 339 + 2325dafe50d1435cbee8ebcaa69688df + + + TECHNOLOGY, METHOD + A method for answering questions that require multiple steps or "hops" to gather information + f35de4d9fb65f1d5a392064b20545c19 + 4 + 340 + 469aeef98cd1421fa123277b93d7b83a + + + TECHNOLOGY, METHOD + An approach that involves generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 2 + 341 + 2fb66f9a0de6406d83b61742a3b52cd6 + + + TECHNOLOGY, METHOD + A method for generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 1 + 342 + b0e6cfd979ea48b997019b059999d3c2 + + + TECHNOLOGY, METHOD + A method for answering multiple interpretations of ambiguous questions by generating a hierarchical structure + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 87 + 2 + 1 + 343 + ef00ec3a324f4f5986141401002af3f6 + + + TECHNOLOGY, METHOD + A process that involves using LLMs to create knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 77 + 2 + 1 + 344 + a542fd7aed7341468028928937ea2983 + + + TECHNOLOGY, METHOD + A process that involves using LLMs to complete existing knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 77 + 2 + 1 + 345 + 1c5e296a5ac541c1b5cac4357537c22d + + + TECHNOLOGY, METHOD + Graphs that represent causal relationships, which can be extracted using LLMs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 77 + 2 + 1 + 346 + 5ecf534a9ffe46e0b1c2144110c691c0 + + + REFERENCE, PUBLICATION + A reference to a publication by Cheng et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 87 + 2 + 1 + 347 + 4d183e7007624fcd98af96b9d752c16d + + + REFERENCE, PUBLICATION + A reference to a publication by Mao et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 87 + 2 + 1 + 348 + 718c507cb8ac49e6a35c251ac951b5ca + + + REFERENCE, PUBLICATION + A reference to a publication by Shao et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 87 + 2 + 1 + 349 + b45ef27279c043269b23b894461d7d8c + + + REFERENCE, PUBLICATION + A reference to a publication by Wang et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 87 + 2 + 1 + 350 + 10983a248cc448c59c94df4d1d0898f0 + + + REFERENCE, PUBLICATION + A reference to a publication by Su et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 87 + 2 + 1 + 351 + e2ec7d3cdbeb4dd086ae6eb399332363 + + + REFERENCE, PUBLICATION + A reference to a publication by Feng et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 87 + 2 + 1 + 352 + 67f10971666240ea930f3b875aabdc1a + + + REFERENCE, PUBLICATION + A reference to a publication by Trivedi et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 87 + 2 + 1 + 353 + 8b95083939ad4771b57a97c2d5805f36 + + + REFERENCE, PUBLICATION + A reference to a publication by Khattab et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 87 + 2 + 1 + 354 + 3c4062de44d64870a3cc5913d5769244 + + + REFERENCE, PUBLICATION + A reference to a publication by Sarthi et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 87 + 2 + 1 + 355 + 24652fab20d84381b112b8491de2887e + + + REFERENCE, PUBLICATION + A reference to a publication by Kim et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 87 + 2 + 1 + 356 + d4602d4a27b34358baa86814a3836d68 + + + REFERENCE, PUBLICATION + "TRAJANOSKA ET AL., 2023" refers to a paper by Trajanoska et al. published in 2023, which focuses on using Large Language Models (LLMs) for knowledge graph creation. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting innovative methodologies for leveraging advanced language models to construct and enhance knowledge graphs. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 77 + 2 + 2 + 357 + 36be44627ece444284f9e759b8cd25c6 + + + REFERENCE, PUBLICATION + "Yao et al., 2023" refers to a paper published by Yao and colleagues in 2023. The study focuses on the application of large language models (LLMs) for the task of knowledge graph completion. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting the potential of advanced LLMs to enhance the accuracy and efficiency of knowledge graph completion processes. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 77 + 2 + 2 + 358 + a64b4b17b07a44e4b1ac33580d811936 + + + REFERENCE, PUBLICATION + A reference to a publication by Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + 77 + 2 + 1 + 359 + 423b72bbd56f4caa98f3328202c1c3c9 + + + TECHNOLOGY, METHOD + A system that combines multiple concepts for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 1 + 360 + 5c7ef01f46a94641bf1ae5cd25f8a538 + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 1 + 361 + aefde1f7617f4c0e9aed31db77f6d862 + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 1 + 362 + ad52ba79a84748a49067e53b1d5095f9 + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + 1 + 363 + 289616058bf4495887292003b27ba216 + + + TECHNOLOGY, METHOD + Strategies used before the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + 0 + 364 + 7ffa3a064bce468082739c5a164df5a3 + + + TECHNOLOGY, METHOD + Strategies used during the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + 0 + 365 + ce36d1d637cf4a4e93f5e37ffbc6bd76 + + + TECHNOLOGY, METHOD + Strategies used after the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + 0 + 366 + eeb9c02c0efa4131b9e95d33c31019fc + + + TECHNOLOGY, METHOD + A pattern in Modular RAG systems for iterative and dynamic cycles of retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + 0 + 367 + 7b2472c5dd9949c58828413387b94659 + + + TECHNOLOGY, METHOD + Cycles of generation that are facilitated by self-memory in Graph RAG + f35de4d9fb65f1d5a392064b20545c19 + 0 + 368 + bdddcb17ba6c408599dd395ce64f960a + + + PUBLICATION + A paper by Ban et al. published in 2023, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 1 + 369 + bc70fee2061541148833d19e86f225b3 + + + PUBLICATION + A paper by Zhang et al. published in 2024, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 1 + 370 + 0fc15cc3b44c4142a770feb4c037a6f7 + + + METHOD + A method where the index is a knowledge graph, developed by Baek et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 2 + 371 + a24e9df02e1b4b43bf6324b039e28285 + + + PUBLICATION + A paper by Baek et al. published in 2023, focusing on the KAPING method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 1 + 372 + ab3a5a6713244fd595a1ace978c3d960 + + + METHOD + A method where subsets of the graph structure are the objects of enquiry, developed by He et al. in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + 2 + 373 + 02a88c0d128e4586b2f1f64329786d3c + + + PUBLICATION + A paper by He et al. published in 2024, focusing on the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 1 + 374 + 1ca41537c47c4752a17a44d1d7086d96 + + + METHOD + A method where derived graph metrics are the objects of enquiry, developed by Zhang in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 2 + 375 + 7e0d14ca308b4796bdc675a64bd3a36e + + + PUBLICATION + A paper by Zhang published in 2023, focusing on the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 1 + 376 + 8323efc8e539419e9ca3c98e758f6609 + + + METHOD + A method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, developed by Kang et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 2 + 377 + a80c7c98c0b647f8b9f6f8cc09168e44 + + + PUBLICATION + A paper by Kang et al. published in 2023, focusing on the SURGE method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 1 + 378 + 2d66a15939294d21b83b3e277f0a4e46 + + + METHOD + A method where retrieved event-plot subgraphs are serialized using narrative templates, developed by Ranade and Joshi in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 2 + 379 + 47f6d6573cf34e1096c95e36251dd60c + + + PUBLICATION + A paper by Ranade and Joshi published in 2023, focusing on the FABULA method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 1 + 380 + 2fbd74d5ccca4be99c5257b3ac95cfba + + + PUBLICATION + A paper by Wang et al. published in 2023, focusing on a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + 1 + 381 + a2b1621a3e424ae29a6a73f00edbeca3 + + + ORGANIZATION + LangChain is an organization that developed Langchain graphs and supports a variety of graph databases. + 71f6daf11e64e5273a3847d46bf228e1,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + 5 + 382 + ec45e1c400654c4f875046926486ded7 + + + ORGANIZATION + LlamaIndex is an organization that developed the LlamaIndex Knowledge Graph Index and supports a variety of graph databases. + 6cd82819982879bd164547d2773ba5c7,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + 3 + 383 + 047cd93e9d704c7d8dadb6e79f9458df + + + TECHNOLOGY + Neo4J is both a graph database format supported by various Retrieval-Augmented Generation (RAG) applications and an organization that developed Project NaLLM. The graph database format of Neo4J is widely recognized for its efficiency in handling complex relationships and structures, making it a valuable tool in the field of Natural Language Processing and Information Retrieval. As an organization, Neo4J has contributed significantly to the advancement of these domains through innovative projects like NaLLM, which further underscores its pivotal role in the community. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + 4 + 384 + 5b71ee73a5b6484495b2a0a75219426c + + + METHOD + A method that can create and reason over knowledge graphs in Neo4J format, developed by Neo4J in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + 1 + 385 + e1f524d4b9754ce2b64a0a4c8f73b854 + + + TECHNOLOGY + NebulaGraph is both a graph database format and an organization that has made significant contributions to the field of graph databases and retrieval-augmented generation (RAG) applications. As a graph database format, NebulaGraph is supported by various RAG applications, facilitating the efficient handling and querying of complex graph data structures. Additionally, NebulaGraph, as an organization, has pioneered the industry-first graph RAG, which integrates retrieval-augmented generation with large language models (LLMs) based on knowledge graphs. This innovation underscores NebulaGraph's role in advancing the capabilities of knowledge graph-based applications and enhancing the performance of LLMs in generating contextually relevant information. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + 4 + 386 + ae1fe1c014c54ec4bcdf10dbdaed5068 + + + METHOD + A method that can create and reason over knowledge graphs in NebulaGraph format, developed by NebulaGraph in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + 1 + 387 + 92646910ee624bd7909fac2b5c0232e3 + + + METHOD + A method for comparing fabrication rates, developed by Manakul et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + 1 + 388 + 05913bee89a94bca88449249e35ba74d + + + PUBLICATION + "MANAKUL ET AL., 2023" refers to a paper by Manakul et al. published in 2023, which focuses on the SelfCheckGPT method. This work by Manakul and colleagues is centered around the development and application of SelfCheckGPT, a technique likely aimed at enhancing the performance and reliability of GPT models. The paper contributes to the field of Natural Language Processing and Information Retrieval by addressing specific challenges and proposing innovative solutions through the SelfCheckGPT method. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + PUBLICATION + 2 + 389 + 57b8930790c34dcba4a32c6be703ed78 + + + STAKEHOLDER + END USERS are individuals who are the final users of the system or analysis. They play a crucial role in validating sensemaking questions and target metrics, ensuring that the system or analysis meets the intended objectives and provides meaningful insights. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + STAKEHOLDER + 2 + 390 + 838c4498bc3c437f8d65428b580766a2 + + + CONCEPT + Considerations and compromises involved in building a graph index + 92e93fc6449756c0a60200636b297f65 + CONCEPT + 1 + 391 + 1b893f24eb98477aad6ce49c0f26737e + + + METRIC + The effectiveness of RAG systems, which varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + METRIC + 1 + 392 + 6573bc2af4f94596a3f4452a602d6fc4 + + + CONCEPT + Various forms of data used in RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + 1 + 393 + 0dddcca0e5df4b16bc03a51a2d2d8e16 + + + METRIC + The scale of datasets used in RAG systems, which affects performance + 92e93fc6449756c0a60200636b297f65 + METRIC + 1 + 394 + df40ad480a3c47299a6c8fad05349304 + + + PROCESS + The process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + PROCESS + 1 + 395 + fe98fb197d294b0b837aee8d5a98dfb1 + + + DATASET + Collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + DATASET + 1 + 396 + feb9ddd0ac2949178f26a36949aa5422 + + + CONCEPT + Different categories of questions used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + 1 + 397 + b4e4fa2e3dfc46e68d532d659b18d17d + + + METHOD + SelfCheckGPT is an approach used to compare fabrication rates in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 398 + f58813d090b947a48c1b4614b92c3ec3 + + + METHOD + A method for global summarization of source texts that does not use a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 399 + 30a251bc3d04430d82b5a1a98c7b8c75 + + + RESOURCE + The amount of computational resources allocated for a task + e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 400 + 93e1d19f9bfa4c6b8962d56d10ea9483 + + + METRIC + The expected number of queries over the lifetime of a dataset + e4d9b12cf2b4c691c74019eefff4fb39 + 0 + 401 + 8046335ba70b434aa3188392a746fd78 + + + DATA + Annotations that provide detailed information about the text + e4d9b12cf2b4c691c74019eefff4fb39 + 87 + 2 + 1 + 402 + 5c02b1ab32064c64a0f8b27b219e358a + + + METHOD + A method that uses embeddings to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + 3 + 403 + c5f77ba0c261408780db3d50346f16b7 + + + METHOD + RAG schemes that combine embedding-based matching with other approaches + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 404 + 453ecf5476f64f4a8d5020b95baf1314 + + + METHOD + Mechanisms used in map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 405 + 6a1d83c9ce2b483dbd7de5ab3ae2487d + + + DATA + A hierarchical organization of communities + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 406 + 66c3dffb7d7a4fa8bb6b48a22ca917a6 + + + METHOD + A method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) for global text summarization + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 407 + 6f3dd1fd6d7f4df4af0656ed0525c92e + + + METRIC + The cost associated with the number of tokens used in a text generation task + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 408 + 711eb39432794b0a91110358dd536517 + + + TECHNOLOGY + An implementation of Graph RAG approaches using the Python programming language + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 409 + 0e00585b08044954a254116665400463 + + + PERSON + A person who contributed to the work mentioned in the acknowledgements + e4d9b12cf2b4c691c74019eefff4fb39 + 87 + 2 + 1 + 410 + db0147eff2204a20b5e5e6bec7a8bae5 + + + METRIC + The rates at which fabrications occur in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 411 + 67bb4f4678284819add02ba04f3b1103 + + + METRIC + The expected number of queries over the lifetime of a specific dataset + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 412 + 2033ec0487f04240abb3bdbe77b39087 + + + METRIC + The benefits or value obtained from using a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 413 + f026fab8fec948ae9e7baa2ad715e6ef + + + METHOD + Different methods related to retrieval-augmented generation that utilize graph structures + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 414 + d0d7ed36d6f54b5d986dfd854096b728 + + + METHOD + Graph RAG approaches that operate in a more localized manner + e4d9b12cf2b4c691c74019eefff4fb39 + 87 + 2 + 1 + 415 + bf6a4c18f44042799eb7456a6b85b54a + + + DATA + Annotations made on the graph to provide additional information + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 416 + fac4a59c2278498d83f9f1b4231ad62e + + + DATA + Reports generated from community summaries + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 417 + d6d2b5862ddc4c4d87deee3423506817 + + + METHOD + An operation that aggregates information across multiple levels of a hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 418 + 47d588d26e2b4cccb68fe2af4c147c8f + + + METHOD + A mechanism that allows for exploring detailed information by following higher-level summaries + e4d9b12cf2b4c691c74019eefff4fb39 + 2 + 419 + c0f2dc03d8df400db4997c1a0babd6ad + + + DATA + The trail of information that guides users to more detailed data + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 420 + 0211d61aae834229a3a1e004ff5cc658 + + + DATA + The top-level communities in a hierarchical structure + e4d9b12cf2b4c691c74019eefff4fb39 + 99 + 2 + 1 + 421 + ccbbbcc055c34709abcf103208c2c299 + + + DATA + A graph index organized around entities + e4d9b12cf2b4c691c74019eefff4fb39 + 87 + 2 + 1 + 422 + 989add81cf874018a569239b68d17ff2 + + + TECHNOLOGY + A publicly available implementation of a technology + e4d9b12cf2b4c691c74019eefff4fb39 + 1 + 423 + fd7d94fbab084bc380480abeef6bfade + + + PERSON + Alonso Guevara Fernández is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 21 + 424 + cfb915c95caf41c6a25e99a9f37f03a2 + + + PERSON + Amber Hoak is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 19 + 425 + 8815ed80f9b741dbb458d902024f34a4 + + + PERSON + Andrés Morales Esquivel is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 426 + dddb831546354e088d29aebd154e3a31 + + + PERSON + Ben Cutler is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 427 + 005d2154da754b21adcd90ac921bd5f7 + + + PERSON + Billie Rinaldi is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 428 + 711ba818354546cea69f1532b92a2f26 + + + PERSON + Chris Sanchez is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 429 + 5c4d8a8f9c104176b87d2bfdf04ae0bd + + + PERSON + Chris Trevino is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 430 + 5a781604f1fb4719b730f43f534627f6 + + + PERSON + Christine Caggiano is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 431 + ecdc1020b10e49ca869d399825e16fa3 + + + PERSON + David Tittsworth is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 432 + 0d8fde01d7234726a00d7e73e2e01d66 + + + PERSON + Dayenne de Souza is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 433 + 9c4bd60958fd4e09a6d5b9e2ab163b5a + + + PERSON + Douglas Orbaker is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 434 + 39d31f770cf740e78d526a2e1101a1db + + + PERSON + Ed Clark is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 435 + 9d282b2250f7408888504f1f93c202a8 + + + PERSON + Gabriel Nieves-Ponce is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 436 + c063484895794a0eaae1b0ff070ad4c9 + + + PERSON + Gaudy Blanco Meneses is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 437 + e8868920e21b4431aad16e86db977ecb + + + PERSON + Kate Lytvynets is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 438 + aea3378bfff842e5b3f4b7a4b55b3879 + + + PERSON + Katy Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 439 + d562223c17d948bf98e34b4d97dde932 + + + PERSON + Mónica Carvajal is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 440 + cde2d75c51d245879265b79d14b8699b + + + PERSON + Nathan Evans is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 441 + 44594467054849d4a1fadb46ddd51641 + + + PERSON + Richard Ortega is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 442 + 2918130221f94f4387da049b647bfe6a + + + PERSON + Rodrigo Racanicci is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2 + 443 + fd139ac75b0e4777ab67b7423eaaa37f + + + PERSON + Sarah Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1 + 444 + a701c349eb7142d48ba7efad89caf9d2 + + + PERSON + Shane Solomon is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1 + 445 + e5d40a1b17f74b1db5d18279caedb04a + + + PUBLICATION + A technical report on GPT-4 published as an arXiv preprint + 086021a89900a39bcb62036981737bfa + 0 + 446 + de25d06733d04385825ee082792f5e52 + + + METHOD + A method for zero-shot knowledge graph question answering described in an arXiv preprint + 086021a89900a39bcb62036981737bfa + 0 + 447 + 32f6f11a7845416b8c6eb9fb0b382140 + + + METHOD + A method for harnessing large language models for advanced causal discovery from data + 086021a89900a39bcb62036981737bfa + 0 + 448 + 91407be8c3e54e23918d3a7183d962db + + + METHOD + A method incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models + 086021a89900a39bcb62036981737bfa + 0 + 449 + 3831134696584d83bbf676a6b3bfa8f9 + + + PERSON + J. Achiam is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 450 + 50e512a5dbe941f5af68bfdf74b1c3c0 + + + PERSON + S. Adler is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 451 + edc717747e904728b57185f5013461f9 + + + PERSON + S. Agarwal is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 452 + 8fba1fea719d49d380ac2d9c310d68b3 + + + PERSON + L. Ahmad is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 453 + 532da08f04f645708e747c57e9c4ee05 + + + PERSON + I. Akkaya is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 454 + 3cf0ab4cf14e47ddabd49d500a3dc488 + + + PERSON + F. L. Aleman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 455 + a39b72f8921f43ef8ef295c7cc8f7294 + + + PERSON + D. Almeida is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 456 + 9f5adbeb6cf04f089abe78d86cfa6aba + + + PERSON + J. Altenschmidt is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 457 + efb6350e65964659bc20396c0166b296 + + + PERSON + S. Altman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 458 + e095cc36da784300b27c6f8c60a96440 + + + PERSON + S. Anadkat is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9 + 459 + c68893ca39d74ba08c6eb138f24441e1 + + + PERSON + R. Anil is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 460 + 472b23bb92834173b4118d101040c726 + + + PERSON + S. Borgeaud is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 461 + 81869985b45a4fefbbbb23ea118a3de4 + + + PERSON + Y. Wu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 462 + 42b8584c5a874eb08fbd61f0c18f3ca0 + + + PERSON + J.-B. Alayrac is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 463 + 824d93d9840a4b7c8b1f31bc6816b497 + + + PERSON + J. Yu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 464 + f209a808f1f04a5699601e672f4abd06 + + + PERSON + R. Soricut is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 465 + ccb335166f6c4564ac1c61549d8ded50 + + + PERSON + J. Schalkwyk is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 466 + cbe1a41a82aa4f268e8264568b25938f + + + PERSON + A. M. Dai is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 467 + 28e7639f55ce464c8a080cbb2c745fa2 + + + PERSON + A. Hauth is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + 8 + 468 + 3f3a2d7aa1294116814f0b4d89baa23d + + + PERSON + J. Baek is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 2 + 469 + 3073b33926bd4f33807ffa3befacefaf + + + PERSON + A. F. Aji is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 2 + 470 + 2b916117691c4872a9c4e4888d4fe4ab + + + PERSON + A. Saffari is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 2 + 471 + 1f7b02bf486e4f42b23e9cb1a63207f3 + + + PERSON + T. Ban is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 1 + 472 + e744c118ae7f4638a01d060bbaedd6e9 + + + PERSON + L. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 1 + 473 + e1c1080c717d437996def1a41772d179 + + + PERSON + X. Wang is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 0 + 474 + 63fba9a7c47a4f14ac0bee6bc90d0fea + + + PERSON + H. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + 0 + 475 + 6bfc2395b4f54a528a1ebac94a43acb8 + + + PERSON + T. Baumel is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 476 + 1cce5cebf437428eb1a60dffbdfa603f + + + PERSON + M. Eyal is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 477 + dc94039d6643460ca3c66150b9087129 + + + PERSON + M. Elhadad is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 478 + f197d75f159943f8a3ff441199790bc7 + + + PUBLICATION + The arXiv preprint identifier for the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0 + 479 + 4d8890c699684c9381105b03b0b41b03 + + + PUBLICATION + The arXiv preprint identifier for the Gemini paper + 086021a89900a39bcb62036981737bfa + 0 + 480 + b1658adfa43847eabad1437db235e858 + + + PUBLICATION + The arXiv preprint identifier for the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 0 + 481 + a1773cac7d4c4939aec965660e5015fe + + + PUBLICATION + The arXiv preprint identifier for the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + 0 + 482 + 6a054cb59fb44cf494b93988b5f88833 + + + PERSON + Baumel, T. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 483 + e7b103a52e384e3e8bf14105223e7e82 + + + PERSON + Eyal, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 484 + 3f1042452c254cecaf7189e89162adc8 + + + PERSON + Elhadad, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 485 + fd31d549420744d1bd1a6b1112a9a6ba + + + PERSON + Blondel, V. D. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 486 + f7ab348030714072a277682b51f7c588 + + + PERSON + Guillaume, J.-L. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 487 + 2139b0906dc541e094138a978d070416 + + + PERSON + Lambiotte, R. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 488 + ff5466607e5d4453b1d833629292f664 + + + PERSON + Lefebvre, E. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 489 + 71f95003936e46a98d90757ffd845d40 + + + PUBLICATION + The journal where the paper "Fast unfolding of communities in large networks" was published + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 490 + bada987ea7da4c939393ee1c3d08ccd4 + + + PERSON + Brown, T. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9 + 491 + d0a274e7934d446fb91847bb53a961a6 + + + PERSON + Mann, B. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9 + 492 + 0a799eab61bc4e6b884db6689f9c2c4a + + + PERSON + Ryder, N. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9 + 493 + 8c34cd494a63438dac219c1dc0f73100 + + + PERSON + Subbiah, M. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 8 + 494 + c6f428af0c5e4f629902fd5455bf19ac + + + PERSON + Kaplan, J. D. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 495 + d1fd271d16c348019c2fcced762b35a2 + + + PERSON + Dhariwal, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 496 + ffa128c9c0c84d39bad1bba8cfa4adc5 + + + PERSON + Neelakantan, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 497 + 058f66cc356b43cc9433bd3c8d57fa46 + + + PERSON + Shyam, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 498 + ff74091eaba246698fcae59c21eec828 + + + PERSON + Sastry, G. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 499 + f6cbbf1b8f4b48a28a16e4dd8976b9bb + + + PERSON + Askell, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 500 + 757ca40654d5476aa949a26b733be8d4 + + + PUBLICATION + "ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS" is a prominent conference where significant papers in the field of Natural Language Processing and Information Retrieval are presented. Notable papers presented at this conference include "Language models are few-shot learners" and "Retrieval-augmented generation for knowledge-intensive NLP tasks." Additionally, it is also the journal where the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" was published. This conference and journal serve as key platforms for disseminating cutting-edge research in neural information processing systems. + 58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,b1bbda43309e8e0e2175ea034aa88e13 + 0 + 501 + 539d55e7c42e44b59d98f59fae3e0ee1 + + + PERSON + Cheng, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 502 + 3785eeadea9042bfb2e50f16c0397a12 + + + PERSON + Luo, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 503 + 48cd97f2297143e09d61ff2a8542c0c5 + + + PERSON + Chen, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 504 + ff95eb0d5f7f49b782027d5c7ae3c3fe + + + PERSON + Liu, L. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 505 + 086da554db5b4ad5806aedeb0024197c + + + PERSON + Zhao, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory"Zhao, D. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + PERSON + 82 + 2 + 2 + 506 + 216ee8a907a0466a88b27f8ada19ffa0 + + + PERSON + Yan, R. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 507 + 6fefb317687d4ac98efe39a52f3e190f + + + PERSON + Dang, H. T. is an author of the paper "Duc 2005: Evaluation of question-focused summarization systems" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 508 + 320d9d91238948a8be67972ccceab878 + + + PUBLICATION + The conference where the paper "Duc 2005: Evaluation of question-focused summarization systems" was presented + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 509 + bdcbcccadd474b3bbe9a8f56c811bab4 + + + PERSON + Es, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 2 + 510 + f127fc4d87f94794be89134406ba0694 + + + PERSON + James, J. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 2 + 511 + c27966a4e3be434686454204ac7b3ab4 + + + PERSON + Espinosa-Anke, L. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 2 + 512 + dab39f92d0ed468c80699f28c05c45fa + + + PERSON + Schockaert, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 2 + 513 + 3076f330d121489aa50964ce54a3b1ac + + + PERSON + Feng, Z. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 82 + 2 + 2 + 514 + c8e5d3afdcb54c8589e280f0c4a87417 + + + PERSON + Feng, X. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 82 + 2 + 2 + 515 + f3d30627e19245649e497ab49bf0fa30 + + + PERSON + Yang, M. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 82 + 2 + 2 + 516 + e3f1098c3d984bc7b5f30b9c0101f7a6 + + + PERSON + Qin, B. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 82 + 2 + 2 + 517 + 24b4a5f4db67418cbfa08c5316f0ab51 + + + PERSON + Fortunato, S. is an author of the paper "Community detection in graphs" + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 518 + e4b707e3e6964197855b82fc66ef59e7 + + + PUBLICATION + The journal where the paper "Community detection in graphs" was published + 58ae80c41cfe46db39da26b6a83584e5 + 0 + 519 + 109b8be5a8ee4180a1465cd23f019d7b + + + PERSON + Gao, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 520 + 49f771e31a0c4b35bc39e389f3623509 + + + PERSON + Xiong, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models. The paper provides a comprehensive survey of the methodologies and applications of retrieval-augmented generation, highlighting its significance in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 521 + aa946d4379694a74ba0da37e69d2810a + + + PERSON + Gao, X. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 522 + 268446fc52a54fd2837f73aeb3e0b74f + + + PERSON + Jia, K. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant area of research within the domains of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 523 + f6ddfa8491ff40d2839bb5b2e105df22 + + + PERSON + Pan, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 524 + db1295504da645b69d9786d54f233fed + + + PERSON + Bi, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 525 + 6ff4ed0dda4f4158af37be99f505565f + + + PERSON + Dai, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance and capabilities of large language models, a significant area of research within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 526 + 5d398b88ee4242a59c32feb188683ec3 + + + PERSON + Sun, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 527 + 0a784e00c9464bd3aeb830b908f73170 + + + PERSON + Wang, H. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + 8 + 528 + b0966a0f455e44229e6c9705d57bfca9 + + + PUBLICATION + The arXiv identifier for the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 3 + 529 + 99761e9b89cc4060be3ed6b34532e7ff + + + PUBLICATION + The arXiv identifier for the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + 4 + 530 + 8130a1a82bde46048952cf147690e630 + + + PUBLICATION + The arXiv identifier for the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + 82 + 2 + 5 + 531 + 79c99026b7ef4946b9b8e0be841fd4c5 + + + PERSON + Goodwin, T. R. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 2 + 532 + fdcb1673254842f1935f53d0c38c467e + + + PERSON + Savery, M. E. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 1 + 533 + dcb3f4cc8abc46faabc193d9885e91d0 + + + PERSON + Demner-Fushman, D. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 1 + 534 + 3295be59128d451bb720c6688adc1e0b + + + CONFERENCE + COLING (International Conference on Computational Linguistics) is the conference where the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" was presented + 00e8e4e881bd0862022f4dfc913b900b + 0 + 535 + aca3eb8924ac494486fe0bfe892f7f2e + + + PERSON + He, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 536 + 66689accdd974295b7eb779e43578748 + + + PERSON + Tian, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 537 + 6b49c78aa1524609ab7aa74aeaa3e01d + + + PERSON + Sun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 538 + 7ff31ce54f424f0bbb297b0b3ba7c757 + + + PERSON + Chawla, N. V. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 539 + bac51e00d486420c8e91e824d8e17411 + + + PERSON + Laurent, T. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 540 + 4adee3aad6524a4aa4c4711c1ee05e64 + + + PERSON + LeCun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 541 + d034e4fd8ac849278e658daad1a1f033 + + + PERSON + Bresson, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 542 + 091e998370dd42d1b05ab0fcf6595a7e + + + PERSON + Hooi, B. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 543 + 1e6cabc18fab4c048281fd29d3044438 + + + PERSON + Jacomy, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 544 + dc08f6d7398b4b798a3bdccf508a2ad4 + + + PERSON + Venturini, T. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 545 + 1c7fd5af8d8041e186eae2431fc627cd + + + PERSON + Heymann, S. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 546 + b16eda56dcec40f2b3e109fb9246bee3 + + + PERSON + Bastian, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 547 + 43c68f9a86654a32a2215e23957ed184 + + + PUBLICATION + PLOS ONE is the journal where the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" was published + 00e8e4e881bd0862022f4dfc913b900b + 0 + 548 + 1ba06fe2e86140a59bbc4f4e969d0f71 + + + PERSON + Jin, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 549 + 36caa0a230c8422c8acb4dc62e35bb32 + + + PERSON + Yu, Z. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 550 + 09940fed9d154504948bba2df1789a50 + + + PERSON + Jiao, P. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 551 + 4d6608557eed49368a6d09c7c5c664c5 + + + PERSON + Pan, S. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 552 + eb7c93eeb9dc41aab57d29e97ebb4951 + + + PERSON + He, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 553 + 3b6e2ac584b64847b53828c9d779fed3 + + + PERSON + Wu, J. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 554 + e9b68002e035447baae848208cea5503 + + + PERSON + Philip, S. Y. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 555 + fe18353546824ca98294ce4be7b96e02 + + + PERSON + Zhang, W. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 556 + 0e9740e25f5a460c81318336e00ac880 + + + PUBLICATION + IEEE Transactions on Knowledge and Data Engineering is the journal where the paper "A survey of community detection approaches: From statistical modeling to deep learning" was published + 00e8e4e881bd0862022f4dfc913b900b + 0 + 557 + b7cd9a62710849778fdadced0d754687 + + + PERSON + Kang, M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 558 + 432a6b4962544200949421a96a405142 + + + PERSON + Kwak, J. M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 559 + d6700b360ac141d282cdb567414bf4ce + + + PERSON + Baek, J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 560 + c1b40a4039b44061a358e098867f7412 + + + PERSON + Hwang, S. J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + 0 + 561 + 4643a7a319674adfb732b6f6122c7c64 + + + PERSON + Khattab, O. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 562 + 46e8056fb2ec4811ab33cb34a0dc9fb3 + + + PERSON + Santhanam, K. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 563 + 8b57a9f43a1942a49b58cf881835f974 + + + PERSON + Li, X. L. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 564 + f78b01b0d93948c283644ec58f7be74a + + + PERSON + Hall, D. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text, indicating its relevance within the domain of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 565 + 8dbe8f9867e4448f998416c18923eac4 + + + PERSON + Liang, P. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Liang, P. contributed to the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP," which explores the integration of retrieval and language models to enhance knowledge-intensive tasks in NLP. Additionally, Liang, P. authored the paper "Lost in the middle: How language models use long contexts," which investigates the utilization of extended contexts by language models. These contributions highlight Liang, P.'s significant role in advancing the understanding and application of language models in complex NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 7 + 566 + fe8ea8bf1395434393e04e8f7a33025f + + + PERSON + Potts, C. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 567 + 7d58b089bfc549e8951e91ad62541119 + + + PERSON + Zaharia, M. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 7 + 568 + 1fa6d3118bd846c8837b5fa9fb78f262 + + + PERSON + Kim, G. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 569 + 62c65bbae33c4ee9a21b61f6f454c4b4 + + + PERSON + Kim, S. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 570 + 30b7034c4468473f98ee18d00ee73b33 + + + PERSON + Jeon, B. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 571 + 00f78b85e5b84999a810e311e540037b + + + PERSON + Park, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 572 + 3e460d9f011d4b0b9ccaae7b6a5202de + + + PERSON + Kang, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 5 + 573 + 9d98dece22eb401aa1a5ce9c88c603f0 + + + PERSON + Klein, G. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 574 + 81446ea789b24eaf9eab02dc07c3d984 + + + PERSON + Moon, B. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 575 + 79f4b1c1b2be4cf7aa828846e20a4eb6 + + + PERSON + Hoffman, R. R. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 576 + de04830d6e414fd5b39a9e90769d9452 + + + PUBLICATION + The journal where the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" were published + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 577 + 69db426b97714835bf4937180774787a + + + PERSON + Koesten, L. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 578 + 9c7bc862339d4a5bb21ee5154d9b33bb + + + PERSON + Gregory, K. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 579 + 17bad53a0ebe4569839e5e151ff78593 + + + PERSON + Groth, P. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 580 + 53d98f08e7c74158b7318357b6c660b3 + + + PERSON + Simperl, E. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 581 + cd601f77419c403889aadeee591915b5 + + + PUBLICATION + The journal where the paper "Talking datasets–understanding data sensemaking behaviours" was published + 71f6daf11e64e5273a3847d46bf228e1 + 4 + 582 + 0f564ebd53e940fba9d16674ac7bc038 + + + PERSON + Kuratov, Y. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 583 + 7deb75816e4f473480e0c79ae99b5bf4 + + + PERSON + Bulatov, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 584 + 7f85b181f1184f77aeb3ea2155cf4027 + + + PERSON + Anokhin, P. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 585 + d148b2b2033048618f1a090a492a40a5 + + + PERSON + Sorokin, D. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 586 + 4d839a10353e4144a26563b0966721d5 + + + PERSON + Sorokin, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 587 + 521a862bb196488389f17c0b0f4b6f4d + + + PERSON + Burtsev, M. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 588 + 22ea3328fb6343f4ad2862495ea27640 + + + TECHNOLOGY + Langchain graphs is a technology developed by LangChain + 71f6daf11e64e5273a3847d46bf228e1 + 1 + 589 + 3f9a2a2c1c0a424e8b4980ea9d48bdbe + + + PERSON + Laskar, M. T. R. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" and also contributed to the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models." Both works highlight Laskar's expertise in leveraging transformer models and transfer learning techniques to enhance the performance of query-focused abstractive text summarization, demonstrating a significant contribution to the field of Natural Language Processing and Information Retrieval. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 5 + 590 + aa2ec452728a4703ae1bdabe85b6c079 + + + PERSON + Hoque, E. is an author of two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning to improve the effectiveness of transformer models in query-focused abstractive summarization tasks. Both works contribute to advancing the understanding and application of transformer models in specialized summarization contexts. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 5 + 591 + c5ddb31e0a9c4b2683e4631283dd505b + + + PERSON + Huang, J. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 592 + 07d8eeb549044ac88d2e788c146a0ef1 + + + PUBLICATION + The conference where the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" was presented + 71f6daf11e64e5273a3847d46bf228e1 + 3 + 593 + 47df2815030c4f1c99facd5cf2482526 + + + PUBLICATION + arXiv preprint refers to a preprint of a paper that is available on the arXiv repository + 71f6daf11e64e5273a3847d46bf228e1 + 18 + 594 + ae521508bdc244f99c4fce4ab5214c79 + + + EVENT + The 33rd Canadian Conference on Artificial Intelligence, held in Ottawa, ON, Canada, from May 13–15, 2020 + 6cd82819982879bd164547d2773ba5c7 + 2 + 595 + 6315b4bf135c40358823ed7e4e4060e2 + + + EVENT + The 2020 edition of the Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + 1 + 596 + 33905debec1a45ecae1c65daac1d854c + + + PUBLISHER + Springer is the publisher of the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + 1 + 597 + bfbe904780fe47daad1a04126b12923c + + + PERSON + Huang, J. X. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + 3 + 598 + 0614f00e932c4cd0b53928053811ebc1 + + + PUBLICATION + The journal where the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" was published + 6cd82819982879bd164547d2773ba5c7 + 3 + 599 + 9ef487dd0b574b108c60a56d6a2f146c + + + PERSON + Lewis, P. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9 + 600 + 4067269e7f6943cdbc299ce02b7eadbd + + + PERSON + Perez, E. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9 + 601 + 094a736ba43c4da48c556437f47f88d1 + + + PERSON + Piktus, A. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9 + 602 + 563c2af32bb3476299e9b24a646097ab + + + PERSON + Petroni, F. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks"Petroni, F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + PERSON + 9 + 603 + d59b49eb94ce442d89907e90c5d3a44e + + + PERSON + Karpukhin, V. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 5 + 604 + 8ea7cef407df48098046551e303e1c64 + + + PERSON + Goyal, N. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 5 + 605 + 186e60d2176547bf84e5bf87bd16bb40 + + + PERSON + Küttler, H. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 606 + e65017091c8d4c7daa45b6c8414e0465 + + + PERSON + Lewis, M. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 607 + a0f326b9597b49dda6563e9208316117 + + + PERSON + Yih, W.-T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 608 + bff3db70f9af4f2c87a93df48ecbb6bc + + + PERSON + Rocktäschel, T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4 + 609 + bf91f36307cb43e1ab1e967cb3ba8274 + + + PERSON + Liu, N. F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 610 + cd58a8740ba54d86a77db9bb9544ef0d + + + PERSON + Lin, K. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 611 + e96d3475d43b42a781b297ae7e650afe + + + PERSON + Hewitt, J. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 612 + 1ce76a5547854d458878bd445f0ccbd6 + + + PERSON + Paranjape, A. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 613 + 11e4325f59394ff1bc89892f79288702 + + + PERSON + Bevilacqua, M. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + 0 + 614 + 71743537a07c440ea1710a269da8b538 + + + PERSON + Liu, Y. is an author of the paper "Hierarchical transformers for multi-document summarization" + 6cd82819982879bd164547d2773ba5c7 + 0 + 615 + 1389192ce5464be6b3b5749bc9536709 + + + PERSON + Lapata, M. is an author known for significant contributions to the field of Natural Language Processing and Information Retrieval. Notably, Lapata, M. has authored the paper "Hierarchical transformers for multi-document summarization," which explores advanced techniques in summarizing information from multiple documents using hierarchical transformer models. Additionally, Lapata, M. has contributed to the paper "Text summarization with latent queries," which delves into innovative methods for summarizing text by leveraging latent query representations. These works highlight Lapata, M.'s expertise and active research in the domain of text summarization, showcasing a focus on developing sophisticated models and methodologies to enhance the efficiency and accuracy of summarization tasks. + 6cd82819982879bd164547d2773ba5c7,fc4b27d64f055b7fc30176ba110dd02e + 2 + 616 + b349041c0be64c62b964ab1234e055e6 + + + TECHNOLOGY + LlamaIndex Knowledge Graph Index is a technology developed by LlamaIndex + 6cd82819982879bd164547d2773ba5c7 + 0 + 617 + 969e1ea0b1e443a68e9a65dfef91d161 + + + PERSON + Manakul, P. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + 0 + 618 + 8e09e7cfea7d405db8b22ae2f836ccb1 + + + PERSON + Liusie, A. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + 0 + 619 + 490583524d394bf79289c5fe34f7dcf1 + + + PERSON + Gales, M. J. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + 0 + 620 + d7db38bb599c42cab7066f3fdd282282 + + + PERSON + Mao, Y. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 621 + efd87a59d01e47c8adc02f63ef2c5c3e + + + PERSON + He, P. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 622 + 80e3ce3de41e4601823a333e22b7bb3f + + + PERSON + Liu, X. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 623 + 50eabc166e8944a49197e79c32f27597 + + + PERSON + Shen, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Shen, Y.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 0 + 624 + 5197a3fb02ef4677abd1900aa87e4efa + + + PERSON + Gao, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 625 + 887f444240bb474da23cdfb6abf7a998 + + + PERSON + Han, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + 0 + 626 + 5d29053f2ce74442aa1855b327ef3bb7 + + + PERSON + Chen, W. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Chen, W.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 627 + 7e40cd12839a4577a95e33d785147a31 + + + PERSON + Martin, S. is an author of the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing a comprehensive, open-source solution for the layout of large graphs, which is a critical task in the visualization and analysis of complex networks. The toolbox aims to facilitate the understanding and interpretation of large-scale graph data, making it a valuable resource for researchers and practitioners in fields such as computational linguistics, information retrieval, and data science. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 628 + 8fe58de8a04f4f8f807c77fb41829a3a + + + PERSON + Brown, W. M. is an author of the paper "Openord: An open-source toolbox for large graph layout." + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 629 + a9f50861273c4bb697d868a9d049d392 + + + PERSON + KLAVANS, R. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 630 + be4820f29fd942b282049fa49697b4ed + + + PERSON + Boyack, K. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on the development and application of Openord, a comprehensive open-source toolbox designed for the layout of large graphs. The paper likely discusses the methodologies, algorithms, and practical implementations of the toolbox, contributing to the fields of graph theory and data visualization. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 3 + 631 + 6deaefe707f84b3dbda979dea0d095ac + + + EVENT + The conference where the paper "Openord: An open-source toolbox for large graph layout" was presented + 833e7d67dcd30790b26b71c9b5306f6b + EVENT + 0 + 632 + d053ea9432a24fb192e8d6aa993b0caa + + + TECHNOLOGY + GPT-4 is a large language model used in Microsoft's study on scientific discovery + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + 1 + 633 + a3e683d294ed42a28d60d09a36cbeb54 + + + TECHNOLOGY + Project NaLLM is a project developed by Neo4J + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + 1 + 634 + 39887ca8567141d5b857b87a2bca4086 + + + PERSON + Newman, M. E. is the author of the paper "Modularity and community structure in networks" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 1 + 635 + 8df8563ab0394ee9a91b89dea7d59404 + + + PUBLICATION + The journal where the paper "Modularity and community structure in networks" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + 1 + 636 + 12398f70065143839d812fd42ac4b2e7 + + + PERSON + Ram, O. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 637 + 74d43d20f251441baf8e3db64fedca43 + + + PERSON + Levine, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 638 + 1b7a22f76f7741e8b140bdc3d8856d76 + + + PERSON + Dalmedigos, I. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 639 + b823ba1bfe944fa9887edd8faf8a5f17 + + + PERSON + Muhlgay, D. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 640 + d0bfb473fdc64643954cdb4675e2f389 + + + PERSON + Shashua, A. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 641 + a4db1b2a9c3e4d2d838725f8166c36b4 + + + PERSON + Leyton-Brown, K. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 642 + 8dae140578c841ae9373cbc607c4a6e6 + + + PERSON + Shoham, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 6 + 643 + b215cc33cf40434f87f284ff8f3506a4 + + + PUBLICATION + The journal where the paper "In-context retrieval-augmented language models" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + 0 + 644 + c1ff9d8e1b8745d6860c34ce26122d79 + + + PERSON + Ranade, P. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 1 + 645 + 9d1e6ca9ae8e4e068fb74631a633b20b + + + PERSON + Joshi, A. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 1 + 646 + 1d7b0deca7674777bf76c163ac065845 + + + PERSON + Sarthi, P. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 5 + 647 + 03afe9988f864c9fa501bfbf043f74c0 + + + PERSON + Abdullah, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 5 + 648 + 4084f614af494fa8ab73095fb5b6b07b + + + PERSON + Tuli, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 5 + 649 + 3ce25564af6e47f390a0b16b6f9433a1 + + + PERSON + Khanna, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 3 + 650 + 78213664d0eb45d1a9239ba4b85b10f7 + + + PERSON + Goldie, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 3 + 651 + 1226e4a4077b4b3a970db4d2509b590c + + + PERSON + Manning, C. D. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" and the paper "Raptor: Recursive abstractive processing for tree-organized retrieval". These contributions highlight Manning's involvement in advancing the fields of Natural Language Processing and Information Retrieval, particularly in the areas of multi-hop question answering and recursive abstractive processing. + 833e7d67dcd30790b26b71c9b5306f6b,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 4 + 652 + b4c7de7a824a4a71b9f52193d2f1a10d + + + PERSON + Scott, K. is associated with "Behind the Tech" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 0 + 653 + b609f1939dae4c7383c7d199bb3c7dc3 + + + PERSON + Shao, Z. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 0 + 654 + aeee2f443dfb4e3ea80af6ae1d9197ce + + + PERSON + Gong, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + 0 + 655 + 8c46d37bc26e4d4dbd37d6ee26867bc6 + + + PERSON + Huang, M. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + 2 + 656 + 58a8fa7f29e347bdb9689b70b065a779 + + + PERSON + Duan, N. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + 2 + 657 + fae3fe31deb141ab93143ac411f1eaaa + + + PERSON + Su, D. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 658 + a2cb46c226b94831853a5d28c5d94b0a + + + PERSON + Xu, Y. is an author of multiple academic papers in the field of Natural Language Processing and Information Retrieval. Notably, Xu, Y. contributed to the paper titled "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," which addresses the management of scholarly information related to COVID-19 through advanced question answering and summarization techniques. Additionally, Xu, Y. co-authored the paper "Text summarization with latent queries," which explores innovative methods for text summarization by leveraging latent queries. These contributions highlight Xu, Y.'s expertise and active involvement in developing sophisticated systems for information retrieval and summarization. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 7 + 659 + d3511ecd27cd4166bdb39e757e275300 + + + PERSON + Yu, T. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 660 + de3b561f5cce4c83bccb39180e362c97 + + + PERSON + Siddique, F. B. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 661 + 5bfefaa0fce04002851733337bed714c + + + PERSON + Barezi, E. J. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 662 + b5fed5609f154df58c6a9f74e55fc0ba + + + PERSON + Fung, P. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 5 + 663 + 91ae5251eaab4c08afe6cd4cbefcaa6b + + + PERSON + Tang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 664 + bbdd53a15e99452a9deff05d1de2d965 + + + PERSON + Yang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 665 + 532bf54d5a924ff48aee254970efb914 + + + PERSON + Touvron, H. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9 + 666 + 2489232bd2bb492babe00617e7290282 + + + PERSON + Martin, L. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 7 + 667 + d2ed972353af4d1db74702638bfdbb58 + + + PERSON + Stone, K. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 668 + 575befc8d64c47eb95af8b1096e02963 + + + PERSON + Albert, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 669 + d6e6366617e04b0ba6732fd1d2d76429 + + + PERSON + Almahairi, A. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 670 + b4c4354c8edb40db984942799fe0c8b1 + + + PERSON + Babaei, Y. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 671 + 170507a64973429f818067b80506d428 + + + PERSON + Bashlykov, N. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 672 + fd9b298e6aea4685bbb2064b05fcda79 + + + PERSON + Batra, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2 + 673 + eeecb159cc8a4c8989f8da0f3df09f2a + + + PERSON + Bhargava, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 674 + 70f22b1d7336492dbade94b8edefe457 + + + PERSON + Bhosale, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1 + 675 + 66e098dc431146e19fc4bc2ea37efbd9 + + + PERSON + Traag, V. A. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 676 + 932e213c57134098a07073febd51dcc2 + + + PERSON + Waltman, L. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 677 + 9593428ad36746ae8af6d8ce639834ef + + + PERSON + Van Eck, N. J. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 678 + 1bcaeb58479d42a6963a073c09f3f397 + + + PUBLICATION + Scientific Reports is the journal where the paper "From Louvain to Leiden: guaranteeing well-connected communities" was published + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 679 + 1ef0c1c59ce946668ccf1a6a4f5ab7cc + + + PERSON + Trajanoska, M. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 680 + d734746e3d6146f780af91827e578dfd + + + PERSON + Stojanov, R. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 681 + 21ed913271614cbeb1b754cdbbef13af + + + PERSON + Trajanov, D. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 682 + 1505dfebbfb04652b0ba57de1a251d67 + + + PERSON + Trivedi, H. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 683 + 907ec65076e5494a8631efffb81b3178 + + + PERSON + Balasubramanian, N. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 684 + 2dc7f6b230db452190a09643ca3d5ec0 + + + PERSON + Khot, T. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 685 + c20ecfc93b3a4875ade5c92cfe4b94a1 + + + PERSON + Sabharwal, A. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 686 + 4bc7dc91ede345dfb63d7d4f7ac3554f + + + PERSON + Wang, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 86 + 2 + 8 + 687 + 0b2b815c9f834aaaac0c341097def9ba + + + PERSON + Liang, Y. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 86 + 2 + 8 + 688 + 424ae71c56024094a02e6fd9bfcfbb04 + + + PERSON + Meng, F. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 86 + 2 + 8 + 689 + 400d10f2ee1d49be9a66efa34dada0e6 + + + PERSON + Sun, Z. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 86 + 2 + 8 + 690 + 91deb9f152264e958d106d481ff2e1ee + + + PERSON + Shi, H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 86 + 2 + 8 + 691 + 586cf02da9494088aed9b3419725638f + + + PERSON + Li, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through their work on evaluating language models. Specifically, Li, Z. has co-authored the paper titled "Is ChatGPT a Good NLG Evaluator? A Preliminary Study," which explores the effectiveness of ChatGPT as a natural language generation evaluator. Additionally, Li, Z. has co-authored another paper, "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which examines the performance of large language models in evaluative roles using specific benchmarking tools. These contributions highlight Li, Z.'s active involvement in advancing the understanding and assessment of language models within the academic community. + 8d87efac8c50cf20cdf26bf61e5e2035,b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + 85 + 2 + 20 + 692 + 229d85a2783e4a2991f17d2ab5750af7 + + + PERSON + Xu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 85 + 2 + 9 + 693 + b7f97d1909a3433abef8ca8e9334fafa + + + PERSON + Qu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 85 + 2 + 9 + 694 + b7fdfffc38b94bf7872eabe9b022c8fd + + + PERSON + Zhou, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 85 + 2 + 9 + 695 + 6242e0c237a348908d0256ea790a0211 + + + PERSON + Wang, S. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" and also contributed to the paper "Is chatgpt a good nlg evaluator? a preliminary study." These works indicate Wang, S.'s involvement in cutting-edge research within the fields of federated search, retrieval augmented generation, and natural language generation evaluation, showcasing a focus on both the technical and evaluative aspects of Natural Language Processing and Information Retrieval. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 4 + 696 + 7cc9f26737e1442595e53253e98015ef + + + PERSON + Khramtsova is an author mentioned in the text + 8d87efac8c50cf20cdf26bf61e5e2035 + 0 + 697 + 1868fec1493643208dbdcad7bc97dfa0 + + + PERSON + H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 85 + 2 + 4 + 698 + a87aa935dccf49cd98b40fb5afe7ad5c + + + PERSON + Khramtsova, E. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 4 + 699 + 36870a3393f6413e9bf647168eb6977a + + + PERSON + Zhuang, S. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through multiple academic papers. Notably, Zhuang, S. co-authored the paper titled "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," which explores the evaluation of federated search systems within the framework of retrieval-augmented generation. Additionally, Zhuang, S. co-authored another significant paper, "Judging llm-as-a-judge with mt-bench and chatbot arena," which delves into the assessment of large language models (LLMs) using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Zhuang, S.'s active involvement in advancing research in federated search and the evaluation of LLMs. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + 83 + 2 + 15 + 700 + 4fe3ff52700c491f8cc650aadb4d7cb0 + + + PERSON + Zuccon, G. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 4 + 701 + f1f6f6435a444e388d67e16e847afca6 + + + PERSON + Wang, Y. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 702 + 0af2ca1c090843ea92679fd14c1fbc9a + + + PERSON + Lipka, N. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 703 + 1b06d3e53ffd4771952fbef04d1e666c + + + PERSON + Rossi, R. A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 704 + b8e966b34cba4b11b9995106767212ba + + + PERSON + Siu, A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 705 + f6de923de6474d2cab6a9c2f0d81fa59 + + + PERSON + Zhang, R. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 706 + 6915637e8d124fdc8473111d501e3703 + + + PERSON + Derr, T. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 707 + 2233f31929194eac89333ce8731a5584 + + + PERSON + Yang, Z. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 6 + 708 + 61f1dc4267314470ac820b6a46c61f7b + + + PERSON + Qi, P. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 709 + f0c578614b224345974c3e4c110878af + + + PERSON + Zhang, S. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 710 + 7ffb88ebc729492c897ccfb569d7f6d0 + + + PERSON + Bengio, Y. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 711 + 60dce7d8bc1b4729a038178a400b9a59 + + + PERSON + Cohen, W. W. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 712 + 4cbb4e238c5b4656803fb9b4b6c3512e + + + PERSON + Salakhutdinov, R. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 1 + 713 + 652873bcd6d5432187e5deafc4fc5211 + + + CONFERENCE + The conference where the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" was presented + fc4b27d64f055b7fc30176ba110dd02e + 0 + 714 + 78f9b30c08134ac5abb4f4e0bff0f7f2 + + + PERSON + Yao, J.-g. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + 0 + 715 + f33e4e897b1e422bb516e8a2c941d9dc + + + PERSON + Wan, X. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + 0 + 716 + fac4e1553a9840e990bbfff46e64ff27 + + + PERSON + Xiao, J. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + 0 + 717 + 029a55d327ee4fb3a8314b36d52bdf34 + + + PUBLICATION + The journal where the paper "Recent advances in document summarization" was published + fc4b27d64f055b7fc30176ba110dd02e + 0 + 718 + 5a636c894c384532bff66212cf9d5824 + + + PERSON + Yao, L. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models"Yao, L. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 81 + 2 + 1 + 719 + a9c468ef78704e9aabfc0317a5b1b42d + + + PERSON + Peng, J. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 0 + 720 + 5df80c25d33a4d148a14aa614343cc6b + + + PERSON + Mao, C. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 0 + 721 + 6a87f06ed55a46f29b24f77e548a3f1d + + + PERSON + Luo, Y. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 0 + 722 + 0daf88ac4ec94cbb868e27e956c6d7f1 + + + PERSON + Zhang, J. is an author of the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 81 + 2 + 1 + 723 + 9ed120043e6247be9965e4904920991b + + + PERSON + Zhang, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 81 + 2 + 1 + 724 + 94d81d7de9254ae4b3b16fcc69aa22ea + + + PERSON + Gan, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 81 + 2 + 1 + 725 + 60c9212246f84ae5b6ab254127a39262 + + + PERSON + Wang, C. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + 81 + 2 + 1 + 726 + 0f8d0c36a4274526a9eddedae5e63881 + + + PERSON + Zheng, L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zheng, L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Zheng, L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools like MT-Bench and Chatbot Arena. These contributions highlight Zheng, L.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR domains. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 83 + 2 + 12 + 727 + 6aedd377efbe4f07ae42e546996e7bfa + + + PERSON + Chiang, W.-L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Chiang, W.-L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Chiang, W.-L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Chiang, W.-L.'s active involvement in advancing the understanding and capabilities of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 83 + 2 + 12 + 728 + 1aa8484562784f378851c33843c89687 + + + PERSON + Sheng, Y. is an author known for contributing to the field of Natural Language Processing and Information Retrieval. Notably, Sheng, Y. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Sheng, Y. has contributed to the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Sheng, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic and technical community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 83 + 2 + 12 + 729 + f1a65d05dd5d456b889217020475ef80 + + + PERSON + Wu, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Wu, Z. co-authored the paper titled "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Wu, Z. is also credited with co-authoring the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Wu, Z.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 83 + 2 + 12 + 730 + c077d92b48b6477db91e1a0460600f52 + + + PERSON + Zhuang, Y. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zhuang, Y. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness of knowledge graphs. Additionally, Zhuang, Y. has also authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Zhuang, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the domain. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 83 + 2 + 12 + 731 + 5ca888df9b884e54accdd2ff29d125c1 + + + PERSON + Lin, Z. is an author of the paper "Exploring large language models for knowledge graph completion" and also contributed to the paper "Judging LLM-as-a-judge with MT-Bench and Chatbot Arena." These works indicate Lin, Z.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the application of large language models for tasks such as knowledge graph completion and the evaluation of language models in judgment tasks. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 83 + 2 + 12 + 732 + 8290a6212d6c4430ae0056c7e8eccd5f + + + PERSON + Li, D. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant research. Notably, Li, D. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Li, D. has also co-authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Li, D.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 83 + 2 + 12 + 733 + 14f8ac195fdb4e06a0b9ebc6ef391180 + + + PERSON + Xing, E. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Xing, E. contributed to the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Xing, E.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + 83 + 2 + 12 + 734 + 667ee58a79194316ae2b82eadd3fc575 + + + TECHNOLOGY + Chatbot Arena is a platform or tool used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 83 + 2 + 11 + 735 + b0e3ee2324054c88adacdf80db13278f + + + 1.0 + Darren Edge and Ha Trinh co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 28b7457ca5dc4a38a488946a3f8e207e + 0 + 2 + + + 1.0 + Darren Edge and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 8029a14d15404e6db95ddf5e2bf9fc15 + 1 + 2 + + + 1.0 + Darren Edge and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 389314ca89d445888c8d4985864dd733 + 2 + 2 + + + 1.0 + Darren Edge and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 87fe1462b9064d5692641ab48e826301 + 3 + 2 + + + 1.0 + Darren Edge and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + a55175ac57014df696ca09d0def9604b + 4 + 2 + + + 1.0 + Darren Edge and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 1766e8858d7b45ed97f71cb5a39e96ea + 5 + 2 + + + 1.0 + Darren Edge and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 6191e014f3f64e46a0777063ed4ac19a + 6 + 2 + + + 1.0 + Darren Edge is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + 21b0499cf14342269c46170c291d0535 + 7 + 2 + + + 1.0 + Ha Trinh and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + c1ef05b38b3f4d59888150fc0dd26826 + 8 + 2 + + + 1.0 + Ha Trinh and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 74cb9b3510e84498b9aee0d904316e8b + 9 + 2 + + + 1.0 + Ha Trinh and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 043d764b2e1b4d1294651ff938df5391 + 10 + 2 + + + 1.0 + Ha Trinh and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 31f2170fef004f3281c533a4a60dc3f3 + 11 + 2 + + + 1.0 + Ha Trinh and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 57f186c5c2754483ba66750e98222f95 + 12 + 2 + + + 1.0 + Ha Trinh and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 4b3fc569d91f4a7aa6501ad4fcf67b7a + 13 + 2 + + + 1.0 + Ha Trinh is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + cbc1667556f84a5eadf867a823e6986c + 14 + 2 + + + 1.0 + Newman Cheng and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + a876d1ab79864396bc47a039225fd5c7 + 15 + 2 + + + 1.0 + Newman Cheng and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + c09f67d4f25448c99f7c0552c30b7706 + 16 + 2 + + + 1.0 + Newman Cheng and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + c0866306dc8c4da2a8a81c0c3a78b657 + 17 + 2 + + + 1.0 + Newman Cheng and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 3884c37eb13a4c9097ee2c5be4eeefaf + 18 + 2 + + + 1.0 + Newman Cheng and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 90764eb2cab74cffb1c7d72d28b965cc + 19 + 2 + + + 1.0 + Newman Cheng is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + 01abe16e67c241a887aa62abe22d155c + 20 + 2 + + + 1.0 + Joshua Bradley and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 37049be0a2c240c6a06acf9339237b8b + 21 + 2 + + + 1.0 + Joshua Bradley and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + e785c52881704d95bf4ec03d2720f8ae + 22 + 2 + + + 1.0 + Joshua Bradley and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 654689c65613476b9905d7afb3809cd2 + 23 + 2 + + + 1.0 + Joshua Bradley and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 15dfb45a6ffa4d34ad72cfe4b3c5cc0d + 24 + 2 + + + 1.0 + Joshua Bradley is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + 427c3b7458f148d8bace1b768e2b5b7c + 25 + 2 + + + 1.0 + Alex Chao and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 95d506750fd94e72bbd9cf2d3fe18e28 + 26 + 2 + + + 1.0 + Alex Chao and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + bf0138ccbcc740089a55fd0c24897360 + 27 + 2 + + + 1.0 + Alex Chao and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 83cd5df42643494396b00d6cb6376def + 28 + 2 + + + 1.0 + Alex Chao is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + 909d28e443fd4e0bac189373125c8309 + 29 + 2 + + + 1.0 + Apurva Mody and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + efbc2439e5034801af83ac1a0b440535 + 30 + 2 + + + 1.0 + Apurva Mody and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + b9a2ef791a064f038cac2059ebea1138 + 31 + 2 + + + 1.0 + Apurva Mody is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + 1ce2b24bc93442148dc2240d3c6223b1 + 32 + 2 + + + 1.0 + Steven Truitt and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + 804c1e94e7974332a817931363ddb643 + 33 + 2 + + + 1.0 + Steven Truitt is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + 32dc0b572ad84c75a64a2007788eb981 + 34 + 2 + + + 1.0 + Jonathan Larson is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + f5c11a5ac94e40068bca8be178a6bcd6 + 35 + 2 + + + 4.0 + RAG (Retrieval-Augmented Generation) and LLM (Large Language Models) are closely intertwined in the domain of Natural Language Processing and Information Retrieval. RAG is employed to enhance the capabilities of LLMs by enabling them to retrieve pertinent information from external knowledge sources. This symbiotic relationship allows LLMs to generate and assess text more effectively. Specifically, RAG leverages the power of LLMs to access and utilize relevant data, thereby augmenting the overall performance and accuracy of text generation tasks. + e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + ea28ff7f127e4677a913952595dce2f5 + 36 + 2 + + + 7.0 + Graph RAG is a specific implementation of RAG that combines the strengths of RAG with graph-based text indexing. This method leverages the natural modularity of graphs to partition data, facilitating global summarization. As a specialized approach within the RAG framework, Graph RAG enhances the capabilities of RAG by integrating graph structures to improve the efficiency and effectiveness of text data processing and summarization. + 21e52bc06a82796b1f4bcd73edda1f2a,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 192a6d23595045f38b0d46a3d8e52fd6 + 37 + 2 + + + 1.0 + Query-Focused Summarization is a task that RAG fails to address effectively + e8d83e6e7a7c0f57b218cef24976b745 + ef67c9fc60284b50aa15ac655b06a155 + 38 + 2 + + + 1.0 + RAG retrieves relevant information from an external knowledge source + e8d83e6e7a7c0f57b218cef24976b745 + cc8201cce1024b5192056fe8e98fda22 + 39 + 2 + + + 1.0 + Naive RAG is a specific implementation of RAG + e8c8f911135faf3ff35f24107eb3f99c + 97e097f9022540b88ab7c13d2805c25f + 40 + 2 + + + 1.0 + Ram et al., 2023 discusses RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + 829a6299a5fa4e7b8ff4020020a0be05 + 41 + 2 + + + 2.0 + Naïve RAG is a basic form of RAG + f35de4d9fb65f1d5a392064b20545c19 + dde2742459c24fb4a91172aa5c1a7620 + 42 + 2 + + + 2.0 + Modular RAG is an advanced form of RAG + f35de4d9fb65f1d5a392064b20545c19 + 323979a67d79498fa271acdf8cd1a0c2 + 43 + 2 + + + 2.0 + LLMs are used in various RAG tasks such as knowledge graph creation and completion + 92e93fc6449756c0a60200636b297f65 + c7e8b188b45841a0a1bcb22f3445ea6e + 44 + 2 + + + 2.0 + The paper by Trajanoska et al. discusses using LLMs for knowledge graph creation, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 5a4ad077106a4a3f951f43d2e01499b0 + 45 + 2 + + + 2.0 + The paper by Yao et al. discusses using LLMs for knowledge graph completion, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + a7ec8df038d7461689d28f1bdea84d9b + 46 + 2 + + + 2.0 + The paper by Ban et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 8ddefa32e2ed4eaf8f76d17a676f74f3 + 47 + 2 + + + 2.0 + The paper by Zhang et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 95ec30ce8dbe4ca28714e3e3735da8f3 + 48 + 2 + + + 2.0 + The paper by Gao et al. discusses advanced RAG where the index is a knowledge graph + 92e93fc6449756c0a60200636b297f65 + 259e7f5e2ec04418937513413b6d51d1 + 49 + 2 + + + 2.0 + KAPING is a method where the index is a knowledge graph, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 2289f06dd3804a3c84371dda0bab091e + 50 + 2 + + + 2.0 + G-Retriever is a method where subsets of the graph structure are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 34ff8ef897804691842071f9ff78708e + 51 + 2 + + + 2.0 + Graph-ToolFormer is a method where derived graph metrics are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + a7401447d994439993da7cc57f127649 + 52 + 2 + + + 2.0 + SURGE is a method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 754b0f2616064b18abb90f409ef0539a + 53 + 2 + + + 2.0 + FABULA is a method where retrieved event-plot subgraphs are serialized using narrative templates, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + acd35bb6b3cb4979a3f3fb68a86b3b05 + 54 + 2 + + + 2.0 + The paper by Wang et al. discusses a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + 9e1e7f67ba044c7fbf64723af1ade58e + 55 + 2 + + + 2.0 + Sensemaking questions are used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + 57e16ff087a84b8ebd70de1e7e534225 + 56 + 2 + + + 2.0 + The evaluation of RAG systems focuses on corpora in the region of 1 million tokens + 92e93fc6449756c0a60200636b297f65 + bbf4007dc9c0486b8ea76d616045467a + 57 + 2 + + + 2.0 + Trade-offs are considerations involved in building a graph index for RAG systems + 92e93fc6449756c0a60200636b297f65 + 9535f4d754044e128cd3951a9d2e3702 + 58 + 2 + + + 2.0 + A graph index is a data structure used in RAG systems to organize and retrieve information + 92e93fc6449756c0a60200636b297f65 + e1ed13e29ee946d4aaafac50aaa3b68f + 59 + 2 + + + 2.0 + Performance of RAG systems varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + eb961d47a30c4870a1134b4a4672a8b2 + 60 + 2 + + + 2.0 + Different data types are used in RAG systems + 92e93fc6449756c0a60200636b297f65 + 5b019e8652264136b95306bac70a2e25 + 61 + 2 + + + 2.0 + Dataset sizes affect the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + 325fc9e2b37043b7af9f6ad338b09469 + 62 + 2 + + + 2.0 + Evaluation is the process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + 6bb11aa08b414232b5b45f10f5766f62 + 63 + 2 + + + 2.0 + Corpora are collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + 179737fd23c943babdfae01ac5c6bfc3 + 64 + 2 + + + 2.0 + Different question types are used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + 65a31e4da283411fb7c971f63d606723 + 65 + 2 + + + 2.0 + Target metrics are specific measures used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + de31810d43174a52aa2f31b72f4542f5 + 66 + 2 + + + 4.0 + Graph RAG utilizes Large Language Models (LLMs) to construct a graph-based text index, enabling the generation of summaries and the answering of queries. In this approach, LLMs play a crucial role in analyzing and generating text based on the information retrieved through the graph structure. Additionally, LLMs leverage the Graph RAG framework to provide comprehensive overviews of public figures in the entertainment industry. This integration of LLMs within Graph RAG enhances the system's ability to process and synthesize complex text data, making it a powerful tool for information retrieval and natural language processing tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + a6ae1d99330443fcacb06ace15a0d937 + 67 + 2 + + + 1.0 + Text chunks are processed using LLM to extract elements of a graph index + bc9e2c9e369c4108cf4f6dd5f60960f4 + 5174cdabb6024de0975762d3a80b059f + 68 + 2 + + + 1.0 + LLM is used to extract elements of a graph index from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + e379fba901174b529250169e62d98c09 + 69 + 2 + + + 2.0 + LLM (Large Language Model) and Few-Shot Examples are closely related in the context of Natural Language Processing and Information Retrieval. Few-shot examples are provided to the LLM for in-context learning, which helps tailor the extraction prompt. This technique is particularly useful for improving the performance of the LLM in specialized domains. By leveraging a small number of examples, the LLM can better understand and adapt to specific tasks, thereby enhancing its overall effectiveness in extracting and processing information within those specialized areas. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4 + 81ee8bb20bbb4d37bc0db642f1c75b8e + 70 + 2 + + + 1.0 + LLM extracts named entities from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 93b4aa6ce6e44123a861d4c3b3d509a2 + 71 + 2 + + + 1.0 + Kuratov et al. (2024) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + ee8414e314f547eeb369849cdb51bac2 + 72 + 2 + + + 1.0 + Liu et al. (2023) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + 9f77aa8888bd4f94abba8a77c4b0565c + 73 + 2 + + + 1.0 + LLM prompts are instructions given to the LLM for extracting elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + dcf33412678340319e7ec8f7be267ef9 + 74 + 2 + + + 1.0 + Recall degradation occurs with longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + 4aa1e0fa00c048939a5d006bfd305fb4 + 75 + 2 + + + 1.0 + The extraction process involves using LLM to identify and extract elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + 03053ab4a9054384a5f5e88d28841621 + 76 + 2 + + + 1.0 + Default prompt is the standard set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + 9fd0f20997d541bca46c4ec9843a5d0f + 77 + 2 + + + 1.0 + Secondary extraction prompt is an additional set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + 27168beee1ff456696c330c9c3b3259f + 78 + 2 + + + 1.0 + The LLM uses covariate prompts to extract additional attributes associated with detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + e1c20e06aeac436788a9c6e918bcb844 + 79 + 2 + + + 1.0 + The LLM uses multiple rounds of gleanings to ensure no entities are missed + 2c6ed90897310eea2f28e33fff1c32b0 + 344417f626ef4da4be4539ef4037bf3f + 80 + 2 + + + 1.0 + Logit bias is used to force a yes/no decision from the LLM during entity extraction + 2c6ed90897310eea2f28e33fff1c32b0 + 8b1fff87c350475fb1d411a26c3c5b0c + 81 + 2 + + + 1.0 + The LLM extracts element instances from source texts + 2c6ed90897310eea2f28e33fff1c32b0 + 898a9458adfb4c13a1eafacf6a1068f6 + 82 + 2 + + + 1.0 + The LLM detects and summarizes communities of entities + 2c6ed90897310eea2f28e33fff1c32b0 + 5448f05781de44ea96e3dea40b285842 + 83 + 2 + + + 1.0 + LLM generates intermediate answers and scores for each chunk + 1d07b4248c2655081c7af0e373bd70c9 + 76b1e69904b84d09ba05c4b7efc48f32 + 84 + 2 + + + 1.0 + LLM generates a helpfulness score for each answer + 1d07b4248c2655081c7af0e373bd70c9 + 3f5590a604894d268603b4b27c3348b5 + 85 + 2 + + + 2.0 + LLM is used to generate questions for evaluating the Podcast Transcripts dataset + 922778ce1cb2fdd6dbab1746c8795620 + 68f998c9c8c34bb7a994de5a998bb9a0 + 86 + 2 + + + 2.0 + LLM is used to generate questions for evaluating the News Articles dataset + 922778ce1cb2fdd6dbab1746c8795620 + aafc13d02ade40adae13d3bee241817a + 87 + 2 + + + 1.0 + LLM uses Naive RAG to list public figures mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 81a4818e5cf84ea085abf09de385c86e + 88 + 2 + + + 1.0 + LLM-generated responses are evaluated using assessment metrics + e8c8f911135faf3ff35f24107eb3f99c + b69851bf63e34ced83827b0021628543 + 89 + 2 + + + 1.0 + LLM-generated responses are evaluated using specific questions + e8c8f911135faf3ff35f24107eb3f99c + b83a4e11bfa64559954327714b73293f + 90 + 2 + + + 1.0 + Ad-hoc LLM use involves the use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + de23b974cc90497eb4363e26d931a57c + 91 + 2 + + + 2.0 + LLMs are used for knowledge graph creation + f35de4d9fb65f1d5a392064b20545c19 + a9de65176e234a9f9073b8df9d675e90 + 92 + 2 + + + 2.0 + LLMs are used for knowledge graph completion + f35de4d9fb65f1d5a392064b20545c19 + 09a1bd11eb9347a9b466edad1a562cc5 + 93 + 2 + + + 2.0 + LLMs are used for the extraction of causal graphs + f35de4d9fb65f1d5a392064b20545c19 + 11d74eab1dcb4fcba7c45def5f0ee22d + 94 + 2 + + + 2.0 + LLMs are used for knowledge graph creation as per Trajanoska et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 4f6a6fd018a948f4bd0e630266b8bf61 + 95 + 2 + + + 2.0 + LLMs are used for knowledge graph completion as per Yao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 17dbfbecfaf0436bb11ed8f867c0caa1 + 96 + 2 + + + 2.0 + LLMs are used for the extraction of causal graphs as per Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + 2b1ec99684574c2ab26bb050d5b57a4d + 97 + 2 + + + 1.0 + Graph RAG is proposed as a method to combine the strengths of RAG and QFS + e8d83e6e7a7c0f57b218cef24976b745 + 1ccce5d1892a4b6995bbaec22882d34d + 98 + 2 + + + 7.0 + Graph RAG is an approach that utilizes community summaries in various capacities to enhance its functionality. Specifically, community summaries are generated within the Graph RAG framework and are employed to generate partial responses. These summaries are also used to compare against source texts, serving as a form of self-memory for the system. By leveraging community summaries, Graph RAG aims to improve the comprehensiveness and diversity of answers. Additionally, Graph RAG incorporates summaries of root-level communities within an entity-based graph index, further enriching its analytical capabilities. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + 51cd93f89fbe4bcf883cdb2ca6774cd6 + 99 + 2 + + + 1.0 + Graph RAG is designed to handle global sensemaking questions over large datasets + e8d83e6e7a7c0f57b218cef24976b745 + 5f353b18fadb438f95ba0ea8feae137c + 100 + 2 + + + 2.0 + Graph RAG is implemented in Python. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + 947d70dd14b34cf398a1ab6dbdc51161 + 101 + 2 + + + 1.0 + The open-source implementation of Graph RAG will be available at this URL + e8d83e6e7a7c0f57b218cef24976b745 + 90f5597a558a4652bded9001a4ec2e56 + 102 + 2 + + + 1.0 + Graph RAG uses an entity knowledge graph to index text + e8d83e6e7a7c0f57b218cef24976b745 + 9532cf83e9324ea0a46e5ac89bac407d + 103 + 2 + + + 3.0 + Graph RAG is an approach that is evaluated for its comprehensiveness, a target quality used to assess its effectiveness. The method aims to improve the comprehensiveness of generated answers, ensuring that the information provided is thorough and complete. This evaluation metric is crucial in determining the success of Graph RAG in producing detailed and accurate responses. + 21e52bc06a82796b1f4bcd73edda1f2a,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + 8919fa72a9e74d1daff801e8f4c15b2b + 104 + 2 + + + 3.0 + Graph RAG is an approach in the domain of Natural Language Processing and Information Retrieval that focuses on improving the diversity of generated answers. Diversity, in this context, is a target quality used to evaluate the performance of the Graph RAG approach. By enhancing the diversity of responses, Graph RAG aims to provide a broader range of answers, thereby improving the overall effectiveness and robustness of the system. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745 + bef38889bb86413895d7dd25b4c3137c + 105 + 2 + + + 3.0 + Graph RAG uses a knowledge graph for global summarization + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + f770bc07cecf4aba8fe2d2c33fdc5542 + 106 + 2 + + + 1.0 + Community detection algorithms are used in the Graph RAG approach to partition graphs + 21e52bc06a82796b1f4bcd73edda1f2a + 13cd49512d5642989c2c72bb5e674807 + 107 + 2 + + + 1.0 + Podcast transcripts are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + e5c5c87a281b43868c344ff60f44c100 + 108 + 2 + + + 1.0 + News articles are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + a562ffbe986247b7943990e7151f4d69 + 109 + 2 + + + 2.0 + Graph RAG is evaluated using the target quality of Empowerment. Empowerment is specifically utilized to assess Graph RAG's capability in aiding users to achieve an informed understanding. This evaluation metric underscores the importance of user comprehension and the effectiveness of the Graph RAG approach in facilitating informed decision-making processes. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4 + 7ea0bc1467e84184842de2d5e5bdd78e + 110 + 2 + + + 2.0 + Graph RAG is compared to naive RAG in the evaluation. In this comparison, Graph RAG outperformed naive RAG on comprehensiveness and diversity. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + 056f23eb710f471393ae5dc417d83fd9 + 111 + 2 + + + 1.0 + Graph RAG is compared to global map-reduce summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + e1ae27016d63447a8dfa021370cba0fa + 112 + 2 + + + 1.0 + Query-focused summarization is a method used in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + f8c10f61a8f344cea7bdafa2d8af14b8 + 113 + 2 + + + 1.0 + Activity-centered sensemaking questions are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + aa7d003f25624e19bc88d3951d4dc943 + 114 + 2 + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 1c97184ce5ea4049be417a3fd125357b + 115 + 2 + + + 2.0 + The "Graph RAG" approach is evaluated in terms of its performance by considering "Token Costs." Token costs are measured to assess the efficiency of the Graph RAG method, indicating that the computational expense associated with processing tokens is a critical factor in determining the overall effectiveness of this approach. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 13a044c404394c34af1e9b07c48aa985 + 116 + 2 + + + 1.0 + Data flow describes the high-level process of the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 69ef1ac7b1f44372979149e82ecbc860 + 117 + 2 + + + 3.0 + Design parameters are key settings in the Graph RAG approach and significantly influence the Graph RAG approach and pipeline. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + 6e26ce67bacc4fa089296843463f69ad + 118 + 2 + + + 1.0 + Graph RAG uses global summarization to summarize information from a large dataset + 21e52bc06a82796b1f4bcd73edda1f2a + ae0d3104647f4e6ab3ec2cf8e60be5ca + 119 + 2 + + + 1.0 + Graph RAG aims to answer specific queries + 21e52bc06a82796b1f4bcd73edda1f2a + 49e24b5f2c1d40d7857afe327db4f554 + 120 + 2 + + + 1.0 + Graph RAG uses a corpus for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + 587f39a32e93412395d9c22ad0ac2f94 + 121 + 2 + + + 1.0 + Activity-centered sensemaking is used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 8d9ded5fc9cf4c4faba8c6c8cd50e2f4 + 122 + 2 + + + 1.0 + Real-world datasets are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 595a841aa6034c93bd3dc55681e17710 + 123 + 2 + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + d0e58b78e8e84a0c8796e707b1f95f65 + 124 + 2 + + + 1.0 + Graph RAG is compared to source text summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + 215fcc6a3b5e452da123aa7f9ef0cbc9 + 125 + 2 + + + 1.0 + Low-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 0d0fc5d4ecb548079b28979186f19bf6 + 126 + 2 + + + 1.0 + Intermediate-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + e7d3fe0f87ff47f5a4c8d9572d27245a + 127 + 2 + + + 1.0 + High-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + 6f7165b558ae427ca14b2b16d1e8e204 + 128 + 2 + + + 1.0 + The Graph RAG approach involves a specific pipeline for processing and summarizing text + bc9e2c9e369c4108cf4f6dd5f60960f4 + 2ec093d2a76d45f88ec508e45ba8c6a3 + 129 + 2 + + + 1.0 + Techniques are specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 16d5a528d6374612b87a5656e8d95193 + 130 + 2 + + + 1.0 + Implementation details are specific configurations used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + 40293e74dbc643e8ab6546dff759ac7c + 131 + 2 + + + 2.0 + Graph RAG is a specific implementation of RAG systems + 922778ce1cb2fdd6dbab1746c8795620 + 1834b753dc7f4a8b98c2317a551b56ee + 132 + 2 + + + 2.0 + Graph RAG is a system that utilizes root-level community summaries, denoted as C0, to answer user queries. C0 represents these root-level community summaries within the Graph RAG analysis, serving as a foundational element in the system's ability to map out relationships and understand the structural dynamics within specialized communities. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + d9b127eab2f64e338d7adcd186786a45 + 133 + 2 + + + 1.0 + Graph RAG uses high-level community summaries (C1) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + a18f7c9f58ca49d6acf18e1ca69d3033 + 134 + 2 + + + 1.0 + Graph RAG uses intermediate-level community summaries (C2) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + f3c3dd44cf50495c81e362174991242e + 135 + 2 + + + 2.0 + Graph RAG utilizes low-level community summaries, represented by C3, to answer user queries. C3 plays a crucial role in the Graph RAG analysis by providing detailed summaries of community structures, which are essential for effectively addressing user inquiries. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + 86c2b3749a3c4342bbb3a8c70c3a76a0 + 136 + 2 + + + 2.0 + Graph RAG is a key entity in the analysis, serving both as a condition being compared and as a tool for comparing multiple conditions. This dual role highlights its significance in the study, where it functions not only as a subject of comparison but also as a methodological framework for evaluating other conditions. The analysis likely involves a detailed examination of various conditions, with Graph RAG playing a central role in facilitating these comparisons. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + 571f65acb3134490932feeb91b01cca3 + 137 + 2 + + + 1.0 + Graph RAG uses different levels of graph communities to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + d3faf86c153f440eaa410305b3dc6617 + 138 + 2 + + + 1.0 + The Graph RAG mechanism uses an LLM evaluator for head-to-head comparison + 322e02986c8724eedbcf3ebfa20b989c + f85786004b0540349192d2ca05b15264 + 139 + 2 + + + 1.0 + Graph RAG is a multi-stage mechanism + 322e02986c8724eedbcf3ebfa20b989c + cf56bfc9fa7d47fe9cb553dd09f2b412 + 140 + 2 + + + 1.0 + Graph RAG mentions Taylor Swift as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + a077dbcd38b644f6929cf05272c2fb9d + 141 + 2 + + + 1.0 + Graph RAG mentions Travis Kelce as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + d2659a32b9de406eb750a35d078c9774 + 142 + 2 + + + 1.0 + Graph RAG mentions Britney Spears as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + 0b26876307ad4cc48839b61a21a1d03a + 143 + 2 + + + 1.0 + Graph RAG mentions Justin Timberlake as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + c68e6c694a554256846d12178ddb12dc + 144 + 2 + + + 1.0 + Graph RAG is determined to be the winner based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + ff25ce2e8ace4bdcb765c863b483852b + 145 + 2 + + + 1.0 + Answer 1 was generated using the Graph RAG method, which provides a comprehensive and structured overview of public figures across various sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + 757e402cb7ee4601ac1bc8c4fafb5207 + 146 + 2 + + + 1.0 + Graph RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + 62e8f5f04cd04384b246291cef3a9e4d + 147 + 2 + + + 1.0 + Graph RAG is compared with source texts for answer comprehensiveness and diversity + 6f33a085ff3304e5994f7fbb86c881a4 + c04abbd5e59b4c64b023908f6db05498 + 148 + 2 + + + 1.0 + TS represents source text summarization in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 6bb9bed2e39c4e31a81f12479af3d16c + 149 + 2 + + + 1.0 + Root-level summaries are used in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 26c926c6016d4639b05427f01ba629f5 + 150 + 2 + + + 1.0 + Answer comprehensiveness is used to evaluate the performance of Graph RAG + 6f33a085ff3304e5994f7fbb86c881a4 + 8f6872eeb81b432b91405d327636113c + 151 + 2 + + + 1.0 + Win rate is used to measure the success rate of Graph RAG in providing comprehensive and diverse answers + 6f33a085ff3304e5994f7fbb86c881a4 + ac80a99fda2b488285d29596dd4d1471 + 152 + 2 + + + 1.0 + Element extraction prompts are used in Graph RAG to retain specific details in the index + 6f33a085ff3304e5994f7fbb86c881a4 + 67d6a3481e4b419292247cef5cd5b737 + 153 + 2 + + + 2.0 + Graph RAG incorporates the concept of self-memory + f35de4d9fb65f1d5a392064b20545c19 + 904cd052ec194654bb72f4027e43daa3 + 154 + 2 + + + 2.0 + Graph RAG incorporates the concept of iterative retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + 7e88fd2e835147fbb71866612735e8d4 + 155 + 2 + + + 2.0 + Graph RAG incorporates the concept of federated retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + 029d1a8c3b184aa5bb21228f40cd12fd + 156 + 2 + + + 2.0 + Graph RAG incorporates concepts used in multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + a1ebc53a0bc74a0eb6dbdd18cf3c88cd + 157 + 2 + + + 2.0 + Graph RAG incorporates concepts used in multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + a51d063ad4c744049edb359eb88407b7 + 158 + 2 + + + 2.0 + Graph RAG uses a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + b67268f90338474e8e53b9a6715b6833 + 159 + 2 + + + 2.0 + Graph RAG incorporates the concept of a tree of clarifications + f35de4d9fb65f1d5a392064b20545c19 + acb53370e72b4430a752d9ea18c17352 + 160 + 2 + + + 3.0 + Graph RAG utilizes a self-generated graph index. This self-generated graph index is a crucial component of Graph RAG, enabling it to efficiently manage and retrieve information within its graph-based framework. The use of a self-generated graph index suggests that Graph RAG has an inherent capability to construct and maintain its indexing structure, which likely enhances its performance and adaptability in handling complex data relationships. + e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + a8738c7de11543df930169741381c252 + 161 + 2 + + + 2.0 + Graph RAG incorporates concepts from Gao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 3f8b5b2727924ba0b62e6286063b6861 + 162 + 2 + + + 2.0 + Graph RAG incorporates concepts from Cheng et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + bb5010633113442eaf814852995cfa22 + 163 + 2 + + + 2.0 + Graph RAG incorporates concepts from Mao et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + 9eb8c635538243a690366f8bc1de34e0 + 164 + 2 + + + 2.0 + Graph RAG incorporates concepts from Shao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 050c5b770d51409cb40f9c52f02d1329 + 165 + 2 + + + 2.0 + Graph RAG incorporates concepts from Wang et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + 9e12f514d26d48dfab65807568a6cff9 + 166 + 2 + + + 2.0 + Graph RAG incorporates concepts from Su et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + 97f98b1623104f48aa93196a1f7dede2 + 167 + 2 + + + 2.0 + Graph RAG incorporates concepts from Feng et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 87718ef799a34104b6ef9c2df6621cbc + 168 + 2 + + + 2.0 + Graph RAG incorporates concepts from Trivedi et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + f64e87431d674f298c533f6878458b95 + 169 + 2 + + + 2.0 + Graph RAG incorporates concepts from Khattab et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + e6d44d0db58f42799a02eacbd6b14543 + 170 + 2 + + + 2.0 + Graph RAG incorporates concepts from Sarthi et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + 64961fbc3a1641378be10bcb3b0955e1 + 171 + 2 + + + 2.0 + Graph RAG incorporates concepts from Kim et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + 59bcc2ec512c4c1ba44272446b419230 + 172 + 2 + + + 2.0 + Graph RAG generates community answers in parallel + f35de4d9fb65f1d5a392064b20545c19 + 8f39ae56f8b54b1b94faf04dbd0b9d11 + 173 + 2 + + + 1.0 + Graph RAG is compared to a graph-free approach for global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + f3018b934ac241639a33c925c24bc507 + 174 + 2 + + + 1.0 + Graph RAG is compared to map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + ddedfd5179e64700adced4803c75cdba + 175 + 2 + + + 1.0 + Graph RAG uses rich text annotations + e4d9b12cf2b4c691c74019eefff4fb39 + 07d501edd4614e1d9d08d01b702688a3 + 176 + 2 + + + 1.0 + Graph RAG uses a hierarchical community structure + e4d9b12cf2b4c691c74019eefff4fb39 + f745075dedcf444daa9370cf32403d31 + 177 + 2 + + + 1.0 + Graph RAG can operate using embedding-based matching + e4d9b12cf2b4c691c74019eefff4fb39 + 1ef48284d238405f94190125092a3e28 + 178 + 2 + + + 1.0 + Graph RAG can be part of hybrid RAG schemes + e4d9b12cf2b4c691c74019eefff4fb39 + 8806b817446447e3b50f5bc85ff497e1 + 179 + 2 + + + 1.0 + Graph RAG can employ map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + 7303ee20690449db8c168df3fe008bc5 + 180 + 2 + + + 1.0 + Graph RAG can extend operations across the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + 2f1c535a14b14758bf1cacca81c74878 + 181 + 2 + + + 1.0 + Alonso contributed to the work on Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + 3b78cc7ce8224afcab3e4bbe550cde10 + 182 + 2 + + + 1.0 + Graph RAG includes local graph RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + 29ec9dd9f5864170a7e75c46c11c0090 + 183 + 2 + + + 1.0 + Graph RAG uses an entity-based graph index + e4d9b12cf2b4c691c74019eefff4fb39 + 7893ee15f0e941cbacad8cc1feaacbaf + 184 + 2 + + + 2.0 + NebulaGraph launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs + 833e7d67dcd30790b26b71c9b5306f6b + f53397f743ca4d7397c0a694fe787da0 + 185 + 2 + + + 1.0 + Community summaries are used to generate partial responses + e8d83e6e7a7c0f57b218cef24976b745 + 0041db9da3694ad397f37c76f8477770 + 186 + 2 + + + 1.0 + Community summaries are created from graph communities + f0306814bf64f5c9e79603fc6a52f4ea + a7c2a64e06374091adce74adb36801ab + 187 + 2 + + + 2.0 + Community answers are created from community summaries. These answers are generated by synthesizing information from the summaries provided by the community, ensuring that the responses are comprehensive and reflective of the collective knowledge and insights within the community. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + 107568a67cac472c89dfce4bbe11157c + 188 + 2 + + + 1.0 + Domain-tailored summarization is used to create community summaries + f0306814bf64f5c9e79603fc6a52f4ea + 3d78aa9d14714ac189e4020f78b15d24 + 189 + 2 + + + 1.0 + Community descriptions are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + ce0366abadef410d9b65e2bfbbf0b0f9 + 190 + 2 + + + 1.0 + Partial answers are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + f4370806deb84d0eb7e85e742e7d4bbf + 191 + 2 + + + 1.0 + Community summaries are created for each level in the hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56 + c92392d168c2443e8ed7b04992d0c92b + 192 + 2 + + + 1.0 + Community summaries are useful for understanding the global structure and semantics of the dataset + 7fb7d9ce2da9c940a32afdd87d1d9e56 + b5800c807edd4087a2420007272d15d0 + 193 + 2 + + + 1.0 + Community summaries are used to answer global queries + 843fc5421e086120ffa1c75856ecf6cd + aa247540e90d4a7abc5bca6fafaaffa1 + 194 + 2 + + + 1.0 + Community summaries are generated from root communities + 843fc5421e086120ffa1c75856ecf6cd + 34537afa1e954e08bdb52ead3a49e2f3 + 195 + 2 + + + 1.0 + Community summaries are generated from sub-communities + 843fc5421e086120ffa1c75856ecf6cd + ae043af0299f4b32a98cf187efd2a5db + 196 + 2 + + + 1.0 + Community summaries are added to the LLM context window until the token limit is reached + 843fc5421e086120ffa1c75856ecf6cd + 6016863be3414d5a92397f2d45fdfd78 + 197 + 2 + + + 1.0 + Global answers are generated from community summaries + 843fc5421e086120ffa1c75856ecf6cd + a9b900821b8444d69f432da08a77539f + 198 + 2 + + + 1.0 + The level of summary detail affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + 1fee51d6f4614127a3e1cc80d018506e + 199 + 2 + + + 1.0 + The scope of information affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + 00dc2c0748214e52bc799ca3e25204e9 + 200 + 2 + + + 1.0 + Community summaries are used for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + 42d1a9e749ad40daa34c7b0b695f8751 + 201 + 2 + + + 2.0 + Community summaries are divided into chunks of pre-specified token size + 843fc5421e086120ffa1c75856ecf6cd + 20de9a1af6ab4e88acf003cb7be0217c + 202 + 2 + + + 1.0 + Summary detail and scope affect the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + f3229f10a5a54cb1b91a26ffa6ee77a3 + 203 + 2 + + + 1.0 + Community summaries are divided into chunks + 1d07b4248c2655081c7af0e373bd70c9 + 5154b4a4f3ac43729703c69fccb54633 + 204 + 2 + + + 1.0 + Community summaries are prepared to answer user queries + 1d07b4248c2655081c7af0e373bd70c9 + 2091070e709e45f5ae56d40a9da45520 + 205 + 2 + + + 1.0 + Intermediate answers are generated from community summaries + 1d07b4248c2655081c7af0e373bd70c9 + 09045ef5c4314dde9a631a206274563f + 206 + 2 + + + 1.0 + Community summaries are part of the graph community hierarchy + 36db32c37e1987e2c5863898ad882190 + 1b9baa98ede84164883e8cdcbc7000c1 + 207 + 2 + + + 1.0 + Community summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + e4f3fcc475a74756925b730caffcb70d + 208 + 2 + + + 1.0 + Community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 7f3d5282303f4fc3a009e04f7de0ad84 + 209 + 2 + + + 1.0 + Summaries of root-level communities are used in Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + 1219a14eaf5f49ab84c9287ebf58db7a + 210 + 2 + + + 1.0 + Global sensemaking questions are evaluated over datasets in the 1 million token range + e8d83e6e7a7c0f57b218cef24976b745 + efaa386bd5e9454b87e1851cd8b28ac3 + 211 + 2 + + + 1.0 + Global sensemaking questions are directed at an entire text corpus + e8d83e6e7a7c0f57b218cef24976b745 + 073241be9b6a4952ad01dd14b94fb89c + 212 + 2 + + + 1.0 + The Python-based implementation of Graph RAG approaches will be available at this URL + e4d9b12cf2b4c691c74019eefff4fb39 + f7ac6bc4a9ca4250ad29a3adb5d08657 + 213 + 2 + + + 1.0 + Query-focused summarization is used to produce the global answer + f0306814bf64f5c9e79603fc6a52f4ea + ac2ee54e75a2492c8db372dadfccd083 + 214 + 2 + + + 1.0 + Map-reduce is used for query-focused summarization of an entire corpus + 21e52bc06a82796b1f4bcd73edda1f2a + ee895ad0b8cd40c29465e8527748d847 + 215 + 2 + + + 1.0 + Query-focused summarization is used for answering global queries + 7fb7d9ce2da9c940a32afdd87d1d9e56 + fe38c996c2d64bc899eabd6389034075 + 216 + 2 + + + 1.0 + An entity knowledge graph is derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + c0e28ae832c94405b8ddd4d2ad978be5 + 217 + 2 + + + 2.0 + In the domain of Natural Language Processing and Information Retrieval, "SOURCE DOCUMENTS" and "TEXT CHUNKS" are closely related entities. Text chunks are extracted from source documents, specifically for processing in the Graph RAG (Retrieval-Augmented Generation) approach. This method involves breaking down the source documents into smaller, manageable pieces, or text chunks, which can then be effectively utilized in various computational processes, enhancing the efficiency and accuracy of information retrieval and generation tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 7a4573a19ef94e25b4480cb4d953ae7a + 218 + 2 + + + 1.0 + Intermediate-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + 05f6639803524537b67a7f2b0c66ad23 + 219 + 2 + + + 1.0 + Low-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + 21bfd14cbc1f4cbc8ac59f7fd8c75b31 + 220 + 2 + + + 1.0 + Document corpus consists of source documents being processed + bc9e2c9e369c4108cf4f6dd5f60960f4 + c19cf2d7b067421990ab9f3acec9e736 + 221 + 2 + + + 1.0 + Partial responses are summarized to generate a final response + e8d83e6e7a7c0f57b218cef24976b745 + 3e1981b9301c4d339a9228ae7a089a04 + 222 + 2 + + + 1.0 + The LLM evaluator assesses answers based on the comprehensiveness metric + 322e02986c8724eedbcf3ebfa20b989c + 0948efa844814529b4c023aacbc23d64 + 223 + 2 + + + 1.0 + Naive RAG is evaluated for comprehensiveness + e8c8f911135faf3ff35f24107eb3f99c + fcdc0cc5ff93453eb0b94b9254760999 + 224 + 2 + + + 1.0 + Comprehensiveness is a metric used to determine the decision + e8c8f911135faf3ff35f24107eb3f99c + 0ec4ad4398a8457ab3d71bd2561858dc + 225 + 2 + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + 3c06988555334a389eab093f98679e85 + 226 + 2 + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + 81ceb8db419b4697ad24e9d7f46422ff + 227 + 2 + + + 1.0 + The smallest context window size (8k) was universally better for comprehensiveness + 4c855404ee3d3c94aa2136f1513c666f + fd05d8198d0947b39b8fa1b16f3ecf5f + 228 + 2 + + + 1.0 + The final evaluation prioritized comprehensiveness in answers + 4c855404ee3d3c94aa2136f1513c666f + d984f08ad62f47ab9aabb9aeec1b245e + 229 + 2 + + + 1.0 + Global approaches achieved higher comprehensiveness win rates + 36db32c37e1987e2c5863898ad882190 + 43603c7868164ac38c659bce7a77f45a + 230 + 2 + + + 1.0 + The LLM evaluator assesses answers based on the diversity metric + 322e02986c8724eedbcf3ebfa20b989c + 54a20cc6062d4b7193d023b6ff20461f + 231 + 2 + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + 6bb190069a704ccca3d8e1648a384185 + 232 + 2 + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + 47d2036509bf408095ab440bd052ac24 + 233 + 2 + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on diversity + 4c855404ee3d3c94aa2136f1513c666f + c20e6b1418a140389c31c7b71a6eba0c + 234 + 2 + + + 1.0 + The final evaluation prioritized diversity in answers + 4c855404ee3d3c94aa2136f1513c666f + ad96e5294247465a9c7d5ea8161dc305 + 235 + 2 + + + 1.0 + Global approaches achieved higher diversity win rates + 36db32c37e1987e2c5863898ad882190 + 25c968bf5a4f48369fded6c260f71540 + 236 + 2 + + + 1.0 + Human endeavors rely on sensemaking to understand and reason about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + 269b441634a144219f539202309bc9fb + 237 + 2 + + + 1.0 + Human endeavors rely on analyzing document collections for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + d0baf2392635468db7f5657f89eb2024 + 238 + 2 + + + 1.0 + LLMs are used to automate sensemaking in complex domains + f0306814bf64f5c9e79603fc6a52f4ea + 4f29bcf5377d4c9f94ff3f8ca2f8d941 + 239 + 2 + + + 1.0 + Microsoft uses LLMs for automating sensemaking in scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + e7072a582d9b4c1ea8b171ee940d4d6e + 240 + 2 + + + 1.0 + Ranade uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + cb6fcf84e3d04ef59b01f97ac94823a1 + 241 + 2 + + + 1.0 + Joshi uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + 97a21db5f5954e2c8868b298a3f0090e + 242 + 2 + + + 1.0 + LLM prompts are used to tailor the responses of large language models + f0306814bf64f5c9e79603fc6a52f4ea + c8f3e6cadcf34c8fafe8987e4a9b66f8 + 243 + 2 + + + 1.0 + Ranade and Joshi discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + 26c9c44e5059429bb8abc3308bc6c814 + 244 + 2 + + + 2.0 + GPT is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + 7cea9903153f43b895c0b23d25bc90a3 + 245 + 2 + + + 2.0 + Llama is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + b54436ccc23745c88d24edcc3fdd8ed1 + 246 + 2 + + + 2.0 + Gemini is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + 977c895bb98d4136a76e8749533154b6 + 247 + 2 + + + 2.0 + Kuratov et al., 2024, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + 8d75cfea884248aba1f372de5e1b82a9 + 248 + 2 + + + 2.0 + Liu et al., 2023, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + 90f4ee186bcd4996ad8002888569fffc + 249 + 2 + + + 1.0 + Sensemaking is applied in the domain of scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + 4bb78401581b4240b0967309e96af00b + 250 + 2 + + + 1.0 + Sensemaking is applied in the domain of intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + a18dd9ea4143411cb32e261db056cf0c + 251 + 2 + + + 1.0 + Klein defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + cd8d9795f540413390927ea2a9e77c26 + 252 + 2 + + + 1.0 + Klein et al. defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + 2917f3b478b04ffcacd4b47602f4d0f5 + 253 + 2 + + + 2.0 + Element instances are extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + 3984bd063b384901862e68506c77cc68 + 254 + 2 + + + 1.0 + Entity references are extracted from text chunks during processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + 4137a2c7dd884bc2a8469b7fa937346c + 255 + 2 + + + 1.0 + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + 60b6bf585ccc477d830d4b69b8c7b62a + 256 + 2 + + + 2.0 + Element instances are converted into element summaries by the LLM (Large Language Model). Element summaries are created from element instances, indicating a transformation process facilitated by the LLM. This process involves the LLM taking detailed element instances and generating concise element summaries, which encapsulate the essential information from the instances. + 2c6ed90897310eea2f28e33fff1c32b0,f0306814bf64f5c9e79603fc6a52f4ea + 4330f73cb78a4bb39a384eb29112201b + 257 + 2 + + + 1.0 + Covariates are additional attributes associated with extracted element instances + bc9e2c9e369c4108cf4f6dd5f60960f4 + 45c4ed77967746e485ec9e52c0dcc0d2 + 258 + 2 + + + 1.0 + Domain-tailored summarization is used to create element summaries + f0306814bf64f5c9e79603fc6a52f4ea + 17c2cc25d00347c3bf2422d4f7a4ad7e + 259 + 2 + + + 1.0 + Element summaries include descriptions of entity nodes + 2c6ed90897310eea2f28e33fff1c32b0 + 0057fb2ddc0e4088ae5099b7ffa137da + 260 + 2 + + + 1.0 + Element summaries include descriptions of relationship edges + 2c6ed90897310eea2f28e33fff1c32b0 + d67d67cc3698438db76eb4a7f75e1ea0 + 261 + 2 + + + 1.0 + Element summaries include descriptions of claim covariates + 2c6ed90897310eea2f28e33fff1c32b0 + c23761290af24cf29adc1ee8644bdad0 + 262 + 2 + + + 1.0 + Element summaries are used to understand the structure and semantics of graph communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + de51b828ce1f442bbb19a7b20bce9dda + 263 + 2 + + + 1.0 + Element summaries include descriptions of nodes + 843fc5421e086120ffa1c75856ecf6cd + 4a3ff6a3471945fd8c7fd5c171c56d56 + 264 + 2 + + + 1.0 + Element summaries include descriptions of edges + 843fc5421e086120ffa1c75856ecf6cd + 31bb84eb2a834dabacc0ed51af4fcefd + 265 + 2 + + + 1.0 + Element summaries include descriptions of covariates + 843fc5421e086120ffa1c75856ecf6cd + 5070012e83e7442381bcba1cdacdb7d8 + 266 + 2 + + + 1.0 + Sub-community summaries are used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + 5eda9074df124f5497f17b61badd52ac + 267 + 2 + + + 2.0 + Community detection is a technique used to identify graph communities. Graph communities are groups of nodes within a graph that are more densely connected to each other than to the rest of the graph. This process of identifying such communities is crucial for understanding the structural dynamics and relationships within complex networks, particularly in the domain of Natural Language Processing and Information Retrieval. By leveraging community detection algorithms, researchers can uncover hidden patterns and insights within large datasets, facilitating more effective data analysis and interpretation. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + 4cf4107b0e2842778aaa658a1a85f3b3 + 268 + 2 + + + 1.0 + Global answer is created from community answers + f0306814bf64f5c9e79603fc6a52f4ea + 7f4857f94b4e4e49be7236a42071e167 + 269 + 2 + + + 2.0 + Global answers are generated in response to user queries + 843fc5421e086120ffa1c75856ecf6cd + d21a1fef903f4a399bd3cd366aad3c9e + 270 + 2 + + + 1.0 + Global answer is generated by sorting intermediate answers based on helpfulness scores + 1d07b4248c2655081c7af0e373bd70c9 + fc596a598ff74a4c843e405b597551b5 + 271 + 2 + + + 1.0 + Intermediate answers are combined to form the global answer + 1d07b4248c2655081c7af0e373bd70c9 + e2aacff6b4404574b818e7a3ece57b5b + 272 + 2 + + + 1.0 + The final context window is used to generate the global answer + 1d07b4248c2655081c7af0e373bd70c9 + 2ec5cae98c7a485881f0680fbca6d67f + 273 + 2 + + + 1.0 + Graph RAG pipeline operates at indexing time + f0306814bf64f5c9e79603fc6a52f4ea + c87b815d61af448596d3194a804b57b3 + 274 + 2 + + + 1.0 + Graph RAG pipeline operates at query time + f0306814bf64f5c9e79603fc6a52f4ea + 2f92fc82c3b74417896bad3bd8e61f5e + 275 + 2 + + + 1.0 + Nodes are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + fb61c68efe5b4d69a9623e531e7c639c + 276 + 2 + + + 1.0 + Edges are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + dc61e34c1ca8419e923aeeff7d83d949 + 277 + 2 + + + 1.0 + Covariates are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + 697fb824eef34759852f1d5588921aec + 278 + 2 + + + 1.0 + Leiden method is used in the graph RAG pipeline for community detection + f0306814bf64f5c9e79603fc6a52f4ea + b872fcc5b18a4f32b976f4693f22e88e + 279 + 2 + + + 1.0 + Graph RAG pipeline uses the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + 64be9b98299f4d349e0f4358685ca235 + 280 + 2 + + + 1.0 + The Podcast dataset graph consists of 8564 nodes + 36db32c37e1987e2c5863898ad882190 + 8302a03f6ede471bb955c0bbf44a4b3c + 281 + 2 + + + 1.0 + The News dataset graph consists of 15754 nodes + 36db32c37e1987e2c5863898ad882190 + a02263dd89964a1c8ab2d0e9aba0f4eb + 282 + 2 + + + 1.0 + The Podcast dataset graph consists of 20691 edges + 36db32c37e1987e2c5863898ad882190 + 6b7aa6ce4cac4edbaaab831286e67e5e + 283 + 2 + + + 1.0 + The News dataset graph consists of 19520 edges + 36db32c37e1987e2c5863898ad882190 + 655d40ea08e348ad94ae49785797da90 + 284 + 2 + + + 1.0 + Traag contributed to the development of the Leiden method + f0306814bf64f5c9e79603fc6a52f4ea + 254cea99330f4f2aa062c771146da7ea + 285 + 2 + + + 2.0 + Traag et al. are the authors of the Leiden algorithm and developed the Leiden method. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + a2836232227c4e3383d166db860cb2a3 + 286 + 2 + + + 2.0 + Leiden is a specific type of community detection algorithm used in various analytical pipelines. It is designed to identify and map out the structural dynamics within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. The algorithm is known for its efficiency and accuracy in detecting community structures, making it a valuable tool for researchers and practitioners in the field. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + 8a9247ee9bac45bdbf69c9d0bb8419b5 + 287 + 2 + + + 1.0 + Leiden is known for its ability to recover hierarchical community structures efficiently + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 757a0f78fcdd4bf6b8326a75fcee9e15 + 288 + 2 + + + 1.0 + The Leiden algorithm is used to detect graph communities in the MultiHop-RAG + 7fb7d9ce2da9c940a32afdd87d1d9e56 + b5235cb24b8f440389f250ebd5b6e2f8 + 289 + 2 + + + 1.0 + Figure 3 shows graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + bdee1849252749efa2e671ed87641f61 + 290 + 2 + + + 1.0 + Lewis contributed to the development of the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + 057641c1476247958d8c357e17095d8e + 291 + 2 + + + 1.0 + Lewis et al. developed the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + b61dfd0b24664f37af4046bdf0cb7b19 + 292 + 2 + + + 2.0 + Lewis et al., 2020, are the authors who established the RAG approach + fb3c48579608fa28be585ceb6cd2f0fe + 0bc00f14e6194df7b0fe9ef9ba28d34f + 293 + 2 + + + 1.0 + Kevin Scott is the CTO of Microsoft + 1d07b4248c2655081c7af0e373bd70c9 + b823c5d22037423da919eee6c35c4c8b + 294 + 2 + + + 2.0 + Microsoft conducted a study on the impact of large language models on scientific discovery using GPT-4 + 833e7d67dcd30790b26b71c9b5306f6b + cd7f555e4ab948ba94bade14e262ff84 + 295 + 2 + + + 1.0 + Preprint is available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 86cd53087b2542f898d6cecca31e6145 + 296 + 2 + + + 1.0 + Baumel, T. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 5dc3480806b04fdd8089a3be46e22540 + 297 + 2 + + + 1.0 + Eyal, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 50c91820a91f488d8606198540aba894 + 298 + 2 + + + 1.0 + Elhadad, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + a38eace89e7e40de8f007fde24597e9e + 299 + 2 + + + 1.0 + Es, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 5d75097d065e4b049a1678deab40949b + 300 + 2 + + + 1.0 + James, J. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + c277134d380a42cd886a14a953554792 + 301 + 2 + + + 1.0 + Espinosa-Anke, L. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + b680be879404440885b1d3af5b9af583 + 302 + 2 + + + 1.0 + Schockaert, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 4cc609b1a64a442aac6b72078a315ac6 + 303 + 2 + + + 1.0 + Feng, Z. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + a3ee323c9c9a4f81b5907030122b80d2 + 304 + 2 + + + 1.0 + Feng, X. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 19aa5f0b738c4f4a96668c80c3e93331 + 305 + 2 + + + 1.0 + Zhao, D. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + f8402b10349f4db888ac4fb6fd81723a + 306 + 2 + + + 1.0 + Yang, M. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 5927f9089289429da4adf2bbd5641e44 + 307 + 2 + + + 1.0 + Qin, B. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + 60724b8b268044b69a4b3d939f1757e2 + 308 + 2 + + + 1.0 + LangChain is an organization that has published on arXiv + 71f6daf11e64e5273a3847d46bf228e1 + d931685d35e149909472f736114ca62f + 309 + 2 + + + 1.0 + Wang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 3f5e9927a4114a958d75f5ed313526a8 + 310 + 2 + + + 1.0 + Khramtsova, E. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 4728bf0cb7564bbd85c90ceaa846f290 + 311 + 2 + + + 1.0 + Zhuang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + cdac6338c3234797a0d3a32cd68d1b2e + 312 + 2 + + + 1.0 + Zuccon, G. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 372f78df13f9452b84d898c703a1ba95 + 313 + 2 + + + 1.0 + Wang, Y. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 7af06d2b32a941a4b044579a7c423371 + 314 + 2 + + + 1.0 + Lipka, N. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + a10b8fad74744ae981747dadf7234b78 + 315 + 2 + + + 1.0 + Rossi, R. A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0cb2118ecc87439a91409deef7ef9830 + 316 + 2 + + + 1.0 + Siu, A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + ea27218042d640fd81c23eb64aff6b46 + 317 + 2 + + + 1.0 + Zhang, R. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 9e5d626681094933abf87cf797f2fa46 + 318 + 2 + + + 1.0 + Derr, T. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 545358ff14f84601a22e9f39f5ef1534 + 319 + 2 + + + 1.0 + Xu, Y. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 1b0e7dbc7c5944a7833f6540bde1fa4f + 320 + 2 + + + 1.0 + Lapata, M. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0c0f2d8c623949f1ae89c67d0753aeab + 321 + 2 + + + 1.0 + Zhang, J. published the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 20c3844c80a140ac97b62dc444feee41 + 322 + 2 + + + 1.0 + Zhang, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + c5fac1bea509464d9dc934275d938039 + 323 + 2 + + + 1.0 + Gan, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 45b64fbddd8f4abdb86a9c3c6f53f802 + 324 + 2 + + + 1.0 + Yao, L. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0e504b58cbda4d9188050bc43004c01f + 325 + 2 + + + 1.0 + Wang, C. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + c06bd37e120e4af49ec8bd6ce399473b + 326 + 2 + + + 1.0 + Zheng, L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 5d507985f2f540d8a1fa2d1191eae2a8 + 327 + 2 + + + 1.0 + Chiang, W.-L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 8e0b5b4011d74bbb8dc09fa05d88369c + 328 + 2 + + + 1.0 + Sheng, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 5d8184f5d52040d8bb67d1a6b889e9fe + 329 + 2 + + + 1.0 + Wu, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + b3bf669489ae4913bb60ddfe50e41697 + 330 + 2 + + + 1.0 + Zhuang, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 0eba9d55a3ff46298665a0c292e2237f + 331 + 2 + + + 1.0 + Lin, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 55e3f4a200eb4619ae2b6efb645464d1 + 332 + 2 + + + 1.0 + Li, D. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + c44324c171674d00a743413042e9b944 + 333 + 2 + + + 1.0 + Xing, E. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + 4bdaba79a3274241ab98e27aeaf98f57 + 334 + 2 + + + 1.0 + Preprint is classified under cs.CL on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + 7c8c464ed7044a7896adfeb35f58a04d + 335 + 2 + + + 1.0 + Preprint was submitted on 24 Apr 2024 + f0306814bf64f5c9e79603fc6a52f4ea + 5fa2eec73bec481b85eba22ea7a2a927 + 336 + 2 + + + 1.0 + Preprint has the identifier 2404.16130v1 on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + e6aa5eedca984c56b5fa5e179127951d + 337 + 2 + + + 1.0 + Community detection results in the partition of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 1c4bd4ba4ef64a93acd55faa8fd97ca9 + 338 + 2 + + + 1.0 + The pipeline includes a step for community detection + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 5b85c70d578c4d67b5cb4743552bd559 + 339 + 2 + + + 2.0 + Dang, 2006, is the author who established the QFS approach + fb3c48579608fa28be585ceb6cd2f0fe + 956113fb770840c38bce65bb5832f988 + 340 + 2 + + + 2.0 + Baumel et al., 2018, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + 785bb55e79954b0c84a4a53cd7f0b454 + 341 + 2 + + + 2.0 + Laskar et al., 2020, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + 1239281fd3774b91a99358c9c1e6ee1c + 342 + 2 + + + 2.0 + Yao et al., 2017, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + 32b29a842b224f4c99fa1d5c764efc9a + 343 + 2 + + + 2.0 + Goodwin et al., 2020, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + f5ae7dc11fd64822a3a15e7d3839031a + 344 + 2 + + + 2.0 + Laskar et al., 2022, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + e1e254e67719488894eaa3553112a8cf + 345 + 2 + + + 2.0 + Liu and Lapata, 2019, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + ebdd79169d7d41b99faf09b039a66204 + 346 + 2 + + + 2.0 + Achiam et al., 2023, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + e036534e17b24dd2895167a20873230f + 347 + 2 + + + 2.0 + Brown et al., 2020, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + a00bc5e4be634b08b1f084b6a07abafd + 348 + 2 + + + 2.0 + Touvron et al., 2023, are the authors who worked on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + ce8241c964724429bb361b7b53867007 + 349 + 2 + + + 2.0 + Anil et al., 2023, are the authors who worked on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + 61cd7f168f7f44d6a23415e9497f1e65 + 350 + 2 + + + 1.0 + Modularity is an inherent quality of knowledge graphs + 21e52bc06a82796b1f4bcd73edda1f2a + 3be77a7b57e34c55acc1f1dfbc64ee10 + 351 + 2 + + + 1.0 + Brown et al. (2020) discuss in-context learning with few-shot examples + bc9e2c9e369c4108cf4f6dd5f60960f4 + 751c564f8ff6444d9d4c8de4a677e655 + 352 + 2 + + + 1.0 + Kuratov et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + 96963c158fb64680bded290f442ff9aa + 353 + 2 + + + 1.0 + Liu et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + bdbfbde5dd244447a2a0674b30ae3e8f + 354 + 2 + + + 1.0 + Louvain is a type of community detection algorithm + 21e52bc06a82796b1f4bcd73edda1f2a + f970bfe31db74929abff6ea38e5d18e6 + 355 + 2 + + + 1.0 + Community detection algorithms are used to partition the graph index into communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 6f0c2a8b79e6406a8ab7a20864ae2ce2 + 356 + 2 + + + 1.0 + Fortunato has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 7b09e60e33f44ffdab9c656c5b9c1d50 + 357 + 2 + + + 1.0 + Jin et al. have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 8b7beab7c0a143aea7bffc31df7528d5 + 358 + 2 + + + 1.0 + HotPotQA dataset is used to evaluate the entity extraction prompt with gpt-4-turbo + 21e52bc06a82796b1f4bcd73edda1f2a + d03eb34a0612420680555ab9f10d03d5 + 359 + 2 + + + 1.0 + Yang et al. (2018) introduced the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + b066746cdff7440c8a3591f0c098201d + 360 + 2 + + + 2.0 + Yang et al. are the authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + 1e2eded8ef7b4b458c33fbc2d36c4380 + 361 + 2 + + + 1.0 + GPT-4-Turbo was tested with varying context window sizes + 4c855404ee3d3c94aa2136f1513c666f + c59e3e931b0f4cf888c2eb70857ee753 + 362 + 2 + + + 1.0 + Tech journalist uses podcast transcripts to look for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + 305b80bb4df5488b8a34129daeeae0c7 + 363 + 2 + + + 3.0 + Kevin Scott is a participant in the podcast conversations compiled in the Podcast Transcripts dataset. His conversations are included as part of the podcast transcripts, contributing to the overall content and discussions captured within this dataset. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + 66fa0de756da440bad8da583306410c4 + 364 + 2 + + + 1.0 + Technology leaders participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + d1e9c550a0e74c48ae81c319f26ccafc + 365 + 2 + + + 2.0 + RAG systems are used to evaluate the Podcast Transcripts dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + 3730b5d759ba4fd28a54af0a02151f09 + 366 + 2 + + + 2.0 + C0 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 82b7f7c27e2348f880c94ffb80942de7 + 367 + 2 + + + 2.0 + C1 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 0980c4f558654466b4d691d0cb7ce16d + 368 + 2 + + + 2.0 + C2 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + f1e47cf5daa441649c3474c3339bb704 + 369 + 2 + + + 2.0 + C3 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 0964dcfbff934c92af8961155673ac7f + 370 + 2 + + + 2.0 + TS is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + 51b82bcdffe04056bad1c082c3830047 + 371 + 2 + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + c62bb148852b49a98e2779ca23a0919d + 372 + 2 + + + 1.0 + SS is a category used in the analysis of podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + 72b5a0c357c24b739084d501b9354bc1 + 373 + 2 + + + 1.0 + Units are used to measure the context in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + c827b62ebf134e55a3ccf0b63f976870 + 374 + 2 + + + 1.0 + Tokens are used to measure the word count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + b51ef388758845e880e736309ae791e3 + 375 + 2 + + + 1.0 + % Max is used to measure the percentage of maximum token count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + 0a841cd4b6664423b033f22e3a80f33c + 376 + 2 + + + 1.0 + Both are datasets used in the analysis + 36db32c37e1987e2c5863898ad882190 + 16911c51c65b42f8a2d04c05f45b2c58 + 377 + 2 + + + 1.0 + Educator uses news articles to incorporate current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + fc3f77f29574410d991a2aa333950bf6 + 378 + 2 + + + 2.0 + RAG systems are used to evaluate the News Articles dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + 4f847eb72cbe48678d5634dcf93fc0e2 + 379 + 2 + + + 1.0 + C0 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + 829e64159ae04301982e88e93a2f0e49 + 380 + 2 + + + 1.0 + C1 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + cf37d3d4bc154f65b3d79c831c587763 + 381 + 2 + + + 1.0 + C2 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + 4b4fce341d554012bc73d7886860749e + 382 + 2 + + + 1.0 + C3 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + 9f6e7a08bd814d19b45fac58928027f8 + 383 + 2 + + + 1.0 + TS is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + ff9410fed5e64c04a875e040e3d182b2 + 384 + 2 + + + 1.0 + Units are used to measure the context in news articles + 36db32c37e1987e2c5863898ad882190 + 1161272728914953b568f384d7a9f2f1 + 385 + 2 + + + 1.0 + Tokens are used to measure the word count in news articles + 36db32c37e1987e2c5863898ad882190 + f09c82eb89944ae9846df82135123b90 + 386 + 2 + + + 1.0 + % Max is used to measure the percentage of maximum token count in news articles + 36db32c37e1987e2c5863898ad882190 + d221b743a51d464b87de3b72b85f6b59 + 387 + 2 + + + 1.0 + Map-reduce is the method used in the text summarization condition + 973164fa90bf2b4ee267f4fd795916bf + 9fd31a28e1384b40a9d1658a765871cd + 388 + 2 + + + 1.0 + The LLM evaluator assesses answers based on the empowerment metric + 322e02986c8724eedbcf3ebfa20b989c + 0119f233c8394b9584e55fadcce173f0 + 389 + 2 + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for news articles + ebf5249c888e07fedce6572a4c03f88c + 5c20b469b92446dabb1b68976807be7c + 390 + 2 + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on empowerment + 4c855404ee3d3c94aa2136f1513c666f + 2c2392247a35456da663adfcffd12e73 + 391 + 2 + + + 1.0 + Empowerment has an average win rate of 51.3% + 36db32c37e1987e2c5863898ad882190 + 167a32ff67ce4471baa8cf019ee7c17b + 392 + 2 + + + 1.0 + Naive RAG mentions Taylor Swift as a public figure + e8c8f911135faf3ff35f24107eb3f99c + 3280fc12ef414827838e6ac7089f0618 + 393 + 2 + + + 1.0 + Naive RAG mentions Travis Kelce as a public figure + e8c8f911135faf3ff35f24107eb3f99c + 556fba72a0854ce4831f6cfea6fd035e + 394 + 2 + + + 1.0 + Naive RAG mentions Britney Spears as a public figure + e8c8f911135faf3ff35f24107eb3f99c + 8e2e6eeed5a04c9f80efbcfc624ced95 + 395 + 2 + + + 1.0 + Naive RAG mentions Justin Timberlake as a public figure + e8c8f911135faf3ff35f24107eb3f99c + ea6d546f1caa4b4aaacdad8b8af195ec + 396 + 2 + + + 1.0 + Naive RAG is determined to be the loser based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + 267ce44e6dae43ee94d0d375ec08ef17 + 397 + 2 + + + 1.0 + Global approaches consistently outperformed the naive RAG + 36db32c37e1987e2c5863898ad882190 + b37e5d15f3154ee39df016b8eac8de66 + 398 + 2 + + + 1.0 + Naive RAG produces the most direct responses + 36db32c37e1987e2c5863898ad882190 + e13eb574e885414b80f0b66992767ef2 + 399 + 2 + + + 1.0 + SS represents naive RAG in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 93f4140f654e41ccba908c6f6dc65f17 + 400 + 2 + + + 1.0 + Gao et al., 2023 discusses naive RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + a102d091986749ef90b45d411e707bef + 401 + 2 + + + 1.0 + Community partitions enable divide-and-conquer global summarization + 7fb7d9ce2da9c940a32afdd87d1d9e56 + cd6ae38a5a6742899d14f4a064f42c19 + 402 + 2 + + + 1.0 + Global summarization can be performed using a graph-free approach + e4d9b12cf2b4c691c74019eefff4fb39 + fe18688bd4ef44d1a184ec6d1451a5cf + 403 + 2 + + + 1.0 + Source texts are used in global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + 0f1282bdfedb4f6e8765007a90dd2959 + 404 + 2 + + + 1.0 + Final global answer is generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + 540af5c5d4cd41ceb29c40c5fb02e2fe + 405 + 2 + + + 1.0 + Short descriptions are used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + bbf83708095f47019eaee93d6879bc77 + 406 + 2 + + + 1.0 + Low-level community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 245a56f01d1b48a7b4d88ed0e354155a + 407 + 2 + + + 1.0 + The use of rich descriptive text for homogeneous nodes in a graph index aligns with the capabilities of LLMs + 7fb7d9ce2da9c940a32afdd87d1d9e56 + d3aa564fb4eb430a8ca6813ca76bfff6 + 408 + 2 + + + 1.0 + Graph indexes differ from typical knowledge graphs in their use of rich descriptive text instead of concise knowledge triples + 7fb7d9ce2da9c940a32afdd87d1d9e56 + d9b948357d96419ca135065ce1c360ef + 409 + 2 + + + 1.0 + The graph index supports condition C0 + 973164fa90bf2b4ee267f4fd795916bf + 20a79ddd91ba48e4bb7bc194c79baaf6 + 410 + 2 + + + 1.0 + The graph index supports condition C1 + 973164fa90bf2b4ee267f4fd795916bf + b95728a0b96b405cbccafa6c12fd8722 + 411 + 2 + + + 1.0 + The graph index supports condition C2 + 973164fa90bf2b4ee267f4fd795916bf + 5d6dc034d2014e8c930fde69c31b99cf + 412 + 2 + + + 1.0 + The graph index supports condition C3 + 973164fa90bf2b4ee267f4fd795916bf + 127cbb53940f4efa8e1807b4452375ba + 413 + 2 + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + f1ea6ef9539043ab887bcce22ccf9625 + 414 + 2 + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + b50c4f053f0546029c4095b7b93aa05e + 415 + 2 + + + 1.0 + The graph index was created using generic prompts for entity and relationship extraction + 973164fa90bf2b4ee267f4fd795916bf + 0cea7f7a7fab49339cdd6fb02d0d183e + 416 + 2 + + + 1.0 + Few-shot examples tailored to the domain of the data were used in the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + 5b89f0d8101c419b86e1959cca2db848 + 417 + 2 + + + 1.0 + The graph indexing process used a context window size of 600 tokens + 973164fa90bf2b4ee267f4fd795916bf + cdb407fc600b45caa6f94f82e89d2e4f + 418 + 2 + + + 1.0 + The decision to build a graph index depends on the expected number of lifetime queries per dataset + e4d9b12cf2b4c691c74019eefff4fb39 + 7f4905fcb43e4d6ca23e6d2b40f6958e + 419 + 2 + + + 1.0 + The decision to build a graph index depends on the value obtained from it + e4d9b12cf2b4c691c74019eefff4fb39 + f5ad4fe84df544c69db25f0e30c6eace + 420 + 2 + + + 1.0 + The decision to build a graph index depends on the value obtained from other graph-related RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + 237a46cc973b41dc9af4190c71c5c9e1 + 421 + 2 + + + 1.0 + Recall measures the completeness of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + aaa27aa0b1024e3aa3c87a6ec821a348 + 422 + 2 + + + 1.0 + Precision measures the accuracy of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + f0a28fe3f68546dba7850815f7933275 + 423 + 2 + + + 1.0 + Few-shot examples are used to tailor the default prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + 45b59feba7134bc18632cb42530c189a + 424 + 2 + + + 1.0 + Few-shot examples are used to tailor the secondary extraction prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + 7747cd2048f94d378e83265b9561d921 + 425 + 2 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of science + 2c6ed90897310eea2f28e33fff1c32b0 + c4e9532dbc734264a0e3e827bc8014c6 + 426 + 2 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of medicine + 2c6ed90897310eea2f28e33fff1c32b0 + 003e5d505a01434596c6d65ff20b0bdf + 427 + 2 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of law + 2c6ed90897310eea2f28e33fff1c32b0 + f79358f3535045d9aad3b828df59293b + 428 + 2 + + + 1.0 + A single extraction round is part of the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + 7d375c18c1e2415faecd9f7397068a32 + 429 + 2 + + + 1.0 + Domain refers to the specific area of knowledge of the document corpus + bc9e2c9e369c4108cf4f6dd5f60960f4 + dfa0e847a6704c93a0fe014b01858ff7 + 430 + 2 + + + 1.0 + Covariate prompts are used to extract claims linked to detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + 9e91823feb174cd1b6a3bf8d0a5cb86b + 431 + 2 + + + 1.0 + Source text span is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + ad76c8dc8dd94412a5e79005cf8e0f2f + 432 + 2 + + + 1.0 + Start date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 26a03482961e41918ea049018080af7a + 433 + 2 + + + 1.0 + End date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 1cfd220ff4d2493ca4b92d725d171d32 + 434 + 2 + + + 1.0 + Description is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 97738fe0830d405ba53598b5cb1e5e38 + 435 + 2 + + + 1.0 + Subject is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 354cea4f6e164a48ad12122c28a5b30d + 436 + 2 + + + 1.0 + Object is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + 1ee2380c1eda4ebb8c9304820750ac88 + 437 + 2 + + + 1.0 + Communities of entities help manage variations in a noisy graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + 57e00d4d4e0e4679a150f048deb80af3 + 438 + 2 + + + 1.0 + Common entities are described using rich descriptive text for homogeneous nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + c1e4a9dbe55c4fb89f0d927c9fb067a4 + 439 + 2 + + + 1.0 + LLMs are used to generate metrics for evaluating natural language generation + 973164fa90bf2b4ee267f4fd795916bf + 1474a72a5cff4b72ae6f99e804ceaa95 + 440 + 2 + + + 1.0 + Wang et al. (2023) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + 738fda68df7a49a0bae96673a8711afc + 441 + 2 + + + 1.0 + Zheng et al. (2024) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + 89dd5a0943c64247adae624abbc95afb + 442 + 2 + + + 1.0 + Relationship edges connect homogeneous nodes in a graph + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 405e9907440d4deab71f3960ae36f47b + 443 + 2 + + + 1.0 + Edge weights represent the normalized counts of detected relationship instances on relationship edges + 7fb7d9ce2da9c940a32afdd87d1d9e56 + f91e7c9600ca4623a8cc4a56d2dccd07 + 444 + 2 + + + 1.0 + Each level of the hierarchical community structure provides a community partition + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 3af2a8619c394be6adf06e4bc742b7ec + 445 + 2 + + + 2.0 + The hierarchical community structure is a framework used to organize and understand the relationships and dynamics within specialized communities. Root communities are an integral part of this structure, serving as the top-level communities. These root communities form the foundational layer in the hierarchical community structure, providing a basis for further subdivision and organization of more specific sub-communities. This hierarchical approach allows for a systematic analysis of complex networks, facilitating a deeper understanding of the interconnections and dependencies within the domain of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + c10ffc51dcb54708a1dc757693010bfe + 446 + 2 + + + 2.0 + The hierarchical community structure is a framework that organizes communities into different levels, with sub-communities representing the lower-level communities within this structure. Sub-communities are integral components of the hierarchical community structure, indicating that they are nested within larger communities and contribute to the overall organization and dynamics of the community. This hierarchical arrangement allows for a more detailed and nuanced understanding of the relationships and interactions within the community, facilitating more effective analysis and mapping of complex text data, particularly in specialized domains such as Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + e67ce34d48364422973ccf3a6b57af83 + 447 + 2 + + + 1.0 + Community levels are part of the hierarchical community structure + 843fc5421e086120ffa1c75856ecf6cd + 98773a34c9bb474d8a789ea08f57250e + 448 + 2 + + + 1.0 + The Leiden algorithm is used to detect communities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + ae260498423e4d55aa413423cd0eb20b + 449 + 2 + + + 1.0 + OpenORD is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 4aeecb9d885743ca9373337a43957dd8 + 450 + 2 + + + 1.0 + Force Atlas 2 is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 1121b50f7858427fa679d81861238825 + 451 + 2 + + + 1.0 + Nodes represent entities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 6e3c8aa3abab475bb0148faa9112f0bf + 452 + 2 + + + 1.0 + Edges represent connections between nodes in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 948a00e8ee1246cc90c47b292d03ddff + 453 + 2 + + + 1.0 + Covariates are variables linked to nodes and edges in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + 45c42e619f5e488f914608780dcf0579 + 454 + 2 + + + 2.0 + Tang and Yang are the authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + 2b3bea0d9ede41f193828526bcb8e02c + 455 + 2 + + + 1.0 + Questions are generated based on the target datasets + 1d07b4248c2655081c7af0e373bd70c9 + 6b2586cc1f8e4dc8af64913af63d9837 + 456 + 2 + + + 1.0 + N represents the number of test questions per dataset + 973164fa90bf2b4ee267f4fd795916bf + 7983bfa8d173414685272b3844d6612e + 457 + 2 + + + 1.0 + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + 09294e8220a445e288ea8841f234a440 + 458 + 2 + + + 1.0 + Root communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + d4e043cf972c4d129b6b855f1731caae + 459 + 2 + + + 1.0 + Level 0 represents the root-level communities in the hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + e0d63137270c426dbbfe7fcf78c474de + 460 + 2 + + + 1.0 + Reports provide detailed information about specific subtopics within sub-communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + c50bca18bc454a98b935df012b7fd6f9 + 461 + 2 + + + 1.0 + Sub-communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + 434b133c64bd46219e67c6eb296ad0ff + 462 + 2 + + + 1.0 + Level 1 represents sub-communities within the root-level communities + 843fc5421e086120ffa1c75856ecf6cd + cb895bf7e7c147e6b5d923b6c8f67d63 + 463 + 2 + + + 1.0 + Partitions can be organized into a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 87776e869a01402499a317cb9cf09453 + 464 + 2 + + + 1.0 + Level 0 is the root level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + d1e5359d2e344260bf1b83823df839b7 + 465 + 2 + + + 1.0 + Level 1 is a sub-level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + 0522f6580b824bc39792b695fc8be66b + 466 + 2 + + + 1.0 + The token limit defines the maximum number of tokens in the LLM context window + 843fc5421e086120ffa1c75856ecf6cd + 580fd6d19460460fa40613f66b3ee200 + 467 + 2 + + + 1.0 + Prominence is used to prioritize community edges + 843fc5421e086120ffa1c75856ecf6cd + 84f4684a7a5241c18bb087ccb00550d3 + 468 + 2 + + + 1.0 + Combined source and target node degree is used to measure prominence + 843fc5421e086120ffa1c75856ecf6cd + 9607ba4a796f46be8d4f79bc7065d60b + 469 + 2 + + + 1.0 + Chunks are divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + 236dd7dce9ee4cf5918fddd44b4863e5 + 470 + 2 + + + 1.0 + Helpfulness scores are assigned to intermediate answers + 1d07b4248c2655081c7af0e373bd70c9 + 9e92fed814a64d9d88bfab9a227859d3 + 471 + 2 + + + 1.0 + Tech journalist is interested in episodes dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + 7dccecb29d3a419093b279b22e207539 + 472 + 2 + + + 1.0 + Tech journalist is interested in how guests perceive the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + 89857eb61e63461cbad7c5014f5098f9 + 473 + 2 + + + 1.0 + Tech journalist is interested in discussions about the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + 7b2e7a0d910c4988a7b64489f4159a65 + 474 + 2 + + + 1.0 + Tech journalist is interested in suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + 38630cf0996f4cff8d32b2dbdaa5ba85 + 475 + 2 + + + 1.0 + Tech journalist is interested in discussions about collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + bd0fb68ac7014b91a314c93ec55897f5 + 476 + 2 + + + 1.0 + Educator is interested in current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + f24dcb3cd6d644f8af2b6c47983e280b + 477 + 2 + + + 1.0 + Educator is interested in how news articles address the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + b1cad695afbc4ec3bbcd46ea34bd26ca + 478 + 2 + + + 1.0 + Educator is interested in examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + 72f7974758d74e5d89ddb64ad739abb8 + 479 + 2 + + + 1.0 + Educator is interested in insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + e6ee83249adf4e14b98d1676b1c6b05f + 480 + 2 + + + 1.0 + Educator is interested in highlighting the importance of health literacy through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + f805fd9fe42947a38b92a3db6e8cc986 + 481 + 2 + + + 1.0 + The size of the context window and the prompts used for answer generation are the same across all conditions + 973164fa90bf2b4ee267f4fd795916bf + e8b956218d5c4e5d9d390abcf527a514 + 482 + 2 + + + 1.0 + The task is an activity or goal that the user aims to achieve + 1d07b4248c2655081c7af0e373bd70c9 + 9525aa223d774e62ad856c2201cfab1b + 483 + 2 + + + 1.0 + Questions are generated based on the user's task + 1d07b4248c2655081c7af0e373bd70c9 + 1087596b06d1400a8f863d0ac1af64a4 + 484 + 2 + + + 1.0 + Datasets were used in combination with questions for the analysis + 4c855404ee3d3c94aa2136f1513c666f + 39058965295643c8a7738350cc18ceac + 485 + 2 + + + 1.0 + Questions were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + 9a8a2e5e3f2645619a0403532d935afe + 486 + 2 + + + 2.0 + Zheng et al. are the authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + f0c21c67baac47f097f74f5055b89877 + 487 + 2 + + + 1.0 + Zheng, L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 323a4c7407ac401db79a6023c3a5a17d + 488 + 2 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 686bc2bd59644e398dde88ffd37bf49b + 489 + 2 + + + 1.0 + Sheng, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + e368f8e9c9864acc880fdb5113631f3f + 490 + 2 + + + 1.0 + Zhuang, S. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 05063c19ddb847a89ae1746588464288 + 491 + 2 + + + 1.0 + Wu, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 019b34e800414f7b87f38a14adf2eb67 + 492 + 2 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 1064a663ca4742a78e743128546f6d87 + 493 + 2 + + + 1.0 + Lin, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 9a5e0a4ae34f46b39a5a028cbc135264 + 494 + 2 + + + 1.0 + Li, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 5a224002ecbc4725abeb5a424aaca6a6 + 495 + 2 + + + 1.0 + Li, D. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + 8826a17bbda34012b3ea84d58ae531eb + 496 + 2 + + + 1.0 + Xing, E. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + bab69d76defb402da2a2a358739f1497 + 497 + 2 + + + 1.0 + MT-Bench and Chatbot Arena are both tools used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + ea465e5cd92247829f52ff0c8591d1bb + 498 + 2 + + + 2.0 + Koesten et al. authored a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + 2dbac25b512c4f21965169a95a910a94 + 499 + 2 + + + 2.0 + Xu and Lapata authored a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + 97958ed004f645b1b331fa0e66faa313 + 500 + 2 + + + 1.0 + Text summarization method applies a map-reduce approach directly to source texts (TS) + 973164fa90bf2b4ee267f4fd795916bf + 48129b4ee99f4e30843fd4395d4815c0 + 501 + 2 + + + 1.0 + Text summarization is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + 6de4c00e48b3480883e696e24df9fda4 + 502 + 2 + + + 1.0 + Semantic search RAG is a na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached (SS) + 973164fa90bf2b4ee267f4fd795916bf + 4b3d236101de4904ab348e3e3b11b4be + 503 + 2 + + + 1.0 + Semantic search RAG is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + 5be2ce9957ba404f939b6c8175015619 + 504 + 2 + + + 1.0 + C0 uses root-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + fe77344850214c1cac923094de81098c + 505 + 2 + + + 1.0 + C0 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 798f739abfc14a13bf3911d0a9cfb63b + 506 + 2 + + + 1.0 + C0 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 3105de8188fd41d88d0dbf0a5d48e443 + 507 + 2 + + + 1.0 + C0 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + 8108dde0e62a48008a270138a690a0b9 + 508 + 2 + + + 1.0 + C1 uses high-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + d07207b853c14504a44eea1d4778f902 + 509 + 2 + + + 1.0 + C1 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 406822a1a01a4140baf9bbf1d479f07e + 510 + 2 + + + 1.0 + C1 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + fe47ba3762ae4feda39904d59cbb4160 + 511 + 2 + + + 1.0 + C1 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 53032c2afcb5474a88446ad7c5506980 + 512 + 2 + + + 1.0 + C1 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + 5c66a88612a245cb91fbba9c094f12fc + 513 + 2 + + + 1.0 + C2 uses intermediate-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + b4c54fb7ce0b4b77afd5fbe5a8a2527f + 514 + 2 + + + 1.0 + C2 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + c8b60cdb74104667b5d2b4b70d74d039 + 515 + 2 + + + 1.0 + C2 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + 333e294d7cc34df4abc47ad9ced3d186 + 516 + 2 + + + 1.0 + C2 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 15e66e10d12f4520abca20985d2cb39c + 517 + 2 + + + 1.0 + C2 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + 2a271d9b5d7b46fea4046d5590eed1d7 + 518 + 2 + + + 1.0 + C3 uses low-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + 99e372089bed4a0394af57175679f8e4 + 519 + 2 + + + 1.0 + C3 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 62afe93767684ea38f861d20fb05ff71 + 520 + 2 + + + 1.0 + C3 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + 8fc1fbff7e6c459c93ce2c2f5a62226e + 521 + 2 + + + 1.0 + C3 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 04b3ae04020349a9bc568f26d17eab14 + 522 + 2 + + + 1.0 + C3 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + bbc4d367c60f41ad8a279c12e5cc7da6 + 523 + 2 + + + 1.0 + TS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 9a1aff251eda416ea6270e6158e663fc + 524 + 2 + + + 1.0 + TS is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 7a9e50846c274338ab09e7313b540edb + 525 + 2 + + + 1.0 + TS is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + b268cc3ef860434ba663dd46af633cc5 + 526 + 2 + + + 1.0 + SS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + 1c9f67904a4c4fcc8cdac6a605900248 + 527 + 2 + + + 1.0 + The graph indexing process used 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + e00c403d1dc84ba6a37ee193596e320f + 528 + 2 + + + 1.0 + A graph was created for the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 3f2e726c3b624fe7bf11de9be2c0457e + 529 + 2 + + + 1.0 + Units are used to measure the context in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + f71dc0c394f04771af7e2ed37f85647e + 530 + 2 + + + 1.0 + Tokens are used to measure the word count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 2fea9c1856e54a91b79a9ce85755fbf5 + 531 + 2 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + 7823b4c5b3364c5f890d05f33a46bdde + 532 + 2 + + + 1.0 + Intermediate-level summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + 183f3a0b73ff41c5bb4a19fd7adf0c1d + 533 + 2 + + + 1.0 + The graph indexing process used 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + 392e06f17d724484a9cfb85fe69aac50 + 534 + 2 + + + 1.0 + A graph was created for the News dataset + 36db32c37e1987e2c5863898ad882190 + 6f49e00cdac04a358173ecd40351ee00 + 535 + 2 + + + 1.0 + Units are used to measure the context in the News dataset + 36db32c37e1987e2c5863898ad882190 + 3fef96af4ec343da8c34f8b09518de8a + 536 + 2 + + + 1.0 + Tokens are used to measure the word count in the News dataset + 36db32c37e1987e2c5863898ad882190 + bd403eff654e42c997e5656a2b1c1a20 + 537 + 2 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the News dataset + 36db32c37e1987e2c5863898ad882190 + 5763d829837144f199fac2b490b38110 + 538 + 2 + + + 1.0 + Datasets were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + 234c6f1859f0405ab607f0be53e7b06c + 539 + 2 + + + 1.0 + Wang et al., 2023a discusses the state-of-the-art results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + 21800eab85b94d4880bcada7a60763e5 + 540 + 2 + + + 1.0 + Zheng et al., 2024 discusses the competitive results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + b8bb28a7a9624b6d805be89adfe29eb5 + 541 + 2 + + + 1.0 + Zheng et al., 2024 discusses the LLM-as-a-judge method + 322e02986c8724eedbcf3ebfa20b989c + 61f26f8850504d56a6b7cd764c33299d + 542 + 2 + + + 1.0 + Embedding-based matching is used to match user queries + e4d9b12cf2b4c691c74019eefff4fb39 + d4456fac0ada4b6fbe3cfee873403d00 + 543 + 2 + + + 1.0 + Query-time LLM use was evaluated with different context window sizes + 4c855404ee3d3c94aa2136f1513c666f + f8fd3fcf650b47b2b1692506ebe77762 + 544 + 2 + + + 2.0 + The **CONTEXT WINDOW SIZE** and **FINAL EVALUATION** are closely related in the given data. A fixed context window size of 8k tokens was used for the final evaluation. This indicates that during the final evaluation phase, the system or model was configured to process and analyze text data within a predefined window of 8,000 tokens, ensuring consistency and standardization in the evaluation process. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + d95acc24180c47caa34114627d501592 + 545 + 2 + + + 1.0 + Natural Language Generation achieves state-of-the-art results + 322e02986c8724eedbcf3ebfa20b989c + f4753ab09adc42a9a52754e95440d4b9 + 546 + 2 + + + 1.0 + Natural Language Generation achieves competitive results + 322e02986c8724eedbcf3ebfa20b989c + 12f5a7c56b454a3d8aae97f65908f96b + 547 + 2 + + + 1.0 + Natural Language Generation is compared against human judgements + 322e02986c8724eedbcf3ebfa20b989c + 95f79ff0b8a34080ae2ac8448ce561f1 + 548 + 2 + + + 1.0 + Natural Language Generation can generate reference-based metrics + 322e02986c8724eedbcf3ebfa20b989c + 8733d4602c084e1cab1384dde0306abf + 549 + 2 + + + 1.0 + Natural Language Generation can measure qualities in a reference-free style + 322e02986c8724eedbcf3ebfa20b989c + ded3a49efdf6479a991cad53d0758cf4 + 550 + 2 + + + 1.0 + Es et al., 2023 discusses the RAGAS method + 322e02986c8724eedbcf3ebfa20b989c + 816fceb7e1ca4b5d9277368f78e6ed80 + 551 + 2 + + + 1.0 + RAGAS evaluates context relevance + 322e02986c8724eedbcf3ebfa20b989c + 50539d4503a4495097f49a8ed83e2462 + 552 + 2 + + + 1.0 + RAGAS evaluates faithfulness + 322e02986c8724eedbcf3ebfa20b989c + d6f67aa7ef0e4a19bf5830e777aafea5 + 553 + 2 + + + 1.0 + RAGAS evaluates answer relevance + 322e02986c8724eedbcf3ebfa20b989c + bbf61f9cd3e14f46a010d704e86be008 + 554 + 2 + + + 1.0 + The LLM evaluator assesses answers based on the directness metric + 322e02986c8724eedbcf3ebfa20b989c + 5d34e587bd2f41dba285e9178f179577 + 555 + 2 + + + 1.0 + Table 2 shows an example of LLM-generated assessment + 322e02986c8724eedbcf3ebfa20b989c + 901b491be7344401b4544ff05e591a0e + 556 + 2 + + + 1.0 + The LLM evaluator uses a head-to-head comparison approach + 322e02986c8724eedbcf3ebfa20b989c + ecacbf62b81d485396a56e1730e75a04 + 557 + 2 + + + 1.0 + The LLM evaluator assesses answers based on target metrics + 322e02986c8724eedbcf3ebfa20b989c + ba0ad1bcf02b4928a1b7ff7b23acdd6f + 558 + 2 + + + 1.0 + The LLM evaluator uses a control metric for validity + 322e02986c8724eedbcf3ebfa20b989c + 0e3c66c25d7e43a7960c37d28315e5d8 + 559 + 2 + + + 1.0 + The LLM evaluator accounts for stochasticity + 322e02986c8724eedbcf3ebfa20b989c + a0e0d5b7db9f4efcb5277856db799775 + 560 + 2 + + + 1.0 + The LLM evaluator uses mean scores from multiple comparisons + 322e02986c8724eedbcf3ebfa20b989c + 3f85dab93736440f9776020b6410aa9b + 561 + 2 + + + 1.0 + Directness is used to evaluate the straightforwardness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + 710ed70c346342ff81ccf205e30271bb + 562 + 2 + + + 1.0 + The question asks about public figures mentioned in entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + b3d3e8ba2ede4574a0498f082f0c15ae + 563 + 2 + + + 1.0 + Public figures are repeatedly mentioned across various entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + 8686013390614eca9116ccbab27431d7 + 564 + 2 + + + 1.0 + Answer 1 covers a wide range of public figures from different sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + fd8c8b7e3b9248abb1d8cb8958ab86d3 + 565 + 2 + + + 1.0 + Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports. + 718017a4871c909420f84b85b8ba969d + 039594428123415f95deb246f5097169 + 566 + 2 + + + 1.0 + Controversies involve public figures and impact public discourse. + 718017a4871c909420f84b85b8ba969d + d78ce7696ff14234a544de945ffe40d6 + 567 + 2 + + + 1.0 + Entertainment articles cover topics related to the entertainment industry + 322e02986c8724eedbcf3ebfa20b989c + 59b21508be904875af22b5c1cfdcd211 + 568 + 2 + + + 1.0 + Taylor Swift is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + e9c7a1d505b14229afbbef7c0d04751e + 569 + 2 + + + 1.0 + Travis Kelce is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 4b0efcd54efc40e8a884ac6c31deada2 + 570 + 2 + + + 1.0 + Britney Spears is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 0970f08f3d1a4d638d44e2ccb9237382 + 571 + 2 + + + 1.0 + Justin Timberlake is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + 8f10c11ecb5142029869025521c73431 + 572 + 2 + + + 1.0 + Taylor Swift is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + e36a0e3901864a7eaa5f5ad4280a6471 + 573 + 2 + + + 1.0 + Travis Kelce is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6fce354faa104fe58ba8a565eb3c43f2 + 574 + 2 + + + 1.0 + Britney Spears is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 20585e9a43c04375aa334e946e2dd144 + 575 + 2 + + + 1.0 + Justin Timberlake is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 32e343c0ae454660bdfcd1d3133baf0a + 576 + 2 + + + 1.0 + Actors and Directors are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 505ab840f6cc4fa6a839ebfe82d255ed + 577 + 2 + + + 1.0 + Musicians and Executives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + e38eb1698900424bb7392a74ff0f3351 + 578 + 2 + + + 1.0 + Athletes and Coaches are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 855c57eecf2a45c7aab02ff1ac36938d + 579 + 2 + + + 1.0 + Influencers and Entrepreneurs are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6ee77949c94d4906bd98c24341fdfa03 + 580 + 2 + + + 1.0 + Public Figures in Controversy are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + d06f506604b249feb423915db282ed75 + 581 + 2 + + + 1.0 + Film is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 0f642f63d4af4fc38298822bfc952719 + 582 + 2 + + + 1.0 + Television is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + e017ad1f09b049a7ad41d5a11dc1e3d9 + 583 + 2 + + + 1.0 + Music is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 5cbced0ba7044b7490f520a436261c57 + 584 + 2 + + + 1.0 + Sports is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + d45dea925f8d4e7e93d0e17317001eec + 585 + 2 + + + 1.0 + Digital Media is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 8123eee04a3a4c779f03bdb85de99f9f + 586 + 2 + + + 1.0 + Cultural Narratives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6129d90c83194bcfaede9ff00a011297 + 587 + 2 + + + 1.0 + Trends are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 6ef76e963a564dbe9c9feff4f8ce1683 + 588 + 2 + + + 1.0 + Social Discussions are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 1c8bad73fda646f8b3f413e432f0e351 + 589 + 2 + + + 1.0 + Public Discourse is a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + 7e75749d13d24321b8b10c5be0138805 + 590 + 2 + + + 1.0 + Reference-based metrics require gold standard answers + 322e02986c8724eedbcf3ebfa20b989c + 05bfaf60aa304a288e6789443bd6fd6c + 591 + 2 + + + 1.0 + Gold standard answers are lacking for sensemaking questions + 322e02986c8724eedbcf3ebfa20b989c + 6097e047a74d41ca996a0b7949ef6f0e + 592 + 2 + + + 3.0 + End users play a crucial role in the validation process of sensemaking questions and target metrics. Sensemaking questions are specifically validated with end users to ensure their relevance and accuracy. This collaborative approach ensures that the questions and metrics are aligned with the needs and expectations of the end users, thereby enhancing the overall effectiveness and applicability of the sensemaking process. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + e257439ce5be47a88faaeb0fe01bc4a1 + 593 + 2 + + + 1.0 + Target metrics are validated with end users + e4d9b12cf2b4c691c74019eefff4fb39 + 067b9486d59f45d2963235220f723a41 + 594 + 2 + + + 1.0 + The control metric is used as an indicator of validity + 322e02986c8724eedbcf3ebfa20b989c + 87c46c7ead5447bc8309ab116a316959 + 595 + 2 + + + 1.0 + Taylor Swift is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + f607d795f00347109cab3b2370c414f7 + 596 + 2 + + + 1.0 + Taylor Swift is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + 8f0610c89e9f42e9b8c3d8a947fa2852 + 597 + 2 + + + 1.0 + Travis Kelce is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + 75ef3591790a49748154ddbba20e9cdf + 598 + 2 + + + 1.0 + Travis Kelce is a public figure in the sports sector. + 718017a4871c909420f84b85b8ba969d + 58b7f26cb17b4b2283d3cacbaed15cfc + 599 + 2 + + + 1.0 + Britney Spears is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + 277cdf13617e47ca883b949f495bc243 + 600 + 2 + + + 1.0 + Britney Spears is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + 26080c121c9645b2bb258e4d61d47672 + 601 + 2 + + + 1.0 + Justin Timberlake is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + ee91a06f13b4495f95c800a0c7329ef7 + 602 + 2 + + + 1.0 + Justin Timberlake is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + 6ed8b67be79242e98aa1b9283431d5df + 603 + 2 + + + 1.0 + Answer 1 includes public figures from the film sector. + 718017a4871c909420f84b85b8ba969d + 40c2425cb1c34c1591f7cb89f9f5e0bf + 604 + 2 + + + 1.0 + Answer 1 includes public figures from the television sector. + 718017a4871c909420f84b85b8ba969d + 7cf59650687a435ba26a7c5ffc6c4f4c + 605 + 2 + + + 1.0 + Answer 1 includes public figures from the music sector. + 718017a4871c909420f84b85b8ba969d + 53c2882604b74192a649a4eaa0536c5e + 606 + 2 + + + 1.0 + Answer 2 focuses on public figures primarily from the music sector. + 718017a4871c909420f84b85b8ba969d + 3fbb8aeacea54ca9a957118fba613ccf + 607 + 2 + + + 1.0 + Answer 1 includes public figures from the sports sector. + 718017a4871c909420f84b85b8ba969d + 496ae6a894584a6cb12e50b516341788 + 608 + 2 + + + 1.0 + Answer 2 focuses on public figures primarily from the sports sector. + 718017a4871c909420f84b85b8ba969d + dd1a82c597794ba3a490cb70d488d9dd + 609 + 2 + + + 1.0 + Answer 1 includes public figures from the digital media sector. + 718017a4871c909420f84b85b8ba969d + bbd206ae4c1a4794813fd239fcfef313 + 610 + 2 + + + 1.0 + Answer 1 cites specific data sources from the News article dataset for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + 708ac105e8bb4beeade0472c899f214d + 611 + 2 + + + 1.0 + Answer 1 provides insights into controversies involving public figures and their impact on public discourse. + 718017a4871c909420f84b85b8ba969d + b4fe3c6aea95472db73a5e8bf575895a + 612 + 2 + + + 1.0 + Answer 1 includes public figures from the gaming sector. + 718017a4871c909420f84b85b8ba969d + a861f44aa7dd414790ee82b3f651c609 + 613 + 2 + + + 1.0 + Answer 1 cites specific data sources for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + add9948a221a4aabafbaaed650b1db26 + 614 + 2 + + + 1.0 + Answer 2 was generated using the Naïve RAG method, which directly lists specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d + 6c04f2ab7c9843ea900c3444b014bed8 + 615 + 2 + + + 2.0 + ANSWER 2 is a generated answer for a question in the NEWS ARTICLE DATASET. It relies heavily on a single source from the NEWS ARTICLE DATASET for data. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + 88356435ca9d43ebaf93134b3af8a53e + 616 + 2 + + + 1.0 + Answer 2 relies heavily on a single data source. + 718017a4871c909420f84b85b8ba969d + 233edf428a04436a8d32849af584f9d8 + 617 + 2 + + + 1.0 + Naïve RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + 28a6317db3d141db82a4a22525265fef + 618 + 2 + + + 1.0 + The global approach to Graph RAG shows improvements over naïve RAG + e4d9b12cf2b4c691c74019eefff4fb39 + 90051a1b69cd40f696e440d54085887e + 619 + 2 + + + 1.0 + LLM-generated assessments are used to evaluate the answers produced for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + 973fe6e8a1314a269748f40a98786115 + 620 + 2 + + + 1.0 + Example question is part of the News article dataset used for analysis + ebf5249c888e07fedce6572a4c03f88c + 9a442a8c054d48339aff04923bafe47f + 621 + 2 + + + 1.0 + Head-to-head win rate percentages were used to compare different conditions + 4c855404ee3d3c94aa2136f1513c666f + ffdacb33c3a94b7f9d890d7cc03a1f40 + 622 + 2 + + + 1.0 + Win rate percentages were used to measure the performance of different conditions + 4c855404ee3d3c94aa2136f1513c666f + 8792fc245cc94235a7764481ebad4828 + 623 + 2 + + + 1.0 + The overall winner per dataset and metric was determined for each condition + 4c855404ee3d3c94aa2136f1513c666f + b5982d09c32e4e7387e88f9160b4dd78 + 624 + 2 + + + 1.0 + Self-win rates were shown as the expected 50% for each condition + 4c855404ee3d3c94aa2136f1513c666f + 04ed223f57e44cf18284ba42ba760423 + 625 + 2 + + + 1.0 + The indexing process resulted in the creation of graphs + 36db32c37e1987e2c5863898ad882190 + 0debfb49a28d480db1b7d5ef713cac8f + 626 + 2 + + + 1.0 + Map-reduce summarization requires the highest number of context tokens + 36db32c37e1987e2c5863898ad882190 + 1f9abc7d006f4afa86200385acc3d1ae + 627 + 2 + + + 1.0 + Root-level community summaries require dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + cac3f76fbc11413e92cdfd3064d56ece + 628 + 2 + + + 2.0 + Queries are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + f120d98b793a4276a6f1a0a8e51a589a + 629 + 2 + + + 2.0 + Self-memory is related to generation-augmented retrieval + f35de4d9fb65f1d5a392064b20545c19 + bfda4c94278b49ab98cd3f407980d4d8 + 630 + 2 + + + 2.0 + CAiRE-COVID is a system for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + 7e5e5b80b84749d98cb36f56dbfcb47b + 631 + 2 + + + 2.0 + ITRG is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + 6330604339ca4113b94624bc9bed5ede + 632 + 2 + + + 2.0 + IR-CoT is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + 9638492c4f034be6b3bf88f8abd82edc + 633 + 2 + + + 2.0 + DSP is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + a76322b81f644f3c8733d04fa046b4e4 + 634 + 2 + + + 2.0 + RAPTOR is a method for generating a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + 653ee6584dbc46d1b8e97a05a3eac81e + 635 + 2 + + + 2.0 + The paper by Baek et al. discusses the KAPING method + 92e93fc6449756c0a60200636b297f65 + 9f0d58a479ec404d8e8f493f9269b08d + 636 + 2 + + + 2.0 + The paper by He et al. discusses the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + c5ae09d00a3f417981fc4177ef333eff + 637 + 2 + + + 2.0 + The paper by Zhang discusses the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + 4dd086fcba704d26b976c08a81c1465c + 638 + 2 + + + 2.0 + The paper by Kang et al. discusses the SURGE method + 92e93fc6449756c0a60200636b297f65 + f1ef6375ea84496eaed13c03318d80c6 + 639 + 2 + + + 2.0 + The paper by Ranade and Joshi discusses the FABULA method + 92e93fc6449756c0a60200636b297f65 + ba6829116d114532b99530f101ff0c72 + 640 + 2 + + + 2.0 + Both LangChain and LlamaIndex support a variety of graph databases + 92e93fc6449756c0a60200636b297f65 + 1ab2048463174873883061373d480ac4 + 641 + 2 + + + 2.0 + LangChain supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + 97038fe907af4710859c3daeb13972e9 + 642 + 2 + + + 2.0 + LangChain supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + e0595082eb9f41a4ac2afd9e614b363c + 643 + 2 + + + 1.0 + LangChain developed Langchain graphs + 71f6daf11e64e5273a3847d46bf228e1 + 5bd2ef268d4f4ba18925c17242370e21 + 644 + 2 + + + 2.0 + LlamaIndex supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + 96a21e764d1143fc90de0b2cc7751983 + 645 + 2 + + + 2.0 + LlamaIndex supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + 801b7142ab5b4c5eac41dade999a7c1f + 646 + 2 + + + 2.0 + NaLLM is a method that can create and reason over knowledge graphs in Neo4J format + 92e93fc6449756c0a60200636b297f65 + aac39de4e7e74d1c83f0eb835e635c88 + 647 + 2 + + + 2.0 + Neo4J developed Project NaLLM + 833e7d67dcd30790b26b71c9b5306f6b + c2e801c8221c4806a4f59ba5b793c784 + 648 + 2 + + + 2.0 + GraphRAG is a method that can create and reason over knowledge graphs in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + 351fc21986564103b324540289e2e608 + 649 + 2 + + + 2.0 + The paper by Manakul et al. discusses the SelfCheckGPT method + 92e93fc6449756c0a60200636b297f65 + 1c8a90b0aed7439286bbf85903d423d4 + 650 + 2 + + + 1.0 + SelfCheckGPT is an approach mentioned in the work by Manakul et al., 2023 + e4d9b12cf2b4c691c74019eefff4fb39 + 6c98609312154f118c04d8781663b16a + 651 + 2 + + + 1.0 + SelfCheckGPT is used to compare fabrication rates + e4d9b12cf2b4c691c74019eefff4fb39 + b91a6bf16e334b3ab7ec57665e980ceb + 652 + 2 + + + 1.0 + Embedding-based matching is used to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + c067e41adf9840df89052b111e6c0a6a + 653 + 2 + + + 1.0 + Hybrid RAG schemes combine embedding-based matching against community reports + e4d9b12cf2b4c691c74019eefff4fb39 + 76d7feb8140b4064b5492d3055736ee0 + 654 + 2 + + + 1.0 + The roll-up operation can be extended using map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + 53273797509d45178c49045830ec9fc2 + 655 + 2 + + + 1.0 + The drill down mechanism follows the information scent in the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + 2383fd3c3b4a4249a5a96550c494edb2 + 656 + 2 + + + 1.0 + The global approach to Graph RAG achieves competitive performance at a fraction of the token cost + e4d9b12cf2b4c691c74019eefff4fb39 + 29f172df150042e0a6db5481d5d91cfc + 657 + 2 + + + 1.0 + The open-source implementation of Graph RAG approaches is Python-based + e4d9b12cf2b4c691c74019eefff4fb39 + a243935f440241a281fbabb20422c641 + 658 + 2 + + + 1.0 + The drill down mechanism follows the information scent + e4d9b12cf2b4c691c74019eefff4fb39 + 34b704124fe94c2f933a344c11165f2e + 659 + 2 + + + 1.0 + Alonso Guevara Fernández and Amber Hoak both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + e429a497b01c40f3aef7e2205eaf01d8 + 660 + 2 + + + 1.0 + Alonso Guevara Fernández and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + f521933b91564693b07bd838160083ac + 661 + 2 + + + 1.0 + Alonso Guevara Fernández and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + e0d361b6991b40debf5599e86f2638ca + 662 + 2 + + + 1.0 + Alonso Guevara Fernández and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + cf16005cfadf4e48832ffd0e43f57be1 + 663 + 2 + + + 1.0 + Alonso Guevara Fernández and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + db10b0e690754748b0d75639f3e8d2b8 + 664 + 2 + + + 1.0 + Alonso Guevara Fernández and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + ec13b80209e246588bb5486d516f85eb + 665 + 2 + + + 1.0 + Alonso Guevara Fernández and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 617a76d54ed546e29428a31dea955b96 + 666 + 2 + + + 1.0 + Alonso Guevara Fernández and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + c88ffd2aa7284ac38eb4351c5fad6f44 + 667 + 2 + + + 1.0 + Alonso Guevara Fernández and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0dace3b24df14aae909a2815653e9db1 + 668 + 2 + + + 1.0 + Alonso Guevara Fernández and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + b3cfde857302479aa59b91d6648a40df + 669 + 2 + + + 1.0 + Alonso Guevara Fernández and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 01a52a986b6a444badc83fb11aa7a160 + 670 + 2 + + + 1.0 + Alonso Guevara Fernández and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2516b3485b754bdabf6820863c918e3d + 671 + 2 + + + 1.0 + Alonso Guevara Fernández and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0523ed6330f4429f8468f5b49169c940 + 672 + 2 + + + 1.0 + Alonso Guevara Fernández and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0b7ac591dfd34971b24d38e344b40c37 + 673 + 2 + + + 1.0 + Alonso Guevara Fernández and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2864f68297e94d7e84213833e22da077 + 674 + 2 + + + 1.0 + Alonso Guevara Fernández and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 629cd969d05c4c329bbe24f5d86e0089 + 675 + 2 + + + 1.0 + Alonso Guevara Fernández and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + f7bc5d1fb1404acdb77d50a6b9129141 + 676 + 2 + + + 1.0 + Alonso Guevara Fernández and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + d47535a52c2b40a3bacb3d520b8f0f1c + 677 + 2 + + + 1.0 + Alonso Guevara Fernández and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 557ed8720c2845cabcce0287f7284b3e + 678 + 2 + + + 1.0 + Alonso Guevara Fernández and Sarah Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + cef8ff96a0e842fdae4751933bcb1a28 + 679 + 2 + + + 1.0 + Alonso Guevara Fernández and Shane Solomon both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 93c8356953da486e9630d7e7304a6ff3 + 680 + 2 + + + 1.0 + Amber Hoak and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + dc52f3641c1548bba5b3cf8c65a5c072 + 681 + 2 + + + 1.0 + Amber Hoak and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + d0fdcc6945d84b20aa1de4afe2786592 + 682 + 2 + + + 1.0 + Amber Hoak and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + af6e03260c5946be96737b148b5edd9d + 683 + 2 + + + 1.0 + Amber Hoak and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1949eb874c544c58a71bbd04d6241a22 + 684 + 2 + + + 1.0 + Amber Hoak and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 09384ed4453846cb8c4d0076ecbf928a + 685 + 2 + + + 1.0 + Amber Hoak and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 181833ae880a4d0ab24ba0ccb158138d + 686 + 2 + + + 1.0 + Amber Hoak and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 3426a7ea02f740aeabcb552feee11bcc + 687 + 2 + + + 1.0 + Amber Hoak and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + e47ae0eea85b4f6e86b77fe56396460e + 688 + 2 + + + 1.0 + Amber Hoak and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 1ad7e0ad19334488b5d3b008f93a4ef4 + 689 + 2 + + + 1.0 + Amber Hoak and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 8c76a8cb5951422ba3b3cc6fcb66a391 + 690 + 2 + + + 1.0 + Amber Hoak and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 096dee591c1f4141a73fd628a59ffbe9 + 691 + 2 + + + 1.0 + Amber Hoak and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 9886a385806e46a69d92a726017b99b6 + 692 + 2 + + + 1.0 + Amber Hoak and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 4ecf891e9a5b4daf9e02d5b2ec963079 + 693 + 2 + + + 1.0 + Amber Hoak and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 59207227178147e39296a4059ac1055d + 694 + 2 + + + 1.0 + Amber Hoak and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 0682e47ed49146c0bc5e2b77fb924b6c + 695 + 2 + + + 1.0 + Amber Hoak and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 6022e4aa784f42b88dbcb27a5d9d2614 + 696 + 2 + + + 1.0 + Amber Hoak and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 386047cff59549ea83158b69bbac1870 + 697 + 2 + + + 1.0 + Amber Hoak and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + 2ead31e49ce643ebae4d5f047bb7a37b + 698 + 2 + + + 1.0 + J. Achiam and S. Adler co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + a74ee5b02e1e41b0ac4cf5449f7cdf2c + 699 + 2 + + + 1.0 + J. Achiam and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0f73fcdab31348a880a468124099071c + 700 + 2 + + + 1.0 + J. Achiam and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 24d2dcb7f28144cbad714b0a8b6c9e70 + 701 + 2 + + + 1.0 + J. Achiam and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 093447e0342e490aa6a55bd70ce7c2f2 + 702 + 2 + + + 1.0 + J. Achiam and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0439db7ac7d2484596e02246bd340424 + 703 + 2 + + + 1.0 + J. Achiam and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 81ffef86ebb341bebf145c742fb33dbd + 704 + 2 + + + 1.0 + J. Achiam and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + a0da2872126f43769f75c8533fca5e26 + 705 + 2 + + + 1.0 + J. Achiam and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 7facdc1f91014f42a67e34bac31a95ce + 706 + 2 + + + 1.0 + J. Achiam and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + ca7a635373294067b5f3050c82d38983 + 707 + 2 + + + 1.0 + S. Adler and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 2833c46f05984f729c7ec15e071f0c8e + 708 + 2 + + + 1.0 + S. Adler and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + ada7cf1171b74ad793f7856febc9c6fe + 709 + 2 + + + 1.0 + S. Adler and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 33b355c0a8044ef2b2b8be81bea0d431 + 710 + 2 + + + 1.0 + S. Adler and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 85698526e09a47878e3255a251d95406 + 711 + 2 + + + 1.0 + S. Adler and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 1a625c51e7ad497b86041757d1cde642 + 712 + 2 + + + 1.0 + S. Adler and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + df91c0e5657a4bafa849c8a3079ca582 + 713 + 2 + + + 1.0 + S. Adler and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 091a9788b29443509feda24aa5f5c241 + 714 + 2 + + + 1.0 + S. Adler and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 6f750deaedcb4612b419c3d8dd7e5cb2 + 715 + 2 + + + 1.0 + S. Agarwal and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + d5ea3f061e9c419fb1c07b680bfb287a + 716 + 2 + + + 1.0 + S. Agarwal and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 1f4fd98283df43c69d5537c002b98f58 + 717 + 2 + + + 1.0 + S. Agarwal and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + f3bb299bf6454785a8a406dce9776789 + 718 + 2 + + + 1.0 + S. Agarwal and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 218d3d5a4a544df99caed612e48add5b + 719 + 2 + + + 1.0 + S. Agarwal and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 02393af06d3649549b3e9290b4e46c0a + 720 + 2 + + + 1.0 + S. Agarwal and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + ffa9208f15744978a4ea45c1cff18a86 + 721 + 2 + + + 1.0 + S. Agarwal and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 23ea2a1d78984eb38721adeadee662e1 + 722 + 2 + + + 1.0 + L. Ahmad and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0d0a729e30634e1fb198609ce10c69bf + 723 + 2 + + + 1.0 + L. Ahmad and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 896e7d03cad7450e8044fcb0fd9f6e92 + 724 + 2 + + + 1.0 + L. Ahmad and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + a777a0c3a34b4990899f2e1e1f1c2074 + 725 + 2 + + + 1.0 + L. Ahmad and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 6cd46299702049bcbd39407fa97f0dc0 + 726 + 2 + + + 1.0 + L. Ahmad and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 7af01185ebd648e49bf9a57481e0dc7c + 727 + 2 + + + 1.0 + L. Ahmad and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + d6aad3495b4d4c7ab2a03c44600584ba + 728 + 2 + + + 1.0 + I. Akkaya and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 482d5ecf4ce949e9a5d81f1b368769ee + 729 + 2 + + + 1.0 + I. Akkaya and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 64f7a02f89bd4a37844c482f00d00643 + 730 + 2 + + + 1.0 + I. Akkaya and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 26609be86d614d85ae97deeae4a4be1e + 731 + 2 + + + 1.0 + I. Akkaya and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 5dffe69693734eaeb360de4582d489b0 + 732 + 2 + + + 1.0 + I. Akkaya and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + e4fe96090a7641c68d0b1995d1f238b4 + 733 + 2 + + + 1.0 + F. L. Aleman and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 3fa1055cd26840678d546570e8b423d9 + 734 + 2 + + + 1.0 + F. L. Aleman and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 9802dae4757b42269e93c66b5214a396 + 735 + 2 + + + 1.0 + F. L. Aleman and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + c39e66acec274a5980ce275709a847ba + 736 + 2 + + + 1.0 + F. L. Aleman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 667794d1397a40bb904d406205960864 + 737 + 2 + + + 1.0 + D. Almeida and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 891f50162c0140e4b9c0e4ba33f69a1b + 738 + 2 + + + 1.0 + D. Almeida and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 922773e5841a475d89d5904fe7a324f8 + 739 + 2 + + + 1.0 + D. Almeida and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0338bebae19c41c196ee6c09ccba36e3 + 740 + 2 + + + 1.0 + J. Altenschmidt and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 0b407647077c4288b2324f06ac355985 + 741 + 2 + + + 1.0 + J. Altenschmidt and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + 4e9254fd4b234106843cf8ff91fd3b6f + 742 + 2 + + + 1.0 + S. Altman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + e6a7f4ccd6f54136b784572db0d5cb88 + 743 + 2 + + + 1.0 + R. Anil and S. Borgeaud co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + ccf54360ef954353b71c1c8175cd7f4e + 744 + 2 + + + 1.0 + R. Anil and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + bd1c72f46b81427892b1f415fecce77e + 745 + 2 + + + 1.0 + R. Anil and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 148e7caecdf740e58ee09a9ff549d19c + 746 + 2 + + + 1.0 + R. Anil and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + df78f3e3415a4d47b6dffdd3890f3eee + 747 + 2 + + + 1.0 + R. Anil and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 003abb3c5743482aa63022cf20cf5ccc + 748 + 2 + + + 1.0 + R. Anil and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 4788433078b843079ccd9a64e5430169 + 749 + 2 + + + 1.0 + R. Anil and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + cbecfdbc04c9405aa139566d727d3a33 + 750 + 2 + + + 1.0 + R. Anil and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + f514a867efb948868009b435fecbe372 + 751 + 2 + + + 1.0 + S. Borgeaud and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 6e4c8a7f1da147f5b38103c51c999502 + 752 + 2 + + + 1.0 + S. Borgeaud and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + ac0e52b6b3ae4cc485f9eef2f2dea7e7 + 753 + 2 + + + 1.0 + S. Borgeaud and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + b150af2f3df24f17a7fd836ba663680a + 754 + 2 + + + 1.0 + S. Borgeaud and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + f9740be4adc946149b5941f355d45c74 + 755 + 2 + + + 1.0 + S. Borgeaud and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + c6afe51b28f94c3ba21640387edd2ee8 + 756 + 2 + + + 1.0 + S. Borgeaud and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 040b86f496dc4930a895f2c21cb0731c + 757 + 2 + + + 1.0 + S. Borgeaud and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + e13ed464574f483c9f1db5f569e91445 + 758 + 2 + + + 1.0 + Y. Wu and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + e1131985dc53451fa7543912b2e7db07 + 759 + 2 + + + 1.0 + Y. Wu and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + d6c00e0a975e4adc979afd25d4037d4d + 760 + 2 + + + 1.0 + Y. Wu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 9cac1113be2148ce8abaa957620f9d59 + 761 + 2 + + + 1.0 + Y. Wu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + c277297e3e7b417892e986c8767f58ad + 762 + 2 + + + 1.0 + Y. Wu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 744f492f46d841c0b0fee5f4a9b40b6c + 763 + 2 + + + 1.0 + Y. Wu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + d59386dbfa0349b49f7b904e288b21ad + 764 + 2 + + + 1.0 + J.-B. Alayrac and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 1005971b7e764bffa0a4610ad403976b + 765 + 2 + + + 1.0 + J.-B. Alayrac and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 1eaf3527e2804c75bbd9e3ccac9d760e + 766 + 2 + + + 1.0 + J.-B. Alayrac and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 301ab7b5f81d40788e46dacb09579b50 + 767 + 2 + + + 1.0 + J.-B. Alayrac and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + d6f25f03a08b41b4a2eaa9df3db9dceb + 768 + 2 + + + 1.0 + J.-B. Alayrac and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 6529cf63872440a98aeab73beee3762a + 769 + 2 + + + 1.0 + J. Yu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 65faae6c13f5444b8d71b4b2be38eba3 + 770 + 2 + + + 1.0 + J. Yu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 9187af05545a4c8d92e38c2b46254092 + 771 + 2 + + + 1.0 + J. Yu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + f12578d1ff7b46f5ae84c7672fac8deb + 772 + 2 + + + 1.0 + J. Yu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 6b5c42bba0ec48c1a5de177a7f1b9bfc + 773 + 2 + + + 1.0 + R. Soricut and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 444ab529d10c47f19ef33e931489b8b8 + 774 + 2 + + + 1.0 + R. Soricut and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 0ffea4c3c86849ab828036b67b58acdc + 775 + 2 + + + 1.0 + R. Soricut and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 65b488142164407a81b496b4820ef556 + 776 + 2 + + + 1.0 + J. Schalkwyk and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + b3b006f8629b44df81a266c1e4d81d3f + 777 + 2 + + + 1.0 + J. Schalkwyk and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 2a48f66b6a424b9ebf38562836fe1c82 + 778 + 2 + + + 1.0 + A. M. Dai and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + 54e486668fa94feda285f377fb05d14d + 779 + 2 + + + 1.0 + J. Baek and A. F. Aji co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + d7c4899260084560905ac54dba81f0e6 + 780 + 2 + + + 1.0 + J. Baek and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 880db899ce864932843fe230e3d364ad + 781 + 2 + + + 1.0 + A. F. Aji and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + 59463c48d2fb48308cd38ee8dd869f59 + 782 + 2 + + + 1.0 + T. Ban and L. Chen co-authored the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + ff69f1aae7404c38b8bde6abc5a79b57 + 783 + 2 + + + 1.0 + Baumel, T. and Eyal, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 8db22f709edb4ae98f0fef060ccd24b8 + 784 + 2 + + + 1.0 + Baumel, T. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 0e412e834e62475a9fe1920438f7b75b + 785 + 2 + + + 1.0 + Baumel, T. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + 2e3f6dbac98742ddb213037ae77f0a82 + 786 + 2 + + + 1.0 + Eyal, M. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + 59ced15f5a1d485ebf0eac7fa85c1cdf + 787 + 2 + + + 1.0 + Eyal, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + 7cef69e2a71c4379b0816844799fc71e + 788 + 2 + + + 1.0 + Elhadad, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + 0f565c275f8148d885ae53c315ddc568 + 789 + 2 + + + 1.0 + Blondel, V. D. and Guillaume, J.-L. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 29c05af68cc541b79fdf499eac42b9c6 + 790 + 2 + + + 1.0 + Blondel, V. D. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 8d5d7b8fb7d14d0ba46ce7f0be6de661 + 791 + 2 + + + 1.0 + Blondel, V. D. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + 607a66de21cf42e497c23013327b751f + 792 + 2 + + + 1.0 + Guillaume, J.-L. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + f4f85b6086384211a25248f614bfb786 + 793 + 2 + + + 1.0 + Guillaume, J.-L. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + c1a4bcd4e7874e699f06bc795e291150 + 794 + 2 + + + 1.0 + Lambiotte, R. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + ca585d891433495aa70a3a01b252e50c + 795 + 2 + + + 1.0 + Brown, T. and Mann, B. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + f26b5e1c52e445998b6a63738d203b38 + 796 + 2 + + + 1.0 + Brown, T. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 0945030309e14518a16df16fbb25c76f + 797 + 2 + + + 1.0 + Brown, T. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 35939cc34a734b5f867f8d75df419f37 + 798 + 2 + + + 1.0 + Brown, T. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 807561f61906451b880e04ac6a33687f + 799 + 2 + + + 1.0 + Brown, T. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 0538e2dc04174140a43bc0359fed2d23 + 800 + 2 + + + 1.0 + Brown, T. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 6e87d210775b45e4a09e518492329bce + 801 + 2 + + + 1.0 + Brown, T. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + bde6223f81884473a1acc3b75dd056aa + 802 + 2 + + + 1.0 + Brown, T. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 7beb3a2ecfd5419b950a20a155e06169 + 803 + 2 + + + 1.0 + Brown, T. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 961dda09b0df497a974c38c28eb90686 + 804 + 2 + + + 1.0 + Mann, B. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 1fb946075dd54b218b8dfad20647d33e + 805 + 2 + + + 1.0 + Mann, B. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + bf90efb1858e49b19987cbd280d0e911 + 806 + 2 + + + 1.0 + Mann, B. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + f9949e43ea004014abec1b59f2155b5a + 807 + 2 + + + 1.0 + Mann, B. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 3d89c7fe0b6448e0a0d27bceccc09f09 + 808 + 2 + + + 1.0 + Mann, B. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 185f98d04d9f484ab3d626fd459a23a2 + 809 + 2 + + + 1.0 + Mann, B. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 688e9b443bc44782855aea4afd8a9d16 + 810 + 2 + + + 1.0 + Mann, B. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + cd56c4963a0e49d7bab0e25f0e068779 + 811 + 2 + + + 1.0 + Mann, B. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + c4d4216677af42f5a29a0f4dcb442220 + 812 + 2 + + + 1.0 + Ryder, N. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + d99684f88f2d43eaacd62ba9082b64a5 + 813 + 2 + + + 1.0 + Ryder, N. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9ca07c62b7e146298882e33f3c6cb653 + 814 + 2 + + + 1.0 + Ryder, N. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + eef3aeb29aba43da93b433a816e77203 + 815 + 2 + + + 1.0 + Ryder, N. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 1c2a05515b9f49e1966a4ceb4bb0a3a5 + 816 + 2 + + + 1.0 + Ryder, N. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 86db00646d264b0a922c6b639dc9d16b + 817 + 2 + + + 1.0 + Ryder, N. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 7892ab98e1b0475c97a798aa8b2d7f6c + 818 + 2 + + + 1.0 + Ryder, N. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 9732636cdd50433bb146a241cd72dbc5 + 819 + 2 + + + 1.0 + Subbiah, M. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + a5a8a63d5baf4946b7d7d1696f0e4e0e + 820 + 2 + + + 1.0 + Subbiah, M. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + a69a82dd8773426096c58ddc56832770 + 821 + 2 + + + 1.0 + Subbiah, M. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 72a7215e3e4a4b0db851351dfe5afd37 + 822 + 2 + + + 1.0 + Subbiah, M. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 5f8224530d424618acb32b74a3afe2c9 + 823 + 2 + + + 1.0 + Subbiah, M. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + 99d4510732d843299514461aebd5f176 + 824 + 2 + + + 1.0 + Zhao, D. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 1730cbfab07747508d5b5ea421b97953 + 825 + 2 + + + 1.0 + Es, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + 7dd3aadc8f424988a72f8ba3ccf17155 + 826 + 2 + + + 1.0 + James, J. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + a2e7cbaf38c24564b2abe61680cacd72 + 827 + 2 + + + 1.0 + Espinosa-Anke, L. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + 162f1ccf8dfc46cea4d54a36ed9ec823 + 828 + 2 + + + 1.0 + Schockaert, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + 17127080df794121830177e93631aa3b + 829 + 2 + + + 1.0 + Feng, Z. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 374a8f81e6304b6d90b44cdceb90ecb4 + 830 + 2 + + + 1.0 + Feng, X. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 8c6aafd2a5da496385bea2c69be03a5a + 831 + 2 + + + 1.0 + Yang, M. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + 06b9da0d4d9b4d6bb762bd2eeca7028a + 832 + 2 + + + 1.0 + Qin, B. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + a5f6de474fb24ec9af7403231c616831 + 833 + 2 + + + 1.0 + Gao, Y. and Xiong, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + a0ef87eb823b400594300f5c47e5c9c3 + 834 + 2 + + + 1.0 + Gao, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + ed076834490640acbb5d837aaac9fed5 + 835 + 2 + + + 1.0 + Gao, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + f935684600f34a27906def1902627ff2 + 836 + 2 + + + 1.0 + Gao, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 72fc2f604fc644e39f7d70e25094e347 + 837 + 2 + + + 1.0 + Gao, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 3ddb886b220c4bb2ab3d68f7f29ce5c5 + 838 + 2 + + + 1.0 + Gao, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 2f482bb08d564072a5ff4f2509dfdda6 + 839 + 2 + + + 1.0 + Gao, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + fc86507186da4c6c94fe3b788d77c471 + 840 + 2 + + + 1.0 + Gao, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + c235c2c649484c83967e2a42523028bb + 841 + 2 + + + 1.0 + Xiong, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 7e9d748907ea4b74925a32999a2b40d9 + 842 + 2 + + + 1.0 + Xiong, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + eefaef8a5c7149d18d304f39bf41f280 + 843 + 2 + + + 1.0 + Xiong, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 327114716cea49a79d33ba609158cd87 + 844 + 2 + + + 1.0 + Xiong, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + b45920e405af45f787ab167f54cfd2e9 + 845 + 2 + + + 1.0 + Xiong, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 00db8f6e99254c99be6c6f5c14a79500 + 846 + 2 + + + 1.0 + Xiong, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + aadd82f0e70c4fc49b1bdee3f60c1890 + 847 + 2 + + + 1.0 + Xiong, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 9528d92ccc10454793c4df59e24586db + 848 + 2 + + + 1.0 + Gao, X. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 80618f4e809e4af1bcdb59342c375377 + 849 + 2 + + + 1.0 + Gao, X. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 32d785e275be458fb7178ad2021ecdfc + 850 + 2 + + + 1.0 + Gao, X. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 0757f97d1fbf49748169ba696a364e4c + 851 + 2 + + + 1.0 + Gao, X. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + ca9a355bf38b452cbde62dba747ec65f + 852 + 2 + + + 1.0 + Gao, X. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + ba297c67512447e4b86f0cbc39fbc301 + 853 + 2 + + + 1.0 + Gao, X. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 00a9c8745b404b659c76a694dba9851c + 854 + 2 + + + 1.0 + Jia, K. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + eb338f2214414f0f9fa396f06ca12860 + 855 + 2 + + + 1.0 + Jia, K. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + d119680bac3744e58d2ed3273b1208b6 + 856 + 2 + + + 1.0 + Jia, K. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + d60eefaddf1e4b1db125d8f9ac49bacb + 857 + 2 + + + 1.0 + Jia, K. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 6058343c0824402e9843c92b2991f778 + 858 + 2 + + + 1.0 + Jia, K. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 383a003edc5a4f2387c7dd7865a984c9 + 859 + 2 + + + 1.0 + Pan, J. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 6bf9623c44824e48b7451bdfa1b47816 + 860 + 2 + + + 1.0 + Pan, J. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + cc62f333666e427eb1c66ec3f12a7a55 + 861 + 2 + + + 1.0 + Pan, J. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + c6d99481f5f545278ca8a73650b66e87 + 862 + 2 + + + 1.0 + Pan, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 5a0887b99d8b4bd89286962cd6f07037 + 863 + 2 + + + 1.0 + Bi, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 75174e7af26f434c9154b182087b58dc + 864 + 2 + + + 1.0 + Bi, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 0c43dd117fe6495991d4b4d8c2f5d70e + 865 + 2 + + + 1.0 + Bi, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 65e5d27c4f8a4dfa8ad92f227964b9cf + 866 + 2 + + + 1.0 + Dai, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + 63c4595187884af29aa46d03319acded + 867 + 2 + + + 1.0 + Dai, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + b32482039edd4d50bc43514570500345 + 868 + 2 + + + 1.0 + Sun, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + f3e6e2c82bab4430a33987a19e3d1835 + 869 + 2 + + + 1.0 + Goodwin, T. R. and Savery, M. E. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 6768cc24da5d4a2492ff936dd4b35661 + 870 + 2 + + + 1.0 + Goodwin, T. R. and Demner-Fushman, D. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + 223e1e3e7c4f4282b086e940f8c935c2 + 871 + 2 + + + 2.0 + Khattab, O. and Santhanam, K. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + f5bb365c9a814b909df0351498d79bb5 + 872 + 2 + + + 2.0 + Khattab, O. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and highlights their collaborative work in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + ab4ad26863b44497a1e48aa7c17a096c + 873 + 2 + + + 2.0 + Khattab, O. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 145bc384290c49228a231ac124ce88a8 + 874 + 2 + + + 2.0 + Khattab, O. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + f64c99537adf489ea58940e417cb5924 + 875 + 2 + + + 2.0 + Khattab, O. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + fe98ea566cf6486b85f8ed14aabb2618 + 876 + 2 + + + 2.0 + Khattab, O. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 6bfb83cb716745fcb591c8d2fb54f8f4 + 877 + 2 + + + 1.0 + Khattab, O. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 9866640f891944c7bb0a08748aa8b91f + 878 + 2 + + + 2.0 + Santhanam, K. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This collaboration is mentioned in the text, highlighting their joint contribution to the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + b0d513adad104e14a89a767a66f30848 + 879 + 2 + + + 2.0 + Santhanam, K. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + db37a25770a34437b472fa0038837868 + 880 + 2 + + + 2.0 + Santhanam, K. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 67746ba67d80491da102aab7704dfd30 + 881 + 2 + + + 2.0 + Santhanam, K. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 873a1ca522e6461090d5cdebc2c9ae98 + 882 + 2 + + + 2.0 + Santhanam, K. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + d919ccd28e2248b5ab1dcdd7af8b00cf + 883 + 2 + + + 1.0 + Santhanam, K. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 1883a3ca4d6a4bfd984e7053e2553e16 + 884 + 2 + + + 2.0 + Li, X. L. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 2a0c93cf781a4020aceef7230b286bbf + 885 + 2 + + + 2.0 + Li, X. L. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 06282cc8998f4b2ea43e0a9522383639 + 886 + 2 + + + 2.0 + Li, X. L. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 182efa2240c14212bb021746a18936bd + 887 + 2 + + + 2.0 + Li, X. L. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 727bc610ea9a4393bfa5de453b84340f + 888 + 2 + + + 1.0 + Li, X. L. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 960c3b508a294332ba7c05ffd897db31 + 889 + 2 + + + 2.0 + Hall, D. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + bd0c1d79ef1144a49f3ce09d4cdf099b + 890 + 2 + + + 2.0 + Hall, D. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + b8ae79ed2d6d43f98e0808b5bea884dd + 891 + 2 + + + 2.0 + Hall, D. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + a03b33a4ee97467c808946679e240ddf + 892 + 2 + + + 1.0 + Hall, D. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 00b685bcb7a54c4493cd78da1f4752ab + 893 + 2 + + + 2.0 + Liang, P. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + 9a54e0361b684d24aefdc05fc340cf41 + 894 + 2 + + + 2.0 + LIANG, P. and ZAHARIA, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + d423b97f085947bd89529bc1ed2c41a7 + 895 + 2 + + + 1.0 + Liang, P. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 6fd2b0d5156c424a89cb1c068cf1e076 + 896 + 2 + + + 2.0 + Potts, C. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + a8c8dd8ddbc44363ac2102b9b8989c6d + 897 + 2 + + + 1.0 + Potts, C. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 00cb0db6e46749f7af97701ad26e23be + 898 + 2 + + + 1.0 + Zaharia, M. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + 4baa53e4336d4807964fa8d186b32bc5 + 899 + 2 + + + 1.0 + Kim, G. and Kim, S. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + f325a83e0c854a7ba5d46663ddff1a29 + 900 + 2 + + + 1.0 + Kim, G. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + c2b7e5d9761e423a81149a94537f6def + 901 + 2 + + + 1.0 + Kim, G. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 3d1ccd312d3a4e7387e888aaa137c7c2 + 902 + 2 + + + 1.0 + Kim, G. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 840966e7ac4a4b14ac912e75102d50b7 + 903 + 2 + + + 1.0 + Kim, G. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + f0ede764bdb1437b8cfcc20ca9598712 + 904 + 2 + + + 1.0 + Kim, S. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + dd38d562c21f444190768c8a154280da + 905 + 2 + + + 1.0 + Kim, S. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 1e66c1cbb56b41269555d27e1505ec92 + 906 + 2 + + + 1.0 + Kim, S. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + a653bd47ad3d4009ab6a5b8e6ff18679 + 907 + 2 + + + 1.0 + Kim, S. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 37b7cf055e604ec6927a9f0b15b2698d + 908 + 2 + + + 1.0 + Jeon, B. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 182e0f3d0abd4181820acdd2bf8e5eaf + 909 + 2 + + + 1.0 + Jeon, B. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 6fd6398a9bfd496f9a0505d9f3190362 + 910 + 2 + + + 1.0 + Jeon, B. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 2192d6436ca840a1bce77dbf9fd354af + 911 + 2 + + + 1.0 + Park, J. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + ef5e0bbdb3774a22900cf45e9b8863ad + 912 + 2 + + + 1.0 + Park, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 307de250d35e43a8b122c4232fa8fb7c + 913 + 2 + + + 1.0 + Kang, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + 0e2f68c8ff734b279b7aad333bcf2fda + 914 + 2 + + + 1.0 + Klein, G. and Moon, B. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 172b0d0fa0794494a3c50b135c1f2cd6 + 915 + 2 + + + 1.0 + Klein, G. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + 96f016e343b34ac894b0b7153f474ab0 + 916 + 2 + + + 1.0 + Klein, G. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + d356499ae9a345b6bbfb33b5fa01f47b + 917 + 2 + + + 1.0 + Moon, B. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + d7ebead985b34576821f30d83a416cd2 + 918 + 2 + + + 1.0 + Moon, B. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + 02c1447586fc46dfa65b793e0105a878 + 919 + 2 + + + 1.0 + Hoffman, R. R. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + 1b1560bb4b0447e5860f8ba351af112e + 920 + 2 + + + 1.0 + Koesten, L. and Gregory, K. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 23b929895979486cba3bf6a13f4ce655 + 921 + 2 + + + 1.0 + Koesten, L. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + f1ebe367253a4a4088b363a6cc4601a1 + 922 + 2 + + + 1.0 + Koesten, L. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + cf575adc3eb140f9aec33757ec040eb8 + 923 + 2 + + + 1.0 + Koesten, L. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + 4e581bc7d4424c2fb0023e5b11687e02 + 924 + 2 + + + 1.0 + Gregory, K. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + ae9f8a02ac0f43d4ba67ccce412989d6 + 925 + 2 + + + 1.0 + Gregory, K. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + b972541545604529a30cabc262d83dae + 926 + 2 + + + 1.0 + Gregory, K. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + ec3f73ffbb9742e090b65893d040434b + 927 + 2 + + + 1.0 + Groth, P. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + 20be7b3222174d31839fac6a278f8b61 + 928 + 2 + + + 1.0 + Groth, P. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + 35688e258b0e4cc78c8b92ef8a13d3e3 + 929 + 2 + + + 1.0 + Simperl, E. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + 3a9503f2d69343c396c9b1d842d1aa74 + 930 + 2 + + + 1.0 + Kuratov, Y. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 39516d28d39e49a2b80e6cfac32e2609 + 931 + 2 + + + 1.0 + Bulatov, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 43cf2e01543540789eb8781fdb5f287d + 932 + 2 + + + 1.0 + Anokhin, P. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 54beb6d012d844058715f8ef8a91c5da + 933 + 2 + + + 1.0 + Sorokin, D. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 5a291bf96ac141b98730ac27c96e829e + 934 + 2 + + + 1.0 + Sorokin, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 5e572d8b0a614ce1839ec9a568078cdc + 935 + 2 + + + 1.0 + Burtsev, M. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + 3f7b36b371da40568ce15510a35b58e7 + 936 + 2 + + + 1.0 + Laskar, M. T. R. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + 6db7771bcc674e4ead899fbdd417930f + 937 + 2 + + + 2.0 + Laskar, M. T. R. and Hoque, E. co-authored two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning with transformer models to improve the effectiveness of query-focused abstractive summarization. Both works contribute to advancing the application of transformer models in specialized summarization tasks. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + 176a96262ad64ccbacb1efdfb36bd88a + 938 + 2 + + + 1.0 + Laskar, M. T. R. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + bd120225f7b84bbdb0567048ca803e3c + 939 + 2 + + + 1.0 + Laskar, M. T. R. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + 8ecda003a3d044279b1f0bdc1c96c25e + 940 + 2 + + + 1.0 + Laskar, M. T. R. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + 8e02a61bda6a4470b693e7e234abfc94 + 941 + 2 + + + 1.0 + Hoque, E. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + b62e3858d801445facc3a501c5100723 + 942 + 2 + + + 1.0 + Hoque, E. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + 64e8971970e94ea79d10e46c55b3e761 + 943 + 2 + + + 1.0 + Hoque, E. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + b10fd3628e7a45d29a2814771f53ad60 + 944 + 2 + + + 1.0 + Hoque, E. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + 826bb16575a141d683fb871ec94517e0 + 945 + 2 + + + 1.0 + Huang, J. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + 52384316108d433397224cb36486407c + 946 + 2 + + + 1.0 + The 33rd Canadian Conference on Artificial Intelligence is also known as Canadian AI 2020 + 6cd82819982879bd164547d2773ba5c7 + 9a27717e1a1b499981031fd69c58aff1 + 947 + 2 + + + 1.0 + Springer published the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + 4efbe8fc23a64506b36d6cf29f968baa + 948 + 2 + + + 1.0 + Huang, J. X. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + d3b80330258d412f9ac6a7670fe79044 + 949 + 2 + + + 1.0 + Lewis, P. and Perez, E. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 52703e888bf4493b866186b889d85783 + 950 + 2 + + + 1.0 + Lewis, P. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 5e25a814a9a04bcda6017c9cc99880a7 + 951 + 2 + + + 1.0 + Lewis, P. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + bb2070d133f74049b88c96510fc807ba + 952 + 2 + + + 1.0 + Lewis, P. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + d0734be4aaab40eb9f2be6229f4a803c + 953 + 2 + + + 1.0 + Lewis, P. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + a5839bfcc6c0471c9337257ed05b361b + 954 + 2 + + + 1.0 + Lewis, P. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + bee441f5c41e41ff8220254bbf714eb4 + 955 + 2 + + + 1.0 + Lewis, P. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 8d93d45adbe547f78460a9ef3eb40ab2 + 956 + 2 + + + 1.0 + Lewis, P. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + c6f21de31a6f4fbda2eed1780ffed5b1 + 957 + 2 + + + 1.0 + Lewis, P. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 07c3f03764874b7680710ca030cdb60c + 958 + 2 + + + 1.0 + Perez, E. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 7a7990b6045c440ba606d142bd8ddc02 + 959 + 2 + + + 1.0 + Perez, E. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 803cb895664c40319ca40cc9abb6a03d + 960 + 2 + + + 1.0 + Perez, E. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 4c9e4e36560946699b6cb1e67b1437ae + 961 + 2 + + + 1.0 + Perez, E. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 8a7a2d6266424b9f9006502e82fcd778 + 962 + 2 + + + 1.0 + Perez, E. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 6b38285eedc544b08b444ee781db9f0c + 963 + 2 + + + 1.0 + Perez, E. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9d59b69c7c984abb9d3e281c04e73510 + 964 + 2 + + + 1.0 + Perez, E. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 83ee1d8db753419f8b240f419a139815 + 965 + 2 + + + 1.0 + Perez, E. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 9211b015bb074bcd89ae6c75ec10e6da + 966 + 2 + + + 1.0 + Piktus, A. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + cbc280dac03a4c19bb6737e3789c928f + 967 + 2 + + + 1.0 + Piktus, A. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 166cfa435aac4465943f59c2d04a0da1 + 968 + 2 + + + 1.0 + Piktus, A. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 44bced1d9b184aa29376cf3b0cdac625 + 969 + 2 + + + 1.0 + Piktus, A. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + ca58c4e2fae84899a780ff379e1927eb + 970 + 2 + + + 1.0 + Piktus, A. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 2060ce64f028490798a3ed69832e048d + 971 + 2 + + + 1.0 + Piktus, A. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + fa8fa48e2a7542fc8ff2c43c35e1b32b + 972 + 2 + + + 1.0 + Piktus, A. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + a37bd07ff1694b6c90572399f084e1ec + 973 + 2 + + + 1.0 + Petroni, F. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 826e32d5ea1d455f8ae2d3b77cd2b41e + 974 + 2 + + + 1.0 + Petroni, F. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 1a3bd511e04d4929a45a36fb80127353 + 975 + 2 + + + 1.0 + Petroni, F. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + d0cd49577d6a49f4a21fdc389aa84805 + 976 + 2 + + + 1.0 + Petroni, F. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + aa667f462aae45a8a700d83a68c1982f + 977 + 2 + + + 1.0 + Petroni, F. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 37f40795969a42b3b50e61b76a96fa07 + 978 + 2 + + + 1.0 + Petroni, F. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 6db48bd5e4ce4337aaac4648376ed07d + 979 + 2 + + + 1.0 + Karpukhin, V. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + 422e6a449e7e4ce69182113a6493a4e5 + 980 + 2 + + + 1.0 + Xu, Y. and Lapata, M. co-authored the paper "Text summarization with latent queries" + fc4b27d64f055b7fc30176ba110dd02e + 68511afc6e204c0b996d76cb75de081c + 981 + 2 + + + 1.0 + Huang, M. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + 050f02fbf9e64d08b108c5b921581335 + 982 + 2 + + + 1.0 + Duan, N. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + 97367a34fd6b4451b6be397496d646ea + 983 + 2 + + + 3.0 + Martin, S. and Brown, W. M. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with large graph structures, making it a valuable resource for researchers and practitioners in the domain of graph theory and network analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 80da3caff5344d56b6ca12660594949a + 984 + 2 + + + 3.0 + Martin, S. and Klavans, R. co-authored the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + e7efb7b459ae4ed4aa412cd20d808970 + 985 + 2 + + + 3.0 + Martin, S. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the capabilities and applications of the Openord toolbox, emphasizing its utility in handling extensive graph data efficiently. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + f0cc756b32314a1aae3e3cbb507850a2 + 986 + 2 + + + 3.0 + Brown, W. M. and Klavans, R. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + f3082b58c8a54c538cf3a0110296955b + 987 + 2 + + + 3.0 + Brown, W. M. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + c5f2254b67c04ad4add88875e5623e5a + 988 + 2 + + + 3.0 + KLAVANS, R. and BOYACK, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + 6e151c66c5574df39a7f243858e2ad3f + 989 + 2 + + + 2.0 + Newman, M. E. published the paper "Modularity and community structure in networks" in the Proceedings of the National Academy of Sciences + 833e7d67dcd30790b26b71c9b5306f6b + 166366ae9ec842ec9a1deeb13c94026e + 990 + 2 + + + 2.0 + Ram, O. and Levine, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + cdecc293edb847ae92c3bf8ff39d1e9a + 991 + 2 + + + 2.0 + Ram, O. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 3a615d980a124616a659136b4fd277b7 + 992 + 2 + + + 2.0 + Ram, O. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + a9b46d1f9ef747b69d6211386b5aaa20 + 993 + 2 + + + 2.0 + Ram, O. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 891a6dbec2ef4a039efaca78040b00c1 + 994 + 2 + + + 2.0 + Ram, O. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 981d45442a11448097acebc6080da414 + 995 + 2 + + + 2.0 + Ram, O. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + fad1c8144b504954bea46ede106d93ec + 996 + 2 + + + 2.0 + Levine, Y. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 53edb7e587204ed48e523c6f1f8f4056 + 997 + 2 + + + 2.0 + Levine, Y. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + f386b02e36884167a5db1a12ee6fcb1a + 998 + 2 + + + 2.0 + Levine, Y. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 2809d8a73b71495ca4220571dd54ba1e + 999 + 2 + + + 2.0 + Levine, Y. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 49f82fa775fb466bb9ae3db14db5b29a + 1000 + 2 + + + 2.0 + Levine, Y. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + f9312ee0bac24ea1b497e16e0958d621 + 1001 + 2 + + + 2.0 + Dalmedigos, I. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + c84bc84ffea84df9ad25ae9f972b4ec0 + 1002 + 2 + + + 2.0 + Dalmedigos, I. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + be1068c6efb24cde96e5a523eb04aee8 + 1003 + 2 + + + 2.0 + Dalmedigos, I. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + ecc4f28a7d574a5886f4c80a0b7cddd4 + 1004 + 2 + + + 2.0 + Dalmedigos, I. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + b506a4813da44600b136e949db4f2832 + 1005 + 2 + + + 2.0 + Muhlgay, D. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 77735209cabb415289c8ae4e102ff6df + 1006 + 2 + + + 2.0 + Muhlgay, D. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 1f1d88a6f6ce46bab94a4b50693c89ff + 1007 + 2 + + + 2.0 + Muhlgay, D. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 03454aaf00c54112a09ea4e52185b195 + 1008 + 2 + + + 2.0 + Shashua, A. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 3c5cd9cbad0d456cab4c76f1dfcde25b + 1009 + 2 + + + 2.0 + Shashua, A. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 8503eae9f6c746afae0caa58070f25e6 + 1010 + 2 + + + 2.0 + Leyton-Brown, K. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + 6e3c655e5b544277a62832a0974aa0ed + 1011 + 2 + + + 2.0 + Ranade, P. and Joshi, A. co-authored the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + bd0363bace6b42b0b3879bed5a064274 + 1012 + 2 + + + 2.0 + Sarthi, P. and Abdullah, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 99969eec8bf8441eaf9cb004cb61a13e + 1013 + 2 + + + 2.0 + Sarthi, P. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 0bcceb946a94486faf935f58dabea978 + 1014 + 2 + + + 2.0 + Sarthi, P. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 4d8421b4a6c74627afaa45aefa08c43a + 1015 + 2 + + + 2.0 + Sarthi, P. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + f879674860784f9eb4289aeb91728351 + 1016 + 2 + + + 2.0 + Sarthi, P. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + dea50b1765e54936b3d0b1e499ab2053 + 1017 + 2 + + + 2.0 + Abdullah, S. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 26634913d18f4629b39dffa19c1df734 + 1018 + 2 + + + 2.0 + Abdullah, S. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + a69bde74fc9d41cfa669f148c7c43dd8 + 1019 + 2 + + + 2.0 + Abdullah, S. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + e65667ec99e145fea2055d6b583cb05b + 1020 + 2 + + + 2.0 + Abdullah, S. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + cdfcbba5664d42508cd34df9af42b0dc + 1021 + 2 + + + 2.0 + Tuli, A. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 570de818eba04615a6afb3a573e82ff1 + 1022 + 2 + + + 2.0 + Tuli, A. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + 473533c454d34975a17a0193e39e0bac + 1023 + 2 + + + 2.0 + Tuli, A. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + e22af264c702440f93070465f45e630e + 1024 + 2 + + + 1.0 + Yang, Z. and Manning, C. D. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + ad95fbd957ae4c22859f58446dd8c9cc + 1025 + 2 + + + 1.0 + Huang, M. and Duan, N. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + 37d42129ca4d49dea240f66d1fdd4b78 + 1026 + 2 + + + 1.0 + Su, D. and Xu, Y. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1f40481f4ee342d4be51d33ffafc17d1 + 1027 + 2 + + + 1.0 + Su, D. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + e892d46e07e44bd5a2d1626875cc024f + 1028 + 2 + + + 1.0 + Su, D. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + fcb033a54d734ce5a87e0d8ad555867a + 1029 + 2 + + + 1.0 + Su, D. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 964fc01bfd9a400eb668761539dc9d9f + 1030 + 2 + + + 1.0 + Su, D. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 86b0d920fa504eba81c26cfc3f4d2b9f + 1031 + 2 + + + 1.0 + Xu, Y. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + e0221df7b7e44dd7956c8d0348d46b6d + 1032 + 2 + + + 1.0 + Xu, Y. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + a20d7f4fee104273b9628d648c05a5ac + 1033 + 2 + + + 1.0 + Xu, Y. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + b352398c5b1742d8a61acd8534ef0f53 + 1034 + 2 + + + 1.0 + Xu, Y. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 38b778af3c3f4be2a23e3932c94390c3 + 1035 + 2 + + + 1.0 + Yu, T. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 35d528e52a6441a58e58385d85bfae4b + 1036 + 2 + + + 1.0 + Yu, T. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 06d0d34ab3d043c689044a0fbfc65e10 + 1037 + 2 + + + 1.0 + Yu, T. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9e0ec036c91e44daa8e1a2af50df2081 + 1038 + 2 + + + 1.0 + Siddique, F. B. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 580b2395e68442539a606d37ddba691d + 1039 + 2 + + + 1.0 + Siddique, F. B. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 370fd1c6563045499e1d711fcd9ef9d5 + 1040 + 2 + + + 1.0 + Barezi, E. J. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + 3b6c4319026844ecb645f650e30b7d1a + 1041 + 2 + + + 1.0 + Tang, Y. and Yang, Y. co-authored the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + e044334b7d2e426ca2cab7eb763d8bc9 + 1042 + 2 + + + 1.0 + Touvron, H. and Martin, L. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 4a2fa382e77946d2be8e95edc04c6a64 + 1043 + 2 + + + 1.0 + Touvron, H. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + cd015281a069460e844faeb327b7d65f + 1044 + 2 + + + 1.0 + Touvron, H. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + c7a9fbf22a054056bf4f4562eaecfc08 + 1045 + 2 + + + 1.0 + Touvron, H. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 44bf341f78d74c4bb15ae209649d0ca9 + 1046 + 2 + + + 1.0 + Touvron, H. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 045ff6885d424b4caeabc76c50468c7c + 1047 + 2 + + + 1.0 + Touvron, H. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 543a52396f0b4f7f99ea755fba11d290 + 1048 + 2 + + + 1.0 + Touvron, H. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 33e487870db646e5b3d9c1f2962a7c6a + 1049 + 2 + + + 1.0 + Touvron, H. and Bhargava, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + b5b628d809514bfe9bbb3bd362815e79 + 1050 + 2 + + + 1.0 + Touvron, H. and Bhosale, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9d9ae51d2af44ebe8324dd2dd1dcd83b + 1051 + 2 + + + 1.0 + Martin, L. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + b7c606aa6ad1416e9f934628acce5f24 + 1052 + 2 + + + 1.0 + Martin, L. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 4521342f29774fab85e6acb0490d46e5 + 1053 + 2 + + + 1.0 + Martin, L. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + e7ac741e4aa4433ca5f2379726f90b33 + 1054 + 2 + + + 1.0 + Martin, L. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 43645eb9258244a8bd334ce77216b1c0 + 1055 + 2 + + + 1.0 + Martin, L. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0c9c52488ad647abbaf2b4589c976957 + 1056 + 2 + + + 1.0 + Martin, L. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0e8fb49a531e4ea48fece73957bd8a54 + 1057 + 2 + + + 1.0 + Wang, J. and Liang, Y. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 6187217c38ca4225b97d04d9644dcdf0 + 1058 + 2 + + + 1.0 + Wang, J. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8c8330abe276487294eba3a341ee9e0c + 1059 + 2 + + + 1.0 + Wang, J. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 200df54d93964e81ae2dcf727bffb23c + 1060 + 2 + + + 1.0 + Wang, J. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 4d0478c05f614675b336a76a0c088b3e + 1061 + 2 + + + 1.0 + Wang, J. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + c0dc4d618b5e49f7a18efa34dbf450ac + 1062 + 2 + + + 1.0 + Wang, J. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + dec05f79120940b78cd921a0a67f1540 + 1063 + 2 + + + 1.0 + Wang, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8edbf3e4f0d94f6ab78127c61bf87b76 + 1064 + 2 + + + 1.0 + Wang, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 1d722426930a42eeadfa624a6eb2408f + 1065 + 2 + + + 1.0 + Liang, Y. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 6d22f2009f6a4df9a242f03e2642981e + 1066 + 2 + + + 1.0 + Liang, Y. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + db5035c544214c72987eed4d4d9e327f + 1067 + 2 + + + 1.0 + Liang, Y. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + fd00e337b5c4465cbcbdf07bc294a3a8 + 1068 + 2 + + + 1.0 + Liang, Y. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + ea1e51b558c149649711a29157f4e604 + 1069 + 2 + + + 1.0 + Liang, Y. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + b888ad14e84347f8831a7dd2cea294fd + 1070 + 2 + + + 1.0 + Liang, Y. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 0e2323c0256d40579e7526dbdd019a8d + 1071 + 2 + + + 1.0 + Liang, Y. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + deb3bc5537a14352b22a0a473a59d8c7 + 1072 + 2 + + + 1.0 + Meng, F. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + ba445c400c8e405bb646387eab98a62b + 1073 + 2 + + + 1.0 + Meng, F. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 95a90d05e82d44ada6f8577ca49dd491 + 1074 + 2 + + + 1.0 + Meng, F. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + fa15140701a54689835604665d187c54 + 1075 + 2 + + + 1.0 + Meng, F. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8346468f7f4c46bebe1eaafd9753d55f + 1076 + 2 + + + 1.0 + Meng, F. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 46ddbfe32d444912b423dd1769fbaa43 + 1077 + 2 + + + 1.0 + Meng, F. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + ffce5a64e9394d1399319588d7fd4e3e + 1078 + 2 + + + 1.0 + Sun, Z. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + e934202aa3b344ba9fef89ecb42530b4 + 1079 + 2 + + + 1.0 + Sun, Z. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 22d5ff62439047ccaeaa63fd8a30f3e5 + 1080 + 2 + + + 1.0 + Sun, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 9b2f77b408ec4147bd5dd67a01d9f439 + 1081 + 2 + + + 1.0 + Sun, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 96b1264e89394adfaf026471e3b6ad47 + 1082 + 2 + + + 1.0 + Sun, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 00741dfb8f6d477f913d20406dfcd65d + 1083 + 2 + + + 1.0 + Shi, H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 3eb344aa8b05448984dacac93482ebc4 + 1084 + 2 + + + 1.0 + Shi, H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 8db8a8680d534161b0772d7a771df6bd + 1085 + 2 + + + 1.0 + Shi, H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 7f2c628b4fa54c0b9254049602ed20d2 + 1086 + 2 + + + 1.0 + Shi, H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + 2f8f5d33916d4824bec6773bacd37d87 + 1087 + 2 + + + 2.0 + Li, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + b8e1f95f9d3e497393d86e6bd137fe10 + 1088 + 2 + + + 2.0 + Li, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 13a41e2ff8b847ee8073e1e23b0bffc6 + 1089 + 2 + + + 2.0 + Li, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 9c61fb5ee44744a48bc5638bd42f654b + 1090 + 2 + + + 1.0 + H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 946213d345b64cbaa6becb8723b01d87 + 1091 + 2 + + + 1.0 + Zheng, L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e75c65762f064dfc95787fa331c95392 + 1092 + 2 + + + 1.0 + Chiang, W.-L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 9a586c1629464133920fb19d8bd1e690 + 1093 + 2 + + + 1.0 + Sheng, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e7a82e12e4f84f3e82c1ec74d3088235 + 1094 + 2 + + + 1.0 + Zhuang, S. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 51828127e71d40829039e033add265c4 + 1095 + 2 + + + 1.0 + Wu, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 04918b80bc714753b00af559d439a4ec + 1096 + 2 + + + 1.0 + Zhuang, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 3b4dbcb1c7c24bf8b6d55485c0304f7e + 1097 + 2 + + + 1.0 + Lin, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + fdd2a43d9b9f450c899adfb60b05e711 + 1098 + 2 + + + 1.0 + Li, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + bced585ff9d54fb7acd03f54f5729391 + 1099 + 2 + + + 1.0 + Li, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 50aec048280a4cdb8572993faab794dd + 1100 + 2 + + + 1.0 + Li, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + e128a7179b6e476c98d6bbfecf2a3f9a + 1101 + 2 + + + 2.0 + Xu, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + a3fa60c3370e4d5e8147250e2a18104a + 1102 + 2 + + + 2.0 + Xu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 1a7ef91522514b9f8b1ddaf68424351d + 1103 + 2 + + + 1.0 + H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + e5f094bf02d84a0889cd041199156ad7 + 1104 + 2 + + + 2.0 + Qu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + 5c64fc0a74044110906120ca1d5c7919 + 1105 + 2 + + + 1.0 + H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 524d1b6a01d34b0098a0da8af056bfc8 + 1106 + 2 + + + 1.0 + H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + 73af37cadd3c4d3dbfb8bfd697aeef58 + 1107 + 2 + + + 1.0 + Wang, S. and Khramtsova, E. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + c6b26fcc94044c368b2fe0db4b9b72f2 + 1108 + 2 + + + 1.0 + Wang, S. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 36b9f94e17c6481fb83670b70b192eb7 + 1109 + 2 + + + 1.0 + Wang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 24c63641b4e241589336236d5f916e34 + 1110 + 2 + + + 1.0 + Khramtsova, E. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 718e03207dcd44a080806880d08268ea + 1111 + 2 + + + 1.0 + Khramtsova, E. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 6d0efadfba5046eb86869827544c2703 + 1112 + 2 + + + 1.0 + Zhuang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + 2c33b12183ad4722ab1ab2cbd75f8312 + 1113 + 2 + + + 1.0 + Zheng, L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + f334bc9701204b1b943f9ece317ca68a + 1114 + 2 + + + 1.0 + Chiang, W.-L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 4ea3ecb74c91452da866f4c9163386e2 + 1115 + 2 + + + 1.0 + Sheng, Y. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + a12fe8fd9bc34db69d8de6944283d3c9 + 1116 + 2 + + + 1.0 + Zhuang, S. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 02b877f830cc4bb58dfad02f13a6d6ce + 1117 + 2 + + + 1.0 + Zhuang, S. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + abffcf93dc114332a181990ad56b7863 + 1118 + 2 + + + 1.0 + Zhuang, S. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + f894b0ae91eb412d93b6b06d4a73f350 + 1119 + 2 + + + 1.0 + Zhuang, S. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 75e06eb1f93c4ee38b782965ea905b5b + 1120 + 2 + + + 1.0 + Zhuang, S. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 70fd5c73bbe34d918b3dca3fc7294d28 + 1121 + 2 + + + 1.0 + Zhuang, S. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 8f1edc1d00764d6fb23859242c659deb + 1122 + 2 + + + 1.0 + Wang, Y. and Lipka, N. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + c8554314efb44679a898bbce08372abe + 1123 + 2 + + + 1.0 + Wang, Y. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + a41dbc4319f74bc995a93dbe0f4d9aee + 1124 + 2 + + + 1.0 + Wang, Y. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + ef76a41bf9bf45c893c475a7bd5a2938 + 1125 + 2 + + + 1.0 + Wang, Y. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 88f486cbb1904425a5fd5dfa268cf85d + 1126 + 2 + + + 1.0 + Wang, Y. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + ec10a8695b1a4e8787d9d29114e9d5ce + 1127 + 2 + + + 1.0 + Lipka, N. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 9397185bb4d7492b88eaa20fa10c0ae5 + 1128 + 2 + + + 1.0 + Lipka, N. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 2b1b9b1ed49c4ace91ff099752b8c0a5 + 1129 + 2 + + + 1.0 + Lipka, N. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 62c66c25992d4974829678313ed60b1d + 1130 + 2 + + + 1.0 + Lipka, N. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 26a889667b614ab890d863c4b8762e69 + 1131 + 2 + + + 1.0 + Rossi, R. A. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 30461855b0604128a4f10d0b348ce60f + 1132 + 2 + + + 1.0 + Rossi, R. A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + edef06de541f493f98d9281a704d785d + 1133 + 2 + + + 1.0 + Rossi, R. A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 0016a9dec22543e9b203f540860bf2e7 + 1134 + 2 + + + 1.0 + Siu, A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4f79558a259f4de58df5b022b68a459e + 1135 + 2 + + + 1.0 + Siu, A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + ca07919df74f4e5abfbd370c50eacc00 + 1136 + 2 + + + 1.0 + Zhang, R. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + 94b8715919cd49d08ac0ce99b930ea53 + 1137 + 2 + + + 1.0 + Yang, Z. and Qi, P. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4a10c341918b4d888b8b0466bd101b1d + 1138 + 2 + + + 1.0 + Yang, Z. and Zhang, S. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 5173ce4188034717b9c90eef40b94932 + 1139 + 2 + + + 1.0 + Yang, Z. and Bengio, Y. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + cff3415902bf4745992473697570aef0 + 1140 + 2 + + + 1.0 + Yang, Z. and Cohen, W. W. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 4c0cf727ec2843a288aa00b43f25b2de + 1141 + 2 + + + 1.0 + Yang, Z. and Salakhutdinov, R. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + 87ece3a4dcc84c98a291c1138ae56544 + 1142 + 2 + + + 1.0 + Zheng, L. and Chiang, W.-L. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 143dc5f4cb4b4596900ee5158594b1b0 + 1143 + 2 + + + 1.0 + Zheng, L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 3c222c5fdfab46e1ac1352a0f85a5fdd + 1144 + 2 + + + 1.0 + Zheng, L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 8c13a465b483417691c9b8d40b913da3 + 1145 + 2 + + + 1.0 + Zheng, L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 51633d2f0adf4123a23eeb292d95e649 + 1146 + 2 + + + 1.0 + Zheng, L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 849136ae8c9f4f9589a989bfe4c4155d + 1147 + 2 + + + 1.0 + Zheng, L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 84b11b37d1dd4e75b4c453669fbd4df9 + 1148 + 2 + + + 1.0 + Zheng, L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 373c198a0ed2402cb885b8d9f9de92f3 + 1149 + 2 + + + 1.0 + Zheng, L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 64ec8c4eb0734d60a5287e3df62652bd + 1150 + 2 + + + 1.0 + Chiang, W.-L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 7636104f26794a4e9e74b2d6943c879d + 1151 + 2 + + + 1.0 + Chiang, W.-L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + fc4b406a34ea4b2d9f305600aab14ea3 + 1152 + 2 + + + 1.0 + Chiang, W.-L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 5400473bab9d4105a1517fdc55c58f17 + 1153 + 2 + + + 1.0 + Chiang, W.-L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 624a1e6ea1d248f8b5126527e82e76c0 + 1154 + 2 + + + 1.0 + Chiang, W.-L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 66e53a4f6fc740aaaa379aa63d15f0e9 + 1155 + 2 + + + 1.0 + Chiang, W.-L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e676d0167a3f43478a209ec9526c90df + 1156 + 2 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 44fa3d2247904198b1c776e060d35eb2 + 1157 + 2 + + + 1.0 + Sheng, Y. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + a135859c7d3d4d3596f1e4ab218eff8a + 1158 + 2 + + + 1.0 + Sheng, Y. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 5e6fd98384a24a34b80311842661e00a + 1159 + 2 + + + 1.0 + Sheng, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 3a10d5261d4240c7b05b6cdb7838ff24 + 1160 + 2 + + + 1.0 + Sheng, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + ec595c7b07e148dba900040a68ef0fdb + 1161 + 2 + + + 1.0 + Sheng, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 4b43619c5c6a4ea3826bfd3c06aa6e66 + 1162 + 2 + + + 1.0 + Sheng, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 3173671571f14c75bfb9141754424efa + 1163 + 2 + + + 1.0 + Wu, Z. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 1675e75d7b524d4ab236eeaefd2dc992 + 1164 + 2 + + + 1.0 + Wu, Z. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + bb906c8e869141aa9be12118dcd3d3b5 + 1165 + 2 + + + 1.0 + Wu, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 27c95f7d6c3d4732897ae7bffd7c5dc8 + 1166 + 2 + + + 1.0 + Wu, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + b63c467faf714acd8a006431faf7a141 + 1167 + 2 + + + 1.0 + Wu, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 19a5840a67e14c468f9f3d6851eaee5c + 1168 + 2 + + + 1.0 + Zhuang, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 7a58673144d849e7a784caee9d9d4e99 + 1169 + 2 + + + 1.0 + Zhuang, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 8076db94b7214fdf9e006ce5a7e1cbe2 + 1170 + 2 + + + 1.0 + Zhuang, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + e733886404db4d46862bdddb2aee5211 + 1171 + 2 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + b75594a9d9c8404688a5cfe02272cdfc + 1172 + 2 + + + 1.0 + Lin, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + f253ff2311024729a758bb77b14bf72d + 1173 + 2 + + + 1.0 + Lin, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 1f0cb1e7fae64c238efb659d254d6221 + 1174 + 2 + + + 1.0 + Lin, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 6740be36e0e14774a5551a17db648a13 + 1175 + 2 + + + 1.0 + Li, D. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + 0f926e9dfaae4615b16a794e984b85ae + 1176 + 2 + + + 1.0 + Li, D. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + 47851446a4df4f5aa4505c999daaaaf7 + 1177 + 2 + + + 1.0 + Xing, E. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + c3e51a2782ba4f86b49b4038a316d9fb + 1178 + 2 + + + \ No newline at end of file diff --git a/graphfleet/output/graphindex/artifacts/create_base_documents.parquet b/graphfleet/output/graphindex/artifacts/create_base_documents.parquet new file mode 100644 index 000000000..4c27ca76d Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_base_documents.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_base_entity_graph.parquet b/graphfleet/output/graphindex/artifacts/create_base_entity_graph.parquet new file mode 100644 index 000000000..d1edf1d97 Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_base_entity_graph.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_base_extracted_entities.parquet b/graphfleet/output/graphindex/artifacts/create_base_extracted_entities.parquet new file mode 100644 index 000000000..1fc0c7e1e Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_base_extracted_entities.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_base_text_units.parquet b/graphfleet/output/graphindex/artifacts/create_base_text_units.parquet new file mode 100644 index 000000000..a7a577ed5 Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_base_text_units.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_final_communities.parquet b/graphfleet/output/graphindex/artifacts/create_final_communities.parquet new file mode 100644 index 000000000..f2e401889 Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_final_communities.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_final_community_reports.parquet b/graphfleet/output/graphindex/artifacts/create_final_community_reports.parquet new file mode 100644 index 000000000..887e9830f Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_final_community_reports.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_final_covariates.parquet b/graphfleet/output/graphindex/artifacts/create_final_covariates.parquet new file mode 100644 index 000000000..68bb1cb74 Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_final_covariates.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_final_documents.parquet b/graphfleet/output/graphindex/artifacts/create_final_documents.parquet new file mode 100644 index 000000000..ecf10855e Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_final_documents.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_final_entities.parquet b/graphfleet/output/graphindex/artifacts/create_final_entities.parquet new file mode 100644 index 000000000..b9469707c Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_final_entities.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_final_nodes.parquet b/graphfleet/output/graphindex/artifacts/create_final_nodes.parquet new file mode 100644 index 000000000..ec62657e8 Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_final_nodes.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_final_relationships.parquet b/graphfleet/output/graphindex/artifacts/create_final_relationships.parquet new file mode 100644 index 000000000..d3e203dc2 Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_final_relationships.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_final_text_units.parquet b/graphfleet/output/graphindex/artifacts/create_final_text_units.parquet new file mode 100644 index 000000000..23bde4661 Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_final_text_units.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/create_summarized_entities.parquet b/graphfleet/output/graphindex/artifacts/create_summarized_entities.parquet new file mode 100644 index 000000000..839369353 Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/create_summarized_entities.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/embedded_graph.0.graphml b/graphfleet/output/graphindex/artifacts/embedded_graph.0.graphml new file mode 100644 index 000000000..829426858 --- /dev/null +++ b/graphfleet/output/graphindex/artifacts/embedded_graph.0.graphml @@ -0,0 +1,9715 @@ + + + + + + + + + + + PERSON + Darren Edge is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Ha Trinh is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Newman Cheng is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Joshua Bradley is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Alex Chao is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Apurva Mody is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Steven Truitt is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Jonathan Larson is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Research is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Strategic Missions and Technologies is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Office of the CTO is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + METHOD + RAG (Retrieval-Augmented Generation) is a developing research area with multiple established directions, including knowledge graph creation, completion, and extraction of causal graphs. It is a method used for generating responses in text generation tasks by retrieving relevant information from an external knowledge source to enable large language models to answer questions. This approach incorporates the retrieval of relevant data to augment text generation, producing direct responses in various text generation tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY + LLM (Large Language Model) is a type of artificial intelligence model used for a variety of tasks in the field of Natural Language Processing and Information Retrieval. These tasks include generating and assessing text, entity extraction, summarization, understanding relationships in text, and automating human-like sensemaking and reasoning over large collections of documents. LLMs are also employed to generate intermediate answers and scores for text chunks, process these chunks to extract elements of a graph index, and automate the generation of questions for dataset evaluation. Additionally, LLMs can analyze and generate text based on retrieved information and queries, and they possess a context window that can be exceeded by external datasets. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,2c6ed90897310eea2f28e33fff1c32b0,6f33a085ff3304e5994f7fbb86c881a4,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + METHOD + Graph RAG (Retrieval-Augmented Generation) is a sophisticated method that leverages the natural modularity of graphs to partition data for global summarization tasks. This approach integrates multiple stages and concepts, including knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS), to support human sensemaking over text corpora. It is particularly effective in providing comprehensive and structured overviews of public figures across various sectors of the entertainment industry, as well as generating answers for questions in the News article dataset. + +Graph RAG employs a high-level data flow and pipeline for processing and summarizing text, combining both global and local approaches to optimize token usage in text generation tasks. It uses community summaries to improve answer comprehensiveness and diversity while requiring fewer tokens compared to traditional source text summarization methods. This method has been shown to outperform naive RAG in terms of comprehensiveness and diversity in text generation tasks. + +A specific implementation of Graph RAG involves using four levels of graph communities, incorporating concepts from other systems such as self-memory and parallel generation of community answers. This multi-stage mechanism allows for the comparison of multiple conditions, enhancing the overall effectiveness of the summarization process. + +Graph RAG, launched by NebulaGraph, is a retrieval-augmented generation technology based on knowledge graphs. It combines global summarization with graph-based text indexing to answer questions over private text corpora, making it a versatile tool for various text analysis and summarization applications. + 086021a89900a39bcb62036981737bfa,21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,718017a4871c909420f84b85b8ba969d,833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19,fb3c48579608fa28be585ceb6cd2f0fe + + + METHOD + QFS (Query-Focused Summarization) is a method used to generate summaries based on specific user queries + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + Community summaries are generated summaries of data clusters or communities, used to answer queries and tailored to the domain. These summaries are pre-generated for groups of closely-related entities, particularly in the Graph RAG approach, and are derived from community-generated content to compare with source texts. They are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks. Community summaries provide report-like insights into each community within a hierarchical structure, which is useful for understanding the dataset. They are generated from modular communities in the knowledge graph and cover different levels of each graph community hierarchy, including root-level communities in an entity-based graph index. Additionally, these summaries act as a kind of self-memory for generation-augmented retrieval. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + CONCEPT + Global sensemaking questions are questions that require understanding and summarizing large datasets, often beyond the explicit content of the source texts + e8d83e6e7a7c0f57b218cef24976b745 + + + METRIC + 1 million token range refers to the scale of datasets used in the evaluation of the Graph RAG approach + e8d83e6e7a7c0f57b218cef24976b745 + + + TECHNOLOGY + Python is a programming language used for implementing both global and local Graph RAG approaches. Additionally, Python is utilized to implement the open-source version of the Graph RAG approach. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + + + URL + The URL "HTTPS://AKA.MS/GRAPHRAG" is the location where the open-source, Python-based implementation of Graph RAG approaches will be available. This URL serves as the repository for accessing the open-source implementation of the Graph RAG approach. + e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745 + + + METHOD + Query-Focused Summarization (QFS) is a method used to generate summaries that are relevant to specific user queries. This summarization technique focuses on answering specific queries by utilizing the entire corpus of information available. It is designed to provide concise and relevant information based on the specific needs of the user, ensuring that the generated summaries are directly aligned with the queries posed. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + + + CONCEPT + An external knowledge source is a repository of information that can be accessed to retrieve relevant data for answering questions + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + A text corpus is a large collection of written texts used for analysis and research + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + Source documents are the original texts from which information is extracted, retrieved, or summarized. These documents serve as the foundational input texts for various processing tasks, including the Graph RAG (Retrieval-Augmented Generation) approach. In this context, source documents are critical for extracting and analyzing information, ensuring that the data used in computational linguistics and information retrieval tasks is accurate and comprehensive. + bc9e2c9e369c4108cf4f6dd5f60960f4,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + + + CONCEPT + A partial response is an intermediate answer generated from community summaries before being combined into a final response + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + A final response is the comprehensive answer generated after combining all partial responses + e8d83e6e7a7c0f57b218cef24976b745 + + + METRIC + COMPREHENSIVENESS is a metric used to evaluate the quality of generated responses by measuring how much detail an answer provides to cover all aspects and details of a question. It assesses the completeness and thoroughness of answers, ensuring that they encompass all relevant information. This metric is particularly important in evaluating the summarization approach, focusing on the completeness of the summary. In practical applications, such as evaluating Podcast transcripts and News articles, comprehensiveness has shown win rates between 72-83% and 72-80%, respectively. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + + + METRIC + DIVERSITY is a metric used to evaluate the variety and richness of answers generated in response to a question. It measures how varied and rich an answer is in providing different perspectives and insights. This metric is particularly important in assessing the quality of summarization approaches, focusing on the variety of information included in the summary. DIVERSITY is applied to various types of content, including Podcast transcripts, where win rates range from 75-82%, and News articles, with win rates ranging from 62-71%. It is a crucial target quality for evaluating the effectiveness of different methods in generating diverse and informative responses. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + + + ACTIVITY + Human endeavors refer to activities and efforts across various domains that rely on reading and reasoning about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models designed to understand and generate human-like text, playing a crucial role in automating sensemaking in complex domains. Modern language models, such as GPT, Llama, and Gemini, leverage in-context learning to effectively summarize content. These models are integral to the field of Natural Language Processing and Information Retrieval, enabling sophisticated text analysis and generation capabilities. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + + + DOMAIN + Scientific discovery is a complex domain where sensemaking is applied to understand and generate new knowledge from scientific texts + f0306814bf64f5c9e79603fc6a52f4ea + + + DOMAIN + Intelligence analysis is a complex domain where sensemaking is applied to understand and generate insights from intelligence data + f0306814bf64f5c9e79603fc6a52f4ea + + + PROCESS + SENSEMAKING is the process of understanding and making sense of complex information. It involves understanding connections among people, places, and events to anticipate their trajectories and act effectively. This process is crucial for navigating and interpreting intricate data landscapes, enabling individuals and organizations to make informed decisions based on the relationships and patterns identified within the information. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + TEXT CHUNKS are segments of text that are embedded into a vector space for analysis. These segments are extracted from source documents and are used for processing in the Graph RAG (Retrieval-Augmented Generation) approach. By embedding these text chunks into a vector space, they can be analyzed more effectively, facilitating various natural language processing and information retrieval tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + DATA + Element instances are identified and extracted instances of graph nodes and edges from text chunks. They represent individual occurrences of entities, relationships, and claims extracted from source texts. These specific pieces of information are tailored to the domain, providing a structured representation of the underlying data. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Element summaries are concise representations of element instances, tailored to the domain. They are descriptive texts created by the LLM to summarize entities, relationships, and claims extracted from source texts. These summaries provide detailed descriptions of nodes, edges, and covariates within a community, and are used to understand the structure and semantics of the dataset. In essence, element summaries serve as a tool to encapsulate and convey the intricate details of elements within a graph, facilitating a deeper comprehension of the dataset's structural dynamics and semantic relationships. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Graph communities are groups of elements, including nodes, edges, and covariates, detected within a graph index, primarily used for summarization. These communities consist of groups of nodes that exhibit stronger connections to each other than to nodes outside the group. This structural characteristic allows for the identification and analysis of densely connected subgraphs, which can be crucial for understanding the underlying relationships and dynamics within complex networks. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + COMMUNITY ANSWERS are query-focused summaries derived from community summaries. These answers are generated in parallel from the community summaries and are designed to respond to user queries effectively. Essentially, COMMUNITY ANSWERS serve as responses that synthesize information from community summaries to provide concise and relevant answers to specific questions posed by users. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + DATA + GLOBAL ANSWER is a comprehensive response generated from multiple community summaries to answer a user query. It is the final query-focused summary produced from all relevant community summaries. The final answer is generated by combining intermediate community answers based on their helpfulness scores. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + TIME + Indexing time refers to the time when the graph index is created and elements are summarized + f0306814bf64f5c9e79603fc6a52f4ea + + + TIME + Query time refers to the time when a query is made and the relevant summaries are generated + f0306814bf64f5c9e79603fc6a52f4ea + + + PROCESS + Graph RAG pipeline is a process using an LLM-derived graph index to detect, extract, and summarize nodes, edges, and covariates in source documents + f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + NODES are entities detected in the graph index of source documents. They represent the individual elements or points in a graph. For instance, in the Podcast dataset, there are 8,564 nodes, while the News dataset contains 15,754 nodes. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + EDGES are relationships detected in the graph index of source documents. They represent the connections or links between nodes in a graph. For instance, in the Podcast dataset, there are 20,691 edges, while the News dataset contains 19,520 edges. These edges are crucial for understanding the structural dynamics and relationships within the datasets, providing insights into how different nodes (such as topics, entities, or documents) are interconnected. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Covariates are additional attributes associated with extracted node instances in the graph index. They represent claims or additional information detected in the graph index of source documents. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + LEIDEN is a community detection algorithm renowned for its efficiency in recovering hierarchical community structures. It is widely used to partition graphs into modular communities, effectively grouping elements within a graph index. The algorithm's ability to identify and organize these communities makes it a valuable tool in the analysis of complex networks, particularly within the domains of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Retrieval-Augmented Generation (RAG) is an established approach in the field of Natural Language Processing and Information Retrieval, designed to answer user questions over entire datasets. This method involves retrieving relevant text regions to provide grounding for the generation task, thereby enhancing the accuracy and relevance of the generated responses. By combining retrieval and generation processes, RAG effectively synthesizes and presents pertinent information, making it a powerful tool for handling complex queries and large datasets. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + + + ORGANIZATION + Microsoft is a technology company whose Chief Technology Officer, Kevin Scott, actively participates in podcast conversations. The organization is deeply involved in automating sensemaking in scientific discovery through the use of large language models (LLMs). Notably, Microsoft conducted a study examining the impact of large language models, specifically GPT-4, on scientific discovery. + 1d07b4248c2655081c7af0e373bd70c9,833e7d67dcd30790b26b71c9b5306f6b,f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Ranade is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Joshi is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Klein is an author who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Lewis is an author who contributed to the development of the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Traag is an author who contributed to the development of the Leiden community detection method + f0306814bf64f5c9e79603fc6a52f4ea + + + PUBLICATION + arXiv is a preprint repository where several significant papers in the field of Natural Language Processing and Information Retrieval have been published. It serves as a platform for electronic preprints (known as e-prints) that are approved for publication after moderation, but not full peer review. Notable papers published on arXiv include "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models," "Lost in the middle: How language models use long contexts," "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," "Llama 2: Open foundation and fine-tuned chat models," "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy," "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries," "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions," "Enhancing knowledge graph construction using large language models," "Is chatgpt a good nlg evaluator? a preliminary study," "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt," "Causal graph discovery with retrieval-augmented generation based large language models," "Knowledge graph prompting for multi-document question answering," "Text summarization with latent queries," "Retrieval-augmented generation for large language models: A survey," and "Knowledge graph-augmented language models for knowledge-grounded dialogue generation." This repository is a crucial resource for researchers to disseminate their findings rapidly and access the latest advancements in their fields. + 00e8e4e881bd0862022f4dfc913b900b,086021a89900a39bcb62036981737bfa,58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035,f0306814bf64f5c9e79603fc6a52f4ea,fc4b27d64f055b7fc30176ba110dd02e + + + PUBLICATION + Preprint refers to the version of the research paper that is under review and available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + CATEGORY + cs.CL is the category under which the research paper is classified on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + DATE + 24 Apr 2024 is the date when the research paper was submitted to arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + IDENTIFIER + 2404.16130v1 is the identifier for the research paper on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Document collections refer to large sets of documents that are analyzed for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + TECHNOLOGY + LLM PROMPTS are specific instructions given to large language models (LLMs) to tailor their responses to the domain of the dataset. These prompts are also used to extract elements from text chunks, ensuring that the LLMs provide relevant and precise information based on the given context. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Community detection is a method used to identify groups of related elements within a graph. It involves the process of identifying communities within a graph, which are clusters of nodes that are more densely connected internally than with the rest of the network. This technique is crucial in understanding the structural dynamics and relationships within complex networks, such as those found in social networks, biological systems, and information retrieval systems. By uncovering these communities, researchers can gain insights into the underlying structure and function of the network, facilitating more effective analysis and interpretation of the data. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Domain-tailored summarization is a method used to create summaries that are specific to the domain of the dataset + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Klein et al. are authors who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Ranade and Joshi are authors who discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Lewis et al. are authors who developed the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Traag et al. are the authors who developed the Leiden algorithm, a method renowned for its efficiency in recovering hierarchical community structures. This algorithm is widely recognized in the field of Natural Language Processing and Information Retrieval for its ability to accurately detect and map out complex community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + QFS is a task framing that focuses on generating summaries based on specific queries, rather than just concatenating excerpts + fb3c48579608fa28be585ceb6cd2f0fe + + + METHOD + A type of summarization that generates natural language summaries based on specific queries, rather than just extracting text + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A neural network architecture that has shown substantial improvements in various summarization tasks + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + GEMINI is a family of highly capable multimodal models, as described in an arXiv preprint. These models are known for their ability to perform in-context learning and summarization, making them a significant advancement in the field of Natural Language Processing and Information Retrieval. + 086021a89900a39bcb62036981737bfa,fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A knowledge graph is a structured representation of information, utilized in the Graph RAG approach for summarization. This structured representation of knowledge is specifically employed in the Graph RAG approach for global summarization, highlighting its role in organizing and integrating information to facilitate comprehensive and coherent summaries. + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + + + REFERENCE + Authors of a paper on Retrieval-augmented generation (RAG) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Author of a paper on query-focused summarization (QFS) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "BROWN ET AL., 2020" refers to a publication by Brown et al. in 2020, which discusses in-context learning with few-shot examples. The authors of this paper are also known for their work on the GPT series of large language models. + bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "KURATOV ET AL., 2024" refers to a publication by Kuratov and colleagues in 2024. The study discusses the recall degradation and potential for information loss in longer context windows of Large Language Models (LLMs). The authors explore the limitations of these extended context windows, providing insights into how the performance of LLMs can be affected when dealing with longer sequences of text. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "LIU ET AL., 2023" refers to a publication by Liu et al. in 2023, which discusses the recall degradation and potential for information loss in longer context windows of large language models (LLMs). The authors explore the limitations of LLM context windows, highlighting how extended contexts can lead to decreased recall accuracy and information retention. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + TECHNOLOGY + COMMUNITY DETECTION ALGORITHMS are algorithms used to partition a graph into communities of nodes with stronger connections to one another. These algorithms are designed to identify modular communities of closely-related nodes within a graph, thereby revealing the underlying structure and relationships within the network. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + ALGORITHM + Louvain is a community detection algorithm used to partition graphs into modular communities + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + HOTPOTQA is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical resource for evaluating entity extraction prompts, particularly with advanced models like GPT-4-turbo. Additionally, HotPotQA is utilized to observe the behavior of text chunk extraction within the Graph RAG (Retrieval-Augmented Generation) approach, making it a versatile tool in the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNOLOGY + GPT-4-Turbo is a version of the GPT-4 model characterized by its large context size of 128k tokens, which is utilized in various analytical tasks. Specifically, GPT-4-Turbo is employed for entity extraction in evaluations, leveraging its extensive context capacity to enhance the accuracy and comprehensiveness of the analysis. This model is particularly suited for tasks within the Natural Language Processing and Information Retrieval domain, where handling large volumes of text and extracting relevant entities are critical. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + + + DATASET + The "PODCAST TRANSCRIPTS" dataset is a comprehensive collection of compiled transcripts from podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders. This dataset is used for analysis and consists of 1669 text chunks, each containing 600 tokens with 100-token overlaps between chunks, amounting to approximately 1 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620,ebf5249c888e07fedce6572a4c03f88c + + + DATASET + The "NEWS ARTICLES" dataset is a comprehensive collection of news articles used for analysis. It serves as a benchmark dataset comprising news articles published from September 2013 to December 2023. The dataset spans a range of categories, including entertainment, business, sports, technology, health, and science. It consists of 3197 text chunks, each containing 600 tokens, with a 100-token overlap between chunks, amounting to approximately 1.7 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620 + + + METHOD + MAP-REDUCE is a method employed for text summarization by applying a map-reduce approach directly to source texts. It is particularly utilized for query-focused summarization of an entire corpus, enabling efficient processing and extraction of relevant information from large datasets. This technique leverages the map-reduce paradigm to distribute the computational workload, making it suitable for handling extensive text collections in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,973164fa90bf2b4ee267f4fd795916bf + + + METRIC + "EMPOWERMENT" is a concept and metric used in the evaluation of various methods, with an average win rate of 51.3%. It measures how well an answer helps the reader understand and make informed judgments about a topic. Specifically, it evaluates the effectiveness of generated answers in empowering users by developing their understanding of broad issues and themes. Empowerment is a target quality in summarization approaches, focusing on the ability to help users reach an informed understanding. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + Naive RAG is a basic retrieval-augmented generation (RAG) method used as a baseline for comparison in text generation tasks. It converts documents to text, splits them into chunks, and embeds these chunks into a vector space for query matching. While it produces the most direct responses, it is outperformed by global approaches in terms of comprehensiveness and diversity. Naive RAG is also noted for listing public figures mentioned in entertainment articles, focusing on their professional achievements and personal lives. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c + + + METHOD + A method for summarizing source texts using a map-reduce approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Questions generated to evaluate the summarization approach, focusing on understanding activities + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + METRIC + TOKEN COSTS refer to the computational cost measured in tokens used in the summarization process. Specifically, in the context of the Graph RAG (Retrieval-Augmented Generation) approach, token costs denote the number of tokens required for processing text. This metric is crucial for evaluating the efficiency and scalability of text processing methods within the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS + The high-level process of the Graph RAG approach and pipeline + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + Design parameters are key settings and configurations in the Graph RAG approach. These parameters are crucial as they influence the design of the Graph RAG approach and pipeline, determining the effectiveness and efficiency of the overall system. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + PARAMETER + + + METHOD + GLOBAL SUMMARIZATION is a method for summarizing information on a global scale. It aims to encapsulate the overall structure and semantics of a dataset, providing a comprehensive overview of information from large datasets or corpora. This technique is particularly useful in the field of Natural Language Processing and Information Retrieval, where it helps in distilling vast amounts of data into coherent and concise summaries, facilitating better understanding and analysis of the underlying information. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e4d9b12cf2b4c691c74019eefff4fb39 + + + ATTRIBUTE + Modularity is an inherent quality of graphs that allows them to be partitioned into communities of closely-related nodes + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Descriptions generated from modular communities in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + + + INPUT + A specific question or request for information that the summarization methods aim to answer + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + A large collection of texts or documents used for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Intermediate answers generated from community summaries before being combined into a final global answer + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + The comprehensive answer generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + + + METHOD + A method that focuses on generating questions to understand activities from datasets + 21e52bc06a82796b1f4bcd73edda1f2a + + + INPUT + Brief descriptions of datasets used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + Datasets that represent real-world information, such as podcast transcripts and news articles + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + METHOD + A method that summarizes the original source texts directly + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + LOW-LEVEL COMMUNITY SUMMARIES are a type of community summary used in the News dataset for analysis. These summaries provide a detailed overview of the source text and are generated from lower hierarchical levels of the community in the knowledge graph. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + OUTPUT + INTERMEDIATE-LEVEL COMMUNITY SUMMARIES are summaries that provide a mid-level overview of the source text. These summaries are generated from intermediate hierarchical levels of the community in the knowledge graph, offering a balanced perspective that captures essential details without overwhelming the reader with excessive information. This approach ensures that the summaries are both informative and concise, making them valuable for understanding the structural dynamics and relationships within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + OUTPUT + Summaries generated from higher hierarchical levels of the community in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + + + PROCESS, SYSTEM + The entity "PIPELINE" refers to a series of processes or steps used to analyze and summarize a dataset. Specifically, in the context of the Graph RAG approach, the pipeline denotes the sequence of steps and processes involved. This structured sequence is essential for systematically handling data, ensuring that each stage of the analysis is methodically executed to achieve accurate and comprehensive results. + 7fb7d9ce2da9c940a32afdd87d1d9e56,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA STRUCTURE, OUTPUT + The "GRAPH INDEX" is a data structure used in Retrieval-Augmented Generation (RAG) systems to organize and retrieve information. It is a self-generated index that enables Graph RAG by utilizing a graph structure to organize and retrieve data. This index is created from a graph structure and is employed for tasks such as query-focused summarization. The graph index includes various elements extracted from text chunks using Large Language Model (LLM) prompts. Additionally, it supports conditions C0-C3 and is created using generic prompts for entity and relationship extraction. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + + + DATA, UNIT + Entity references are mentions of entities within text chunks, extracted during the processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC + Recall is a metric used to measure the completeness of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC + Precision is a metric used to measure the accuracy of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + FEW-SHOT EXAMPLES are specialized instances provided to the Large Language Model (LLM) to improve its performance in domains with specialized knowledge such as science, medicine, and law. These examples are tailored to the domain of the data used in the graph indexing process and serve as sample inputs for in-context learning. By tailoring the extraction prompt to the document corpus domain, few-shot examples enhance the LLM's ability to understand and process domain-specific information effectively. + 2c6ed90897310eea2f28e33fff1c32b0,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA, UNIT + Named entities are specific types of entities such as people, places, and organizations, extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + REFERENCE, PUBLICATION + A reference to a publication by Yang et al. in 2018, introducing the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METHOD, APPROACH + Techniques refer to the specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Implementation details are specific configurations and settings used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS, METHOD + A single extraction round refers to one complete cycle of extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC, ISSUE + Recall degradation refers to the decrease in recall performance when using longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS, METHOD + The extraction process involves identifying and extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Domain refers to the specific area of knowledge or field to which the document corpus belongs + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA, INPUT + Document corpus refers to the collection of documents being processed in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + Default prompt is the standard set of instructions given to the LLM for extracting named entities + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + Secondary extraction prompt is an additional set of instructions given to the LLM for extracting covariates + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METHOD + A covariate prompt is used to extract additional attributes associated with detected entities, including claims linked to the entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + Claims are statements or assertions linked to detected entities, including details such as subject, object, type, description, source text span, and start and end dates + 2c6ed90897310eea2f28e33fff1c32b0 + + + METHOD + Gleanings refer to multiple rounds of entity extraction to ensure that no entities are missed in the process + 2c6ed90897310eea2f28e33fff1c32b0 + + + TECHNIQUE + Logit bias is a technique used to force a yes/no decision from the LLM during the entity extraction process + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + An entity node is a representation of an entity in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A relationship edge is a representation of a relationship between entities in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A claim covariate is an additional attribute or variable associated with a claim in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + Communities of entities are groups of closely-related entities detected and summarized by the LLM + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + The "NOISY GRAPH STRUCTURE" refers to a graph structure that may contain inconsistencies or errors, making it challenging to analyze. This type of graph often includes duplicate or inconsistent entity elements due to variations in text format. These inconsistencies can arise from various sources, such as data entry errors, differing data formats, or incomplete information, which complicate the process of extracting meaningful insights and relationships from the graph. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + DOMAIN + Science is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + DOMAIN + Medicine is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + DOMAIN + Law is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Source text span is an attribute of a claim that indicates the specific portion of text from which the claim was extracted + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Start date is an attribute of a claim that indicates when the event or fact described in the claim began + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + End date is an attribute of a claim that indicates when the event or fact described in the claim ended + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Description is an attribute of a claim that provides a detailed explanation of the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Subject is an attribute of a claim that indicates the main entity involved in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Object is an attribute of a claim that indicates the entity that is affected by the subject in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A concept referring to an entity that has multiple name variations but is resilient to such variations due to sufficient connectivity to closely-related entities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models capable of understanding and generating human-like text. They are utilized for a variety of tasks, including the creation and completion of knowledge graphs, which are essential for structuring and interlinking information in a meaningful way. Additionally, LLMs serve as evaluators of natural language generation, assessing the quality and coherence of text produced by other AI systems. These models play a crucial role in the field of Natural Language Processing and Information Retrieval, contributing significantly to advancements in how machines comprehend and interact with human language. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf + + + TECHNOLOGY + Structured representations of knowledge in the form of triples (subject, predicate, object) used for reasoning tasks + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + Nodes in a graph that are of the same type and are described using rich descriptive text + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + Edges in a graph that represent relationships between entity nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + METRIC + Weights assigned to edges in a graph, representing the normalized counts of detected relationship instances + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The "HIERARCHICAL COMMUNITY STRUCTURE" is a framework in which communities are organized in a hierarchy, with each level providing a partition of the graph nodes. This structure organizes data into a hierarchy of communities, facilitating a multi-level clustering approach. Hierarchical community structure is utilized to generate community summaries, offering a comprehensive method for understanding the relationships and structural dynamics within specialized communities. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39 + + + CONCEPT + A division of graph nodes into mutually-exclusive, collectively-exhaustive communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + TECHNOLOGY + MULTIHOP-RAG is a benchmark dataset comprising news articles published from September 2013 to December 2023, covering a range of categories including entertainment, business, sports, technology, health, and science. It is specifically designed for open-domain question answering, targeting explicit fact retrieval. Additionally, MULTIHOP-RAG represents a specific implementation or variant of Retrieval-Augmented Generation (RAG) used in the context of graph communities. This dataset is also utilized for community detection and analysis, making it a versatile tool in the field of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author who has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + PERSON + Authors who have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The entity "DATASET" refers to a collection of data used for various purposes such as analysis, summarization, and evaluation. This can include diverse types of data like podcast transcripts and news articles. Specifically, the term encompasses datasets used for evaluation purposes, including notable examples like the Podcast and News datasets. + 1d07b4248c2655081c7af0e373bd70c9,7fb7d9ce2da9c940a32afdd87d1d9e56,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + GLOBAL QUERIES refer to questions or inquiries that require comprehensive answers derived from multiple sources or datasets. These queries aim to retrieve information from a global perspective, covering the entire dataset. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + ROOT COMMUNITIES are the top-level clusters in a hierarchical community structure. These communities represent the highest level of organization within the hierarchy, serving as the primary divisions from which more specific sub-communities branch out. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + SUB-COMMUNITIES are lower-level clusters within root communities in a hierarchical community structure, providing more detailed information. These sub-communities play a crucial role in breaking down the larger, more general root communities into more specific and focused groups, thereby facilitating a deeper and more granular understanding of the overall community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + TECHNOLOGY + Detailed documents that provide information about specific subtopics within a community + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The division of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + A system in which elements are ranked or organized in levels + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + LEVEL 0 represents the root-level communities in the hierarchical clustering with maximum modularity. It serves as the foundational layer in a hierarchical community structure, indicating the initial and most significant division of the dataset into distinct groups. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + LEVEL 1 is a sub-level in a hierarchical community structure, providing more detailed information about the internal organization. Specifically, Level 1 represents sub-communities within the root-level communities, thereby revealing the internal structure and dynamics of these larger groups. This level of granularity helps in understanding the intricate relationships and specialized interactions that occur within the broader community framework. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + A visual representation of graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + METHOD + The Leiden algorithm is a method used for detecting communities in large networks + 843fc5421e086120ffa1c75856ecf6cd + + + TOOL + OpenORD is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + + + TOOL + Force Atlas 2 is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Nodes represent entities in a graph, with size proportional to their degree + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Edges represent connections between nodes in a graph + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Covariates are variables that are linked to nodes and edges in a graph + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + The LLM context window is the token limit within which summaries are added for processing by a language model + 843fc5421e086120ffa1c75856ecf6cd + + + METHOD + Hierarchical clustering is a method of clustering data into a tree-like structure with multiple levels + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + The token limit is the maximum number of tokens that can be processed in a single context window by a language model + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Summary detail refers to the level of detail provided in a summary + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Scope refers to the range or extent of information covered in a summary + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + A "USER QUERY" is a question or inquiry posed by a user seeking information, which the system aims to answer. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd + CONCEPT + + + ELEMENT + Chunks are segments of community summaries divided into pre-specified token sizes + 843fc5421e086120ffa1c75856ecf6cd + ELEMENT + + + METRIC + Prominence is a metric used to prioritize community edges based on the combined degree of source and target nodes + 843fc5421e086120ffa1c75856ecf6cd + + + METRIC + Combined source and target node degree is a metric used to measure the overall prominence of community edges + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Community edges are connections between nodes within a community, prioritized based on prominence + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Sub-community summaries are shorter summaries of sub-communities used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + + + CATEGORY + Community level refers to the different levels in the hierarchical community structure used to generate summaries + 843fc5421e086120ffa1c75856ecf6cd + + + DATA + Chunks are segments of community summaries divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + + + METRIC + A score between 0-100 generated by the LLM to indicate how helpful an answer is in addressing the target question + 1d07b4248c2655081c7af0e373bd70c9 + + + USER + A user looking for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + + + USER + A user incorporating current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic addressing the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + The importance of health literacy highlighted through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + + + OUTPUT + Answers generated for each chunk of community summaries + 1d07b4248c2655081c7af0e373bd70c9 + + + METRIC + The pre-specified size of tokens used to divide community summaries into chunks + 1d07b4248c2655081c7af0e373bd70c9 + + + TECHNOLOGY + The "CONTEXT WINDOW" refers to a window of text used to generate answers, constrained by token size. The size of the context window is consistent across all conditions, ensuring uniformity in answer generation processes. + 1d07b4248c2655081c7af0e373bd70c9,973164fa90bf2b4ee267f4fd795916bf + + + PERSON + Kevin Scott is the Chief Technology Officer (CTO) of Microsoft and actively participates in podcast conversations. His involvement in these discussions is documented and compiled in the dataset, highlighting his contributions to the field of technology and his role in shaping Microsoft's strategic direction. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + Individuals who are leaders in the technology industry and participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + + + INPUT + A specific activity or goal that the user aims to achieve using the datasets + 1d07b4248c2655081c7af0e373bd70c9 + + + INPUT + QUESTIONS refer to specific inquiries generated by the Large Language Model (LLM) based on the user's task and the target datasets. These questions are utilized in the analysis to evaluate the performance of different methods within the domain of Natural Language Processing and Information Retrieval. The generation and subsequent use of these questions are crucial for assessing the effectiveness and accuracy of various computational techniques and models. + 1d07b4248c2655081c7af0e373bd70c9,4c855404ee3d3c94aa2136f1513c666f + + + + + 1d07b4248c2655081c7af0e373bd70c9 + + + DATASET + MT-BENCH is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical tool in evaluating the performance of models in retrieving accurate and relevant information from a broad range of topics. MT-Bench is prominently featured in the academic paper titled "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," where it is utilized to assess the capabilities of large language models in a structured and rigorous manner. This dataset plays a significant role in advancing the field of Natural Language Processing and Information Retrieval by providing a standardized metric for model comparison and evaluation. + 922778ce1cb2fdd6dbab1746c8795620,b1bbda43309e8e0e2175ea034aa88e13 + DATASET + + + PROCESS + The process through which people inspect, engage with, and contextualize data within the broader scope of real-world activities + 922778ce1cb2fdd6dbab1746c8795620 + PROCESS + + + TECHNOLOGY + Retrieval-Augmented Generation systems used for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + TECHNOLOGY + + + AUTHORS + Authors of a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors of a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + AUTHORS + Authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + + + 922778ce1cb2fdd6dbab1746c8795620 + + + PODCAST + "BEHIND THE TECH" is a podcast series featuring conversations between Kevin Scott and other technology leaders. It serves as a media platform associated with Kevin Scott, providing insights and discussions on various technological advancements and industry trends. + 833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + Kevin Scott, Microsoft CTO, who participates in the podcast conversations compiled in the dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + DATASET + A benchmark dataset for open-domain question answering, targeting explicit fact retrieval + 922778ce1cb2fdd6dbab1746c8795620 + + + METRIC + N represents the number of test questions per dataset used in the evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + METHOD + A method applying a map-reduce approach directly to source texts for summarization + 973164fa90bf2b4ee267f4fd795916bf + + + METHOD + A na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached + 973164fa90bf2b4ee267f4fd795916bf + + + CATEGORY + C0 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a root-level community summary, which is utilized to answer user queries by providing the fewest number of summaries. This category is essential for understanding the structural dynamics within the community, particularly in the domain of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C1 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a high-level community summary used to answer user queries, effectively representing sub-communities of C0. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C2 is a category or condition used in the analysis, representing a specific subset of the data. It functions as an intermediate-level community summary used to answer user queries, representing sub-communities of C1. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C3 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a category or condition that encapsulates low-level community summaries, which are instrumental in answering user queries. These summaries represent sub-communities of C2, providing detailed insights into the structural dynamics and relationships within the broader community. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + TS, or "Text Summarization," is a category or condition used in the analysis, representing a specific subset of the data. It is particularly focused on source text summarization within the analysis. TS employs a text summarization method that applies a map-reduce approach directly to source texts, facilitating efficient and scalable summarization processes. This category is integral to understanding and processing large volumes of text data, making it a crucial component in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + "SS" is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a baseline condition and is associated with a na¨ıve RAG (Retrieval-Augmented Generation) approach. In this context, text chunks are retrieved and added to the context window until the token limit is reached. + 4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CONCEPT + The prompts used for answer generation, which are the same across all conditions with minor modifications + 973164fa90bf2b4ee267f4fd795916bf + + + DATASET + The "PODCAST DATASET" is a collection of podcast transcripts utilized for both analysis and evaluation purposes. This dataset is specifically designed to support various analytical tasks, providing a rich source of textual data for researchers and practitioners in the field of Natural Language Processing and Information Retrieval. The transcripts within the dataset offer valuable insights and serve as a critical resource for evaluating different computational models and techniques. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + DATASET + The "NEWS DATASET" is a collection of news articles utilized for both analysis and evaluation purposes. This dataset serves as a valuable resource for examining and assessing various aspects of news content, making it an essential tool in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + METRICS in the context of Natural Language Processing and Information Retrieval are essential tools used to evaluate the performance of natural language generation. These metrics include both reference-based metrics, which compare generated texts to a set of reference texts, and qualities of the generated texts themselves. They are crucial in the analysis to assess the effectiveness of different methods in generating natural language, ensuring that the outputs are both accurate and of high quality. + 4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + "WANG ET AL., 2023A" refers to a study conducted by Wang and colleagues in 2023, which highlights the effectiveness of Large Language Models (LLMs) in evaluation. This study is a significant contribution to the field, providing insights into the capabilities and performance of LLMs in various evaluative tasks. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + "ZHENG ET AL., 2024" refers to a study conducted by Zheng and colleagues in 2024. This study highlights the effectiveness of Large Language Models (LLMs) in evaluation processes. The research, authored by Zheng et al., provides significant insights into the capabilities and applications of LLMs within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + The entity "CONDITIONS" refers to the different scenarios or variables that are compared in an experiment. Specifically, in the context of the analysis, these conditions include Graph RAG, text summarization, and semantic search RAG. These conditions are used to evaluate and compare various aspects of performance and effectiveness within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + USER QUERIES refer to the inquiries made by users to retrieve information. These queries are answered using different methods and conditions, depending on the context and the specific requirements of the information retrieval process. + 973164fa90bf2b4ee267f4fd795916bf,e4d9b12cf2b4c691c74019eefff4fb39 + + + CONCEPT + Types of entities extracted during the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + + + METRIC + The "CONTEXT WINDOW SIZE" refers to the fixed size of the context window used in various stages of natural language processing and information retrieval tasks. For the final evaluation, the context window size is set to 8k tokens. During the analysis phase, different context window sizes are tested, including 8k, 16k, 32k, and 64k tokens. Additionally, in the graph indexing process, the context window size is set to 600 tokens. This variability in context window sizes highlights the importance of adapting the window size to the specific requirements of different tasks within the domain. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + The process of extracting information, with 1 gleaning for the Podcast dataset and 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + TECHNOLOGY + Natural Language Generation (NLG) is a subfield of artificial intelligence that focuses on generating human-like text from data + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method where a Large Language Model (LLM) is used to compare and evaluate competing outputs + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method for evaluating the performance of Retrieval-Augmented Generation (RAG) systems, focusing on context relevance, faithfulness, and answer relevance + 322e02986c8724eedbcf3ebfa20b989c + + + PUBLICATION + A reference to a study or paper authored by Es and others in 2023 + 322e02986c8724eedbcf3ebfa20b989c + + + TOOL + A Large Language Model (LLM) used to evaluate and compare generated texts based on specific metrics + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + DIRECTNESS is a metric that measures how specifically and clearly an answer addresses a question. It is used to evaluate the straightforwardness of the generated answers. Additionally, it serves as a validity test metric to measure the directness of responses, with naive RAG (Retrieval-Augmented Generation) producing the most direct responses. + 322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + DATA + An example of LLM-generated assessment shown in a table format + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The entity "QUESTION" refers to a specific query used in the evaluation process, particularly as a metric to evaluate the generated responses by asking specific questions. This approach is commonly employed in the domain of Natural Language Processing and Information Retrieval to assess the quality and relevance of responses generated by various models or systems. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + ENTITY + Public figures are individuals who have gained fame or notoriety in various sectors such as entertainment, sports, and digital media. These individuals are well-known in the entertainment industry and are frequently mentioned across various articles. Their prominence in public discourse spans multiple domains, reflecting their influence and recognition in society. + 322e02986c8724eedbcf3ebfa20b989c,718017a4871c909420f84b85b8ba969d + + + DATASET + ENTERTAINMENT ARTICLES is a collection of articles focused on the entertainment industry. This dataset consists of articles related to various aspects of the entertainment sector, providing a comprehensive resource for understanding trends, developments, and key topics within this field. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + DOMAIN + The **ENTERTAINMENT INDUSTRY** is a multifaceted sector that encompasses various forms of entertainment, including movies, music, television, sports, and digital media. This industry is characterized by its diverse range of content and mediums, which collectively contribute to its broad appeal and significant cultural impact. The entertainment industry plays a crucial role in shaping public opinion, trends, and cultural norms through its extensive reach and influence across different platforms and genres. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric indicating the highest level of development or achievement in a particular field + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric indicating results that are comparable to or better than those of others in the same field + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric based on evaluations made by humans + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + Metrics that require a gold standard or reference answers for evaluation + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + An evaluation method that does not require reference answers + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how relevant the generated text is to the given context + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how accurately the generated text reflects the source information + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how relevant the generated answer is to the question + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method involving multiple stages or steps + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The correct or ideal answers used as a benchmark in evaluations + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + "SENSEMAKING QUESTIONS" are a class of questions used to evaluate the performance of Retrieval-Augmented Generation (RAG) systems. These questions are specifically designed to help users understand and make sense of complex information, as well as to validate the understanding and interpretation of data. By employing sensemaking questions, researchers and practitioners can assess how effectively a RAG system can retrieve and generate relevant information, thereby ensuring that the system aids in the comprehension and accurate interpretation of intricate datasets. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method where two items are directly compared against each other + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + TARGET METRICS are specific measures used to evaluate the performance of RAG systems. These metrics are aimed to be achieved or measured in the analysis and are the focus of an evaluation. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A metric used as a baseline or standard for comparison + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures the accuracy and reliability of a method or result + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures the randomness or variability in a process + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The average scores obtained from multiple evaluations + 322e02986c8724eedbcf3ebfa20b989c + + + PERSON + Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Britney Spears is a public figure frequently mentioned in entertainment articles, known for her significant contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his significant contributions to the music industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in film and television + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in music + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in sports + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in digital media and business + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry who are involved in controversies + e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric used to determine the winner in the comparison of generated responses + e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric used to evaluate the quality of LLM-generated responses + e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "FILM" refers to a sector within the entertainment industry that encompasses movies and cinema. This sector includes public figures involved in the movie industry, such as actors, directors, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "TELEVISION" refers to a sector within the entertainment industry that encompasses TV shows and series. This sector includes public figures involved in TV shows, such as actors, hosts, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + MUSIC is a sector within the entertainment industry that encompasses musical performances and recordings. This sector includes public figures involved in the music industry, such as singers, musicians, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "SPORTS" refers to a sector within the entertainment industry that encompasses athletic events and competitions. This sector includes public figures involved in sports, such as athletes, coaches, and sports commentators. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + DIGITAL MEDIA is a sector within the entertainment industry that encompasses online content and social media. This sector includes public figures involved in online platforms, such as influencers, content creators, and digital marketers. These individuals play a significant role in shaping digital landscapes through their engagement with audiences and their ability to leverage various online tools and platforms for content dissemination and marketing purposes. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes stories and themes that shape culture + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes popular movements and styles + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes public conversations and debates + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes formal discussions and communications + e8c8f911135faf3ff35f24107eb3f99c + + + RESPONSE + Answer 1 provides a varied and rich response by covering a wide range of public figures from different sectors of the entertainment industry, including film, television, music, sports, gaming, and digital media. It offers insights into the contributions and influence of these figures, as well as controversies and their impact on public discourse. The answer also cites specific data sources for each mentioned figure, indicating a diverse range of evidence to support the claims. + 718017a4871c909420f84b85b8ba969d + + + RESPONSE + "ANSWER 2" is a generated answer for the example question in the News article dataset. It focuses on a smaller group of public figures, primarily from the music industry and sports, and relies heavily on a single source for data. "ANSWER 2" provides concise explanations for the frequent mentions of specific public figures such as Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + Naïve RAG is a baseline method for retrieval-augmented generation (RAG) that does not use advanced techniques. It is a basic form of RAG with certain drawbacks that advanced RAG systems aim to overcome. Naïve RAG is used to generate answers for questions in the News article dataset and to generate responses that directly list specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d,e4d9b12cf2b4c691c74019eefff4fb39,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19 + + + DATASET + The "NEWS ARTICLE DATASET" is a collection of news articles utilized for various analytical purposes. This dataset is specifically employed for generating responses to questions about public figures in the entertainment industry, making it a valuable resource for both analysis and information retrieval tasks within this domain. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + TOPIC + Controversies are events or issues involving public figures that generate public debate and impact public discourse. + 718017a4871c909420f84b85b8ba969d + + + SECTOR + The gaming sector includes public figures involved in the gaming industry, including gamers, developers, and streamers. + 718017a4871c909420f84b85b8ba969d + + + RESOURCE + Data sources are references or reports used to support claims about public figures and their influence. + 718017a4871c909420f84b85b8ba969d + + + METHOD + Assessments generated by large language models (LLMs) to evaluate the answers produced by different methods + ebf5249c888e07fedce6572a4c03f88c + + + DATASET + An example question used in the News article dataset for analysis + ebf5249c888e07fedce6572a4c03f88c + + + DATA + The datasets used in the analysis, consisting of various text sources + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + A metric used to compare the performance of different conditions in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + CATEGORY + A specific setup or scenario used in the analysis, such as C0, C1, C2, C3, TS, and SS + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + WIN RATE is a measure used to evaluate the success rate of different methods in providing comprehensive and diverse answers. It represents the percentage of times a particular approach or method achieves a win in a given context. Specifically, it quantifies the percentage of times a condition outperformed another in the analysis. This metric is crucial in assessing the effectiveness of various strategies within the domain of Natural Language Processing and Information Retrieval, offering insights into the comparative performance of different techniques. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4 + + + METRIC + The condition that performed the best across all comparisons in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + The expected win rate of a condition when compared to itself, shown as 50% for reference + 4c855404ee3d3c94aa2136f1513c666f + + + METHOD + The use of large language models (LLMs) at the time of querying, evaluated in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + METHOD + The "FINAL EVALUATION" is the last stage of the analysis where the best performing context window size was used. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + + + PROCESS + The process that resulted in the creation of graphs for the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + + + STRUCTURE + A data structure consisting of nodes and edges, used to represent the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + + + METHOD + Approaches that consistently outperformed the naive RAG in comprehensiveness and diversity metrics + 36db32c37e1987e2c5863898ad882190 + + + METRIC + The number of context units, such as community summaries or text chunks, used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + + + METRIC + The term "TOKENS" refers to the number of individual words used in the analysis. The evaluation typically focuses on corpora in the region of 1 million tokens. This metric is crucial for understanding the scope and scale of the text data being analyzed, particularly in the fields of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,92e93fc6449756c0a60200636b297f65 + METRIC + + + METRIC + The percentage of the maximum token count used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + + + METHOD + MAP-REDUCE SUMMARIZATION is a method for summarizing source texts using a map-reduce approach. This summarization technique is notably resource-intensive, necessitating the highest number of context tokens compared to other methods. The map-reduce framework, originally popularized for its efficiency in processing large-scale data, is adapted here to handle the complexities of text summarization, ensuring comprehensive and accurate extraction of key information from extensive source texts. + 36db32c37e1987e2c5863898ad882190,e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Summaries at the root level of the community hierarchy, requiring dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + + + DATASET + SOURCE TEXTS are the original texts from which summaries or analyses are derived. These texts serve as the foundational material used for comparison with community summaries in the analysis. + 6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39 + + + REFERENCE + A reference to a paper by Ram et al. in 2023 discussing RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + REFERENCE + "GAO ET AL., 2023" is a paper published in 2023 by Gao et al. that delves into advanced Retrieval-Augmented Generation (RAG) techniques, specifically where the index is a knowledge graph. The publication also touches upon naive RAG approaches, providing a comprehensive examination of both advanced and basic methodologies within the domain of Natural Language Processing and Information Retrieval. + 6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + + + CATEGORY + Intermediate-level summaries are a type of community summary used in the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + CATEGORY + Root-level summaries are a type of community summary used in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + METRIC + Answer comprehensiveness is a measure used to evaluate the completeness of answers provided by different methods + 6f33a085ff3304e5994f7fbb86c881a4 + + + TECHNOLOGY + Ad-hoc LLM use refers to the spontaneous use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + + + TECHNOLOGY + Element extraction prompts are used to extract specific details in the Graph RAG index + 6f33a085ff3304e5994f7fbb86c881a4 + + + CONCEPT, TECHNOLOGY + A mathematical space in which text chunks and queries are embedded to represent similar semantics + f35de4d9fb65f1d5a392064b20545c19 + + + CONCEPT, DATA + Search inputs that are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A type of RAG system that includes patterns for iterative and dynamic cycles of interleaved retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, CONCEPT + A concept related to generation-augmented retrieval that facilitates future generation cycles + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method that facilitates future generation cycles by using self-memory + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A strategy for iterative retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A federated strategy for retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method that combines multiple concepts for summarizing multiple documents + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method for answering questions that require multiple steps or "hops" to gather information + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + An approach that involves generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A method for generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A method for answering multiple interpretations of ambiguous questions by generating a hierarchical structure + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A process that involves using LLMs to create knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A process that involves using LLMs to complete existing knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + Graphs that represent causal relationships, which can be extracted using LLMs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + REFERENCE, PUBLICATION + A reference to a publication by Cheng et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Mao et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Shao et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Wang et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Su et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Feng et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Trivedi et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Khattab et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Sarthi et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Kim et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + "TRAJANOSKA ET AL., 2023" refers to a paper by Trajanoska et al. published in 2023, which focuses on using Large Language Models (LLMs) for knowledge graph creation. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting innovative methodologies for leveraging advanced language models to construct and enhance knowledge graphs. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + "Yao et al., 2023" refers to a paper published by Yao and colleagues in 2023. The study focuses on the application of large language models (LLMs) for the task of knowledge graph completion. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting the potential of advanced LLMs to enhance the accuracy and efficiency of knowledge graph completion processes. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + TECHNOLOGY, METHOD + A system that combines multiple concepts for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + Strategies used before the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Strategies used during the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Strategies used after the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A pattern in Modular RAG systems for iterative and dynamic cycles of retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Cycles of generation that are facilitated by self-memory in Graph RAG + f35de4d9fb65f1d5a392064b20545c19 + + + PUBLICATION + A paper by Ban et al. published in 2023, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + PUBLICATION + A paper by Zhang et al. published in 2024, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where the index is a knowledge graph, developed by Baek et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Baek et al. published in 2023, focusing on the KAPING method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where subsets of the graph structure are the objects of enquiry, developed by He et al. in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by He et al. published in 2024, focusing on the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where derived graph metrics are the objects of enquiry, developed by Zhang in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Zhang published in 2023, focusing on the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, developed by Kang et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Kang et al. published in 2023, focusing on the SURGE method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where retrieved event-plot subgraphs are serialized using narrative templates, developed by Ranade and Joshi in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Ranade and Joshi published in 2023, focusing on the FABULA method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + PUBLICATION + A paper by Wang et al. published in 2023, focusing on a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + ORGANIZATION + LangChain is an organization that developed Langchain graphs and supports a variety of graph databases. + 71f6daf11e64e5273a3847d46bf228e1,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + + + ORGANIZATION + LlamaIndex is an organization that developed the LlamaIndex Knowledge Graph Index and supports a variety of graph databases. + 6cd82819982879bd164547d2773ba5c7,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + + + TECHNOLOGY + Neo4J is both a graph database format supported by various Retrieval-Augmented Generation (RAG) applications and an organization that developed Project NaLLM. The graph database format of Neo4J is widely recognized for its efficiency in handling complex relationships and structures, making it a valuable tool in the field of Natural Language Processing and Information Retrieval. As an organization, Neo4J has contributed significantly to the advancement of these domains through innovative projects like NaLLM, which further underscores its pivotal role in the community. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + + + METHOD + A method that can create and reason over knowledge graphs in Neo4J format, developed by Neo4J in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + TECHNOLOGY + NebulaGraph is both a graph database format and an organization that has made significant contributions to the field of graph databases and retrieval-augmented generation (RAG) applications. As a graph database format, NebulaGraph is supported by various RAG applications, facilitating the efficient handling and querying of complex graph data structures. Additionally, NebulaGraph, as an organization, has pioneered the industry-first graph RAG, which integrates retrieval-augmented generation with large language models (LLMs) based on knowledge graphs. This innovation underscores NebulaGraph's role in advancing the capabilities of knowledge graph-based applications and enhancing the performance of LLMs in generating contextually relevant information. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + + + METHOD + A method that can create and reason over knowledge graphs in NebulaGraph format, developed by NebulaGraph in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + METHOD + A method for comparing fabrication rates, developed by Manakul et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + "MANAKUL ET AL., 2023" refers to a paper by Manakul et al. published in 2023, which focuses on the SelfCheckGPT method. This work by Manakul and colleagues is centered around the development and application of SelfCheckGPT, a technique likely aimed at enhancing the performance and reliability of GPT models. The paper contributes to the field of Natural Language Processing and Information Retrieval by addressing specific challenges and proposing innovative solutions through the SelfCheckGPT method. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + PUBLICATION + + + STAKEHOLDER + END USERS are individuals who are the final users of the system or analysis. They play a crucial role in validating sensemaking questions and target metrics, ensuring that the system or analysis meets the intended objectives and provides meaningful insights. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + STAKEHOLDER + + + CONCEPT + Considerations and compromises involved in building a graph index + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METRIC + The effectiveness of RAG systems, which varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + METRIC + + + CONCEPT + Various forms of data used in RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METRIC + The scale of datasets used in RAG systems, which affects performance + 92e93fc6449756c0a60200636b297f65 + METRIC + + + PROCESS + The process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + PROCESS + + + DATASET + Collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + DATASET + + + CONCEPT + Different categories of questions used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METHOD + SelfCheckGPT is an approach used to compare fabrication rates in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method for global summarization of source texts that does not use a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + RESOURCE + The amount of computational resources allocated for a task + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The expected number of queries over the lifetime of a dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Annotations that provide detailed information about the text + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method that uses embeddings to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + RAG schemes that combine embedding-based matching with other approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Mechanisms used in map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A hierarchical organization of communities + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) for global text summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The cost associated with the number of tokens used in a text generation task + e4d9b12cf2b4c691c74019eefff4fb39 + + + TECHNOLOGY + An implementation of Graph RAG approaches using the Python programming language + e4d9b12cf2b4c691c74019eefff4fb39 + + + PERSON + A person who contributed to the work mentioned in the acknowledgements + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The rates at which fabrications occur in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The expected number of queries over the lifetime of a specific dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The benefits or value obtained from using a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Different methods related to retrieval-augmented generation that utilize graph structures + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Graph RAG approaches that operate in a more localized manner + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Annotations made on the graph to provide additional information + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Reports generated from community summaries + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + An operation that aggregates information across multiple levels of a hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A mechanism that allows for exploring detailed information by following higher-level summaries + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + The trail of information that guides users to more detailed data + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + The top-level communities in a hierarchical structure + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A graph index organized around entities + e4d9b12cf2b4c691c74019eefff4fb39 + + + TECHNOLOGY + A publicly available implementation of a technology + e4d9b12cf2b4c691c74019eefff4fb39 + + + PERSON + Alonso Guevara Fernández is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Amber Hoak is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Andrés Morales Esquivel is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Ben Cutler is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Billie Rinaldi is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Chris Sanchez is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Chris Trevino is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Christine Caggiano is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + David Tittsworth is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Dayenne de Souza is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Douglas Orbaker is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Ed Clark is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Gabriel Nieves-Ponce is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Gaudy Blanco Meneses is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Kate Lytvynets is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Katy Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Mónica Carvajal is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Nathan Evans is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Richard Ortega is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Rodrigo Racanicci is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Sarah Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Shane Solomon is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + A technical report on GPT-4 published as an arXiv preprint + 086021a89900a39bcb62036981737bfa + + + METHOD + A method for zero-shot knowledge graph question answering described in an arXiv preprint + 086021a89900a39bcb62036981737bfa + + + METHOD + A method for harnessing large language models for advanced causal discovery from data + 086021a89900a39bcb62036981737bfa + + + METHOD + A method incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Achiam is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Adler is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Agarwal is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + L. Ahmad is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + I. Akkaya is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + F. L. Aleman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + D. Almeida is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Altenschmidt is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Altman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Anadkat is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + R. Anil is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Borgeaud is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + Y. Wu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J.-B. Alayrac is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Yu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + R. Soricut is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Schalkwyk is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + A. M. Dai is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + A. Hauth is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Baek is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + A. F. Aji is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + A. Saffari is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + T. Ban is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + L. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + X. Wang is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + H. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + T. Baumel is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + M. Eyal is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + M. Elhadad is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + Baumel, T. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Eyal, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Elhadad, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Blondel, V. D. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Guillaume, J.-L. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Lambiotte, R. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Lefebvre, E. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The journal where the paper "Fast unfolding of communities in large networks" was published + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Brown, T. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Mann, B. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Ryder, N. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Subbiah, M. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Kaplan, J. D. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dhariwal, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Neelakantan, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Shyam, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Sastry, G. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Askell, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + "ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS" is a prominent conference where significant papers in the field of Natural Language Processing and Information Retrieval are presented. Notable papers presented at this conference include "Language models are few-shot learners" and "Retrieval-augmented generation for knowledge-intensive NLP tasks." Additionally, it is also the journal where the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" was published. This conference and journal serve as key platforms for disseminating cutting-edge research in neural information processing systems. + 58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,b1bbda43309e8e0e2175ea034aa88e13 + + + PERSON + Cheng, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Luo, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Chen, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Liu, L. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Zhao, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory"Zhao, D. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + PERSON + + + PERSON + Yan, R. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dang, H. T. is an author of the paper "Duc 2005: Evaluation of question-focused summarization systems" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The conference where the paper "Duc 2005: Evaluation of question-focused summarization systems" was presented + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Es, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + James, J. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Espinosa-Anke, L. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Schockaert, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Feng, Z. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Feng, X. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Yang, M. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Qin, B. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Fortunato, S. is an author of the paper "Community detection in graphs" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The journal where the paper "Community detection in graphs" was published + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Gao, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Xiong, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models. The paper provides a comprehensive survey of the methodologies and applications of retrieval-augmented generation, highlighting its significance in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Gao, X. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Jia, K. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant area of research within the domains of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Pan, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Bi, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dai, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance and capabilities of large language models, a significant area of research within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Sun, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Wang, H. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Goodwin, T. R. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Savery, M. E. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Demner-Fushman, D. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + CONFERENCE + COLING (International Conference on Computational Linguistics) is the conference where the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" was presented + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + He, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Tian, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Sun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Chawla, N. V. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Laurent, T. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + LeCun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Bresson, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Hooi, B. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jacomy, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Venturini, T. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Heymann, S. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Bastian, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PUBLICATION + PLOS ONE is the journal where the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" was published + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jin, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Yu, Z. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jiao, P. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Pan, S. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + He, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Wu, J. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Philip, S. Y. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Zhang, W. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PUBLICATION + IEEE Transactions on Knowledge and Data Engineering is the journal where the paper "A survey of community detection approaches: From statistical modeling to deep learning" was published + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Kang, M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Kwak, J. M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Baek, J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Hwang, S. J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Khattab, O. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Santhanam, K. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Li, X. L. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hall, D. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text, indicating its relevance within the domain of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Liang, P. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Liang, P. contributed to the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP," which explores the integration of retrieval and language models to enhance knowledge-intensive tasks in NLP. Additionally, Liang, P. authored the paper "Lost in the middle: How language models use long contexts," which investigates the utilization of extended contexts by language models. These contributions highlight Liang, P.'s significant role in advancing the understanding and application of language models in complex NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Potts, C. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Zaharia, M. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kim, G. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kim, S. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Jeon, B. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Park, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kang, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Klein, G. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Moon, B. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hoffman, R. R. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The journal where the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" were published + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Koesten, L. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Gregory, K. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Groth, P. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Simperl, E. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The journal where the paper "Talking datasets–understanding data sensemaking behaviours" was published + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kuratov, Y. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Bulatov, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Anokhin, P. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Sorokin, D. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Sorokin, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Burtsev, M. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + TECHNOLOGY + Langchain graphs is a technology developed by LangChain + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Laskar, M. T. R. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" and also contributed to the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models." Both works highlight Laskar's expertise in leveraging transformer models and transfer learning techniques to enhance the performance of query-focused abstractive text summarization, demonstrating a significant contribution to the field of Natural Language Processing and Information Retrieval. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hoque, E. is an author of two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning to improve the effectiveness of transformer models in query-focused abstractive summarization tasks. Both works contribute to advancing the understanding and application of transformer models in specialized summarization contexts. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Huang, J. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The conference where the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" was presented + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + arXiv preprint refers to a preprint of a paper that is available on the arXiv repository + 71f6daf11e64e5273a3847d46bf228e1 + + + EVENT + The 33rd Canadian Conference on Artificial Intelligence, held in Ottawa, ON, Canada, from May 13–15, 2020 + 6cd82819982879bd164547d2773ba5c7 + + + EVENT + The 2020 edition of the Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + PUBLISHER + Springer is the publisher of the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Huang, J. X. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + PUBLICATION + The journal where the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" was published + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lewis, P. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Perez, E. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Piktus, A. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Petroni, F. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks"Petroni, F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + PERSON + + + PERSON + Karpukhin, V. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Goyal, N. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Küttler, H. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lewis, M. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Yih, W.-T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Rocktäschel, T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, N. F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lin, K. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Hewitt, J. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Paranjape, A. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Bevilacqua, M. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, Y. is an author of the paper "Hierarchical transformers for multi-document summarization" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lapata, M. is an author known for significant contributions to the field of Natural Language Processing and Information Retrieval. Notably, Lapata, M. has authored the paper "Hierarchical transformers for multi-document summarization," which explores advanced techniques in summarizing information from multiple documents using hierarchical transformer models. Additionally, Lapata, M. has contributed to the paper "Text summarization with latent queries," which delves into innovative methods for summarizing text by leveraging latent query representations. These works highlight Lapata, M.'s expertise and active research in the domain of text summarization, showcasing a focus on developing sophisticated models and methodologies to enhance the efficiency and accuracy of summarization tasks. + 6cd82819982879bd164547d2773ba5c7,fc4b27d64f055b7fc30176ba110dd02e + + + TECHNOLOGY + LlamaIndex Knowledge Graph Index is a technology developed by LlamaIndex + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Manakul, P. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liusie, A. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Gales, M. J. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Mao, Y. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + He, P. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, X. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Shen, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Shen, Y.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Gao, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Han, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Chen, W. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Chen, W.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Martin, S. is an author of the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing a comprehensive, open-source solution for the layout of large graphs, which is a critical task in the visualization and analysis of complex networks. The toolbox aims to facilitate the understanding and interpretation of large-scale graph data, making it a valuable resource for researchers and practitioners in fields such as computational linguistics, information retrieval, and data science. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Brown, W. M. is an author of the paper "Openord: An open-source toolbox for large graph layout." + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + KLAVANS, R. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Boyack, K. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on the development and application of Openord, a comprehensive open-source toolbox designed for the layout of large graphs. The paper likely discusses the methodologies, algorithms, and practical implementations of the toolbox, contributing to the fields of graph theory and data visualization. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + EVENT + The conference where the paper "Openord: An open-source toolbox for large graph layout" was presented + 833e7d67dcd30790b26b71c9b5306f6b + EVENT + + + TECHNOLOGY + GPT-4 is a large language model used in Microsoft's study on scientific discovery + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + + + TECHNOLOGY + Project NaLLM is a project developed by Neo4J + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + + + PERSON + Newman, M. E. is the author of the paper "Modularity and community structure in networks" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PUBLICATION + The journal where the paper "Modularity and community structure in networks" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + + + PERSON + Ram, O. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Levine, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Dalmedigos, I. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Muhlgay, D. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shashua, A. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Leyton-Brown, K. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shoham, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PUBLICATION + The journal where the paper "In-context retrieval-augmented language models" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + + + PERSON + Ranade, P. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Joshi, A. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Sarthi, P. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Abdullah, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Tuli, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Khanna, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Goldie, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Manning, C. D. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" and the paper "Raptor: Recursive abstractive processing for tree-organized retrieval". These contributions highlight Manning's involvement in advancing the fields of Natural Language Processing and Information Retrieval, particularly in the areas of multi-hop question answering and recursive abstractive processing. + 833e7d67dcd30790b26b71c9b5306f6b,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Scott, K. is associated with "Behind the Tech" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shao, Z. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Gong, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Huang, M. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + + + PERSON + Duan, N. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + + + PERSON + Su, D. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Xu, Y. is an author of multiple academic papers in the field of Natural Language Processing and Information Retrieval. Notably, Xu, Y. contributed to the paper titled "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," which addresses the management of scholarly information related to COVID-19 through advanced question answering and summarization techniques. Additionally, Xu, Y. co-authored the paper "Text summarization with latent queries," which explores innovative methods for text summarization by leveraging latent queries. These contributions highlight Xu, Y.'s expertise and active involvement in developing sophisticated systems for information retrieval and summarization. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yu, T. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Siddique, F. B. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Barezi, E. J. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Fung, P. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Tang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Yang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Touvron, H. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Martin, L. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Stone, K. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Albert, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Almahairi, A. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Babaei, Y. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bashlykov, N. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Batra, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bhargava, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bhosale, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Traag, V. A. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Waltman, L. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Van Eck, N. J. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PUBLICATION + Scientific Reports is the journal where the paper "From Louvain to Leiden: guaranteeing well-connected communities" was published + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trajanoska, M. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Stojanov, R. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trajanov, D. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trivedi, H. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Balasubramanian, N. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Khot, T. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Sabharwal, A. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Wang, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Liang, Y. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Meng, F. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Sun, Z. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Shi, H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Li, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through their work on evaluating language models. Specifically, Li, Z. has co-authored the paper titled "Is ChatGPT a Good NLG Evaluator? A Preliminary Study," which explores the effectiveness of ChatGPT as a natural language generation evaluator. Additionally, Li, Z. has co-authored another paper, "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which examines the performance of large language models in evaluative roles using specific benchmarking tools. These contributions highlight Li, Z.'s active involvement in advancing the understanding and assessment of language models within the academic community. + 8d87efac8c50cf20cdf26bf61e5e2035,b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Xu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Qu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhou, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wang, S. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" and also contributed to the paper "Is chatgpt a good nlg evaluator? a preliminary study." These works indicate Wang, S.'s involvement in cutting-edge research within the fields of federated search, retrieval augmented generation, and natural language generation evaluation, showcasing a focus on both the technical and evaluative aspects of Natural Language Processing and Information Retrieval. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Khramtsova is an author mentioned in the text + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Khramtsova, E. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhuang, S. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through multiple academic papers. Notably, Zhuang, S. co-authored the paper titled "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," which explores the evaluation of federated search systems within the framework of retrieval-augmented generation. Additionally, Zhuang, S. co-authored another significant paper, "Judging llm-as-a-judge with mt-bench and chatbot arena," which delves into the assessment of large language models (LLMs) using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Zhuang, S.'s active involvement in advancing research in federated search and the evaluation of LLMs. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zuccon, G. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wang, Y. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Lipka, N. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Rossi, R. A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Siu, A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhang, R. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Derr, T. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yang, Z. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Qi, P. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhang, S. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Bengio, Y. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Cohen, W. W. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Salakhutdinov, R. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + CONFERENCE + The conference where the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" was presented + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yao, J.-g. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wan, X. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Xiao, J. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PUBLICATION + The journal where the paper "Recent advances in document summarization" was published + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yao, L. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models"Yao, L. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Peng, J. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Mao, C. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Luo, Y. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhang, J. is an author of the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhang, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Gan, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Wang, C. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zheng, L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zheng, L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Zheng, L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools like MT-Bench and Chatbot Arena. These contributions highlight Zheng, L.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR domains. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Chiang, W.-L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Chiang, W.-L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Chiang, W.-L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Chiang, W.-L.'s active involvement in advancing the understanding and capabilities of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Sheng, Y. is an author known for contributing to the field of Natural Language Processing and Information Retrieval. Notably, Sheng, Y. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Sheng, Y. has contributed to the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Sheng, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic and technical community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Wu, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Wu, Z. co-authored the paper titled "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Wu, Z. is also credited with co-authoring the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Wu, Z.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhuang, Y. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zhuang, Y. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness of knowledge graphs. Additionally, Zhuang, Y. has also authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Zhuang, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the domain. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Lin, Z. is an author of the paper "Exploring large language models for knowledge graph completion" and also contributed to the paper "Judging LLM-as-a-judge with MT-Bench and Chatbot Arena." These works indicate Lin, Z.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the application of large language models for tasks such as knowledge graph completion and the evaluation of language models in judgment tasks. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Li, D. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant research. Notably, Li, D. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Li, D. has also co-authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Li, D.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Xing, E. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Xing, E. contributed to the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Xing, E.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + TECHNOLOGY + Chatbot Arena is a platform or tool used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Darren Edge and Ha Trinh co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Steven Truitt and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Steven Truitt is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Jonathan Larson is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 4.0 + RAG (Retrieval-Augmented Generation) and LLM (Large Language Models) are closely intertwined in the domain of Natural Language Processing and Information Retrieval. RAG is employed to enhance the capabilities of LLMs by enabling them to retrieve pertinent information from external knowledge sources. This symbiotic relationship allows LLMs to generate and assess text more effectively. Specifically, RAG leverages the power of LLMs to access and utilize relevant data, thereby augmenting the overall performance and accuracy of text generation tasks. + e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 7.0 + Graph RAG is a specific implementation of RAG that combines the strengths of RAG with graph-based text indexing. This method leverages the natural modularity of graphs to partition data, facilitating global summarization. As a specialized approach within the RAG framework, Graph RAG enhances the capabilities of RAG by integrating graph structures to improve the efficiency and effectiveness of text data processing and summarization. + 21e52bc06a82796b1f4bcd73edda1f2a,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Query-Focused Summarization is a task that RAG fails to address effectively + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + RAG retrieves relevant information from an external knowledge source + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Naive RAG is a specific implementation of RAG + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Ram et al., 2023 discusses RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Naïve RAG is a basic form of RAG + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Modular RAG is an advanced form of RAG + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used in various RAG tasks such as knowledge graph creation and completion + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Trajanoska et al. discusses using LLMs for knowledge graph creation, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Yao et al. discusses using LLMs for knowledge graph completion, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Ban et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Zhang et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Gao et al. discusses advanced RAG where the index is a knowledge graph + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + KAPING is a method where the index is a knowledge graph, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + G-Retriever is a method where subsets of the graph structure are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Graph-ToolFormer is a method where derived graph metrics are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + SURGE is a method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + FABULA is a method where retrieved event-plot subgraphs are serialized using narrative templates, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Wang et al. discusses a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Sensemaking questions are used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The evaluation of RAG systems focuses on corpora in the region of 1 million tokens + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Trade-offs are considerations involved in building a graph index for RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + A graph index is a data structure used in RAG systems to organize and retrieve information + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Performance of RAG systems varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Different data types are used in RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Dataset sizes affect the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Evaluation is the process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Corpora are collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Different question types are used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Target metrics are specific measures used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 4.0 + Graph RAG utilizes Large Language Models (LLMs) to construct a graph-based text index, enabling the generation of summaries and the answering of queries. In this approach, LLMs play a crucial role in analyzing and generating text based on the information retrieved through the graph structure. Additionally, LLMs leverage the Graph RAG framework to provide comprehensive overviews of public figures in the entertainment industry. This integration of LLMs within Graph RAG enhances the system's ability to process and synthesize complex text data, making it a powerful tool for information retrieval and natural language processing tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Text chunks are processed using LLM to extract elements of a graph index + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM is used to extract elements of a graph index from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + LLM (Large Language Model) and Few-Shot Examples are closely related in the context of Natural Language Processing and Information Retrieval. Few-shot examples are provided to the LLM for in-context learning, which helps tailor the extraction prompt. This technique is particularly useful for improving the performance of the LLM in specialized domains. By leveraging a small number of examples, the LLM can better understand and adapt to specific tasks, thereby enhancing its overall effectiveness in extracting and processing information within those specialized areas. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM extracts named entities from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Kuratov et al. (2024) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Liu et al. (2023) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM prompts are instructions given to the LLM for extracting elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Recall degradation occurs with longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + The extraction process involves using LLM to identify and extract elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Default prompt is the standard set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Secondary extraction prompt is an additional set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + The LLM uses covariate prompts to extract additional attributes associated with detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM uses multiple rounds of gleanings to ensure no entities are missed + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Logit bias is used to force a yes/no decision from the LLM during entity extraction + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM extracts element instances from source texts + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM detects and summarizes communities of entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + LLM generates intermediate answers and scores for each chunk + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + LLM generates a helpfulness score for each answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + LLM is used to generate questions for evaluating the Podcast Transcripts dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + LLM is used to generate questions for evaluating the News Articles dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + LLM uses Naive RAG to list public figures mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + LLM-generated responses are evaluated using assessment metrics + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + LLM-generated responses are evaluated using specific questions + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Ad-hoc LLM use involves the use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + LLMs are used for knowledge graph creation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph completion + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for the extraction of causal graphs + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph creation as per Trajanoska et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph completion as per Yao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for the extraction of causal graphs as per Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is proposed as a method to combine the strengths of RAG and QFS + e8d83e6e7a7c0f57b218cef24976b745 + + + 7.0 + Graph RAG is an approach that utilizes community summaries in various capacities to enhance its functionality. Specifically, community summaries are generated within the Graph RAG framework and are employed to generate partial responses. These summaries are also used to compare against source texts, serving as a form of self-memory for the system. By leveraging community summaries, Graph RAG aims to improve the comprehensiveness and diversity of answers. Additionally, Graph RAG incorporates summaries of root-level communities within an entity-based graph index, further enriching its analytical capabilities. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is designed to handle global sensemaking questions over large datasets + e8d83e6e7a7c0f57b218cef24976b745 + + + 2.0 + Graph RAG is implemented in Python. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The open-source implementation of Graph RAG will be available at this URL + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Graph RAG uses an entity knowledge graph to index text + e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG is an approach that is evaluated for its comprehensiveness, a target quality used to assess its effectiveness. The method aims to improve the comprehensiveness of generated answers, ensuring that the information provided is thorough and complete. This evaluation metric is crucial in determining the success of Graph RAG in producing detailed and accurate responses. + 21e52bc06a82796b1f4bcd73edda1f2a,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG is an approach in the domain of Natural Language Processing and Information Retrieval that focuses on improving the diversity of generated answers. Diversity, in this context, is a target quality used to evaluate the performance of the Graph RAG approach. By enhancing the diversity of responses, Graph RAG aims to provide a broader range of answers, thereby improving the overall effectiveness and robustness of the system. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG uses a knowledge graph for global summarization + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Community detection algorithms are used in the Graph RAG approach to partition graphs + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Podcast transcripts are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + News articles are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 2.0 + Graph RAG is evaluated using the target quality of Empowerment. Empowerment is specifically utilized to assess Graph RAG's capability in aiding users to achieve an informed understanding. This evaluation metric underscores the importance of user comprehension and the effectiveness of the Graph RAG approach in facilitating informed decision-making processes. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Graph RAG is compared to naive RAG in the evaluation. In this comparison, Graph RAG outperformed naive RAG on comprehensiveness and diversity. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Graph RAG is compared to global map-reduce summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Query-focused summarization is a method used in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Activity-centered sensemaking questions are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 2.0 + The "Graph RAG" approach is evaluated in terms of its performance by considering "Token Costs." Token costs are measured to assess the efficiency of the Graph RAG method, indicating that the computational expense associated with processing tokens is a critical factor in determining the overall effectiveness of this approach. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Data flow describes the high-level process of the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 3.0 + Design parameters are key settings in the Graph RAG approach and significantly influence the Graph RAG approach and pipeline. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Graph RAG uses global summarization to summarize information from a large dataset + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG aims to answer specific queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG uses a corpus for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Activity-centered sensemaking is used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Real-world datasets are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG is compared to source text summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Low-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Intermediate-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + High-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + The Graph RAG approach involves a specific pipeline for processing and summarizing text + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Techniques are specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Implementation details are specific configurations used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Graph RAG is a specific implementation of RAG systems + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + Graph RAG is a system that utilizes root-level community summaries, denoted as C0, to answer user queries. C0 represents these root-level community summaries within the Graph RAG analysis, serving as a foundational element in the system's ability to map out relationships and understand the structural dynamics within specialized communities. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses high-level community summaries (C1) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses intermediate-level community summaries (C2) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 2.0 + Graph RAG utilizes low-level community summaries, represented by C3, to answer user queries. C3 plays a crucial role in the Graph RAG analysis by providing detailed summaries of community structures, which are essential for effectively addressing user inquiries. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + 2.0 + Graph RAG is a key entity in the analysis, serving both as a condition being compared and as a tool for comparing multiple conditions. This dual role highlights its significance in the study, where it functions not only as a subject of comparison but also as a methodological framework for evaluating other conditions. The analysis likely involves a detailed examination of various conditions, with Graph RAG playing a central role in facilitating these comparisons. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses different levels of graph communities to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The Graph RAG mechanism uses an LLM evaluator for head-to-head comparison + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Graph RAG is a multi-stage mechanism + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Graph RAG mentions Taylor Swift as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Travis Kelce as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Britney Spears as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Justin Timberlake as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG is determined to be the winner based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Answer 1 was generated using the Graph RAG method, which provides a comprehensive and structured overview of public figures across various sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Graph RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Graph RAG is compared with source texts for answer comprehensiveness and diversity + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + TS represents source text summarization in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Root-level summaries are used in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Answer comprehensiveness is used to evaluate the performance of Graph RAG + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Win rate is used to measure the success rate of Graph RAG in providing comprehensive and diverse answers + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Element extraction prompts are used in Graph RAG to retain specific details in the index + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Graph RAG incorporates the concept of self-memory + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of iterative retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of federated retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts used in multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts used in multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG uses a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of a tree of clarifications + f35de4d9fb65f1d5a392064b20545c19 + + + 3.0 + Graph RAG utilizes a self-generated graph index. This self-generated graph index is a crucial component of Graph RAG, enabling it to efficiently manage and retrieve information within its graph-based framework. The use of a self-generated graph index suggests that Graph RAG has an inherent capability to construct and maintain its indexing structure, which likely enhances its performance and adaptability in handling complex data relationships. + e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Gao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Cheng et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Mao et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Shao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Wang et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Su et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Feng et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Trivedi et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Khattab et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Sarthi et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Kim et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG generates community answers in parallel + f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is compared to a graph-free approach for global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG is compared to map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses rich text annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses a hierarchical community structure + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can operate using embedding-based matching + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can be part of hybrid RAG schemes + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can employ map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can extend operations across the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Alonso contributed to the work on Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG includes local graph RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses an entity-based graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + 2.0 + NebulaGraph launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Community summaries are used to generate partial responses + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Community summaries are created from graph communities + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Community answers are created from community summaries. These answers are generated by synthesizing information from the summaries provided by the community, ensuring that the responses are comprehensive and reflective of the collective knowledge and insights within the community. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Domain-tailored summarization is used to create community summaries + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Community descriptions are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Partial answers are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Community summaries are created for each level in the hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Community summaries are useful for understanding the global structure and semantics of the dataset + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Community summaries are used to answer global queries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are generated from root communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are generated from sub-communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are added to the LLM context window until the token limit is reached + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Global answers are generated from community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The level of summary detail affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The scope of information affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are used for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Community summaries are divided into chunks of pre-specified token size + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Summary detail and scope affect the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are divided into chunks + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Community summaries are prepared to answer user queries + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Intermediate answers are generated from community summaries + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Community summaries are part of the graph community hierarchy + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Community summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Summaries of root-level communities are used in Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Global sensemaking questions are evaluated over datasets in the 1 million token range + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Global sensemaking questions are directed at an entire text corpus + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The Python-based implementation of Graph RAG approaches will be available at this URL + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Query-focused summarization is used to produce the global answer + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Map-reduce is used for query-focused summarization of an entire corpus + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Query-focused summarization is used for answering global queries + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + An entity knowledge graph is derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + + + 2.0 + In the domain of Natural Language Processing and Information Retrieval, "SOURCE DOCUMENTS" and "TEXT CHUNKS" are closely related entities. Text chunks are extracted from source documents, specifically for processing in the Graph RAG (Retrieval-Augmented Generation) approach. This method involves breaking down the source documents into smaller, manageable pieces, or text chunks, which can then be effectively utilized in various computational processes, enhancing the efficiency and accuracy of information retrieval and generation tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Intermediate-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Low-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Document corpus consists of source documents being processed + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Partial responses are summarized to generate a final response + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The LLM evaluator assesses answers based on the comprehensiveness metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Naive RAG is evaluated for comprehensiveness + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Comprehensiveness is a metric used to determine the decision + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) was universally better for comprehensiveness + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The final evaluation prioritized comprehensiveness in answers + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Global approaches achieved higher comprehensiveness win rates + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The LLM evaluator assesses answers based on the diversity metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on diversity + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The final evaluation prioritized diversity in answers + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Global approaches achieved higher diversity win rates + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Human endeavors rely on sensemaking to understand and reason about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Human endeavors rely on analyzing document collections for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + LLMs are used to automate sensemaking in complex domains + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Microsoft uses LLMs for automating sensemaking in scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Ranade uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Joshi uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + LLM prompts are used to tailor the responses of large language models + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Ranade and Joshi discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + GPT is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Llama is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Gemini is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Kuratov et al., 2024, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Liu et al., 2023, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Sensemaking is applied in the domain of scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Sensemaking is applied in the domain of intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Klein defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Klein et al. defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Element instances are extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Entity references are extracted from text chunks during processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Element instances are converted into element summaries by the LLM (Large Language Model). Element summaries are created from element instances, indicating a transformation process facilitated by the LLM. This process involves the LLM taking detailed element instances and generating concise element summaries, which encapsulate the essential information from the instances. + 2c6ed90897310eea2f28e33fff1c32b0,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Covariates are additional attributes associated with extracted element instances + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Domain-tailored summarization is used to create element summaries + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Element summaries include descriptions of entity nodes + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries include descriptions of relationship edges + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries include descriptions of claim covariates + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries are used to understand the structure and semantics of graph communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Element summaries include descriptions of nodes + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Element summaries include descriptions of edges + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Element summaries include descriptions of covariates + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Sub-community summaries are used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Community detection is a technique used to identify graph communities. Graph communities are groups of nodes within a graph that are more densely connected to each other than to the rest of the graph. This process of identifying such communities is crucial for understanding the structural dynamics and relationships within complex networks, particularly in the domain of Natural Language Processing and Information Retrieval. By leveraging community detection algorithms, researchers can uncover hidden patterns and insights within large datasets, facilitating more effective data analysis and interpretation. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Global answer is created from community answers + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Global answers are generated in response to user queries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Global answer is generated by sorting intermediate answers based on helpfulness scores + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Intermediate answers are combined to form the global answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + The final context window is used to generate the global answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Graph RAG pipeline operates at indexing time + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Graph RAG pipeline operates at query time + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Nodes are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Edges are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Covariates are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Leiden method is used in the graph RAG pipeline for community detection + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Graph RAG pipeline uses the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + The Podcast dataset graph consists of 8564 nodes + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The News dataset graph consists of 15754 nodes + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The Podcast dataset graph consists of 20691 edges + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The News dataset graph consists of 19520 edges + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Traag contributed to the development of the Leiden method + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Traag et al. are the authors of the Leiden algorithm and developed the Leiden method. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Leiden is a specific type of community detection algorithm used in various analytical pipelines. It is designed to identify and map out the structural dynamics within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. The algorithm is known for its efficiency and accuracy in detecting community structures, making it a valuable tool for researchers and practitioners in the field. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Leiden is known for its ability to recover hierarchical community structures efficiently + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The Leiden algorithm is used to detect graph communities in the MultiHop-RAG + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Figure 3 shows graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Lewis contributed to the development of the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Lewis et al. developed the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Lewis et al., 2020, are the authors who established the RAG approach + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Kevin Scott is the CTO of Microsoft + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + Microsoft conducted a study on the impact of large language models on scientific discovery using GPT-4 + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Preprint is available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Baumel, T. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Elhadad, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Es, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + James, J. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Espinosa-Anke, L. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Schockaert, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, Z. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, X. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Zhao, D. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Yang, M. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Qin, B. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + LangChain is an organization that has published on arXiv + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Wang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zuccon, G. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, R. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Derr, T. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Xu, Y. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lapata, M. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, J. published the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Gan, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yao, L. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, C. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Chiang, W.-L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Sheng, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wu, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lin, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Li, D. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Xing, E. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Preprint is classified under cs.CL on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Preprint was submitted on 24 Apr 2024 + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Preprint has the identifier 2404.16130v1 on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Community detection results in the partition of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The pipeline includes a step for community detection + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 2.0 + Dang, 2006, is the author who established the QFS approach + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Baumel et al., 2018, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Laskar et al., 2020, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Yao et al., 2017, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Goodwin et al., 2020, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Laskar et al., 2022, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Liu and Lapata, 2019, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Achiam et al., 2023, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Brown et al., 2020, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Touvron et al., 2023, are the authors who worked on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Anil et al., 2023, are the authors who worked on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Modularity is an inherent quality of knowledge graphs + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Brown et al. (2020) discuss in-context learning with few-shot examples + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Kuratov et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Liu et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Louvain is a type of community detection algorithm + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Community detection algorithms are used to partition the graph index into communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Fortunato has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Jin et al. have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + HotPotQA dataset is used to evaluate the entity extraction prompt with gpt-4-turbo + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Yang et al. (2018) introduced the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Yang et al. are the authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + GPT-4-Turbo was tested with varying context window sizes + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Tech journalist uses podcast transcripts to look for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + + + 3.0 + Kevin Scott is a participant in the podcast conversations compiled in the Podcast Transcripts dataset. His conversations are included as part of the podcast transcripts, contributing to the overall content and discussions captured within this dataset. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Technology leaders participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + RAG systems are used to evaluate the Podcast Transcripts dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + C0 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C1 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C2 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C3 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + TS is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + SS is a category used in the analysis of podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Units are used to measure the context in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Both are datasets used in the analysis + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Educator uses news articles to incorporate current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + RAG systems are used to evaluate the News Articles dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + C0 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Map-reduce is the method used in the text summarization condition + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The LLM evaluator assesses answers based on the empowerment metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on empowerment + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Empowerment has an average win rate of 51.3% + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Naive RAG mentions Taylor Swift as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Travis Kelce as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Britney Spears as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Justin Timberlake as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG is determined to be the loser based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Global approaches consistently outperformed the naive RAG + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Naive RAG produces the most direct responses + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + SS represents naive RAG in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Gao et al., 2023 discusses naive RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Community partitions enable divide-and-conquer global summarization + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Global summarization can be performed using a graph-free approach + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Source texts are used in global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Final global answer is generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Short descriptions are used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Low-level community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + The use of rich descriptive text for homogeneous nodes in a graph index aligns with the capabilities of LLMs + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Graph indexes differ from typical knowledge graphs in their use of rich descriptive text instead of concise knowledge triples + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The graph index supports condition C0 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C1 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C2 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C3 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index was created using generic prompts for entity and relationship extraction + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Few-shot examples tailored to the domain of the data were used in the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The decision to build a graph index depends on the expected number of lifetime queries per dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The decision to build a graph index depends on the value obtained from it + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The decision to build a graph index depends on the value obtained from other graph-related RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Recall measures the completeness of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Precision measures the accuracy of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to tailor the default prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to tailor the secondary extraction prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of science + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of medicine + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of law + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + A single extraction round is part of the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Domain refers to the specific area of knowledge of the document corpus + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Covariate prompts are used to extract claims linked to detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Source text span is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Start date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + End date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Description is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Subject is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Object is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Communities of entities help manage variations in a noisy graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Common entities are described using rich descriptive text for homogeneous nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + LLMs are used to generate metrics for evaluating natural language generation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Wang et al. (2023) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Zheng et al. (2024) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Relationship edges connect homogeneous nodes in a graph + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Edge weights represent the normalized counts of detected relationship instances on relationship edges + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Each level of the hierarchical community structure provides a community partition + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 2.0 + The hierarchical community structure is a framework used to organize and understand the relationships and dynamics within specialized communities. Root communities are an integral part of this structure, serving as the top-level communities. These root communities form the foundational layer in the hierarchical community structure, providing a basis for further subdivision and organization of more specific sub-communities. This hierarchical approach allows for a systematic analysis of complex networks, facilitating a deeper understanding of the interconnections and dependencies within the domain of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + The hierarchical community structure is a framework that organizes communities into different levels, with sub-communities representing the lower-level communities within this structure. Sub-communities are integral components of the hierarchical community structure, indicating that they are nested within larger communities and contribute to the overall organization and dynamics of the community. This hierarchical arrangement allows for a more detailed and nuanced understanding of the relationships and interactions within the community, facilitating more effective analysis and mapping of complex text data, particularly in specialized domains such as Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community levels are part of the hierarchical community structure + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The Leiden algorithm is used to detect communities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + OpenORD is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Force Atlas 2 is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Nodes represent entities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Edges represent connections between nodes in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Covariates are variables linked to nodes and edges in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Tang and Yang are the authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Questions are generated based on the target datasets + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + N represents the number of test questions per dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Root communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Level 0 represents the root-level communities in the hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Reports provide detailed information about specific subtopics within sub-communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Sub-communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Level 1 represents sub-communities within the root-level communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Partitions can be organized into a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Level 0 is the root level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Level 1 is a sub-level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The token limit defines the maximum number of tokens in the LLM context window + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Prominence is used to prioritize community edges + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Combined source and target node degree is used to measure prominence + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Chunks are divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Helpfulness scores are assigned to intermediate answers + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in episodes dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in how guests perceive the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in discussions about the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in discussions about collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in how news articles address the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in highlighting the importance of health literacy through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + The size of the context window and the prompts used for answer generation are the same across all conditions + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The task is an activity or goal that the user aims to achieve + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Questions are generated based on the user's task + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Datasets were used in combination with questions for the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Questions were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + + + 2.0 + Zheng et al. are the authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Zheng, L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Xing, E. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + MT-Bench and Chatbot Arena are both tools used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 2.0 + Koesten et al. authored a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + Xu and Lapata authored a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Text summarization method applies a map-reduce approach directly to source texts (TS) + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Text summarization is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Semantic search RAG is a na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached (SS) + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Semantic search RAG is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C0 uses root-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C0 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C0 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C0 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 uses high-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C1 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C1 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C1 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 uses intermediate-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C2 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C2 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C2 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 uses low-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C3 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C3 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C3 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + TS is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + SS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The graph indexing process used 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + A graph was created for the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Intermediate-level summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + The graph indexing process used 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + A graph was created for the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Datasets were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Wang et al., 2023a discusses the state-of-the-art results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Zheng et al., 2024 discusses the competitive results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Zheng et al., 2024 discusses the LLM-as-a-judge method + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Embedding-based matching is used to match user queries + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Query-time LLM use was evaluated with different context window sizes + 4c855404ee3d3c94aa2136f1513c666f + + + 2.0 + The **CONTEXT WINDOW SIZE** and **FINAL EVALUATION** are closely related in the given data. A fixed context window size of 8k tokens was used for the final evaluation. This indicates that during the final evaluation phase, the system or model was configured to process and analyze text data within a predefined window of 8,000 tokens, ensuring consistency and standardization in the evaluation process. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Natural Language Generation achieves state-of-the-art results + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation achieves competitive results + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation is compared against human judgements + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation can generate reference-based metrics + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation can measure qualities in a reference-free style + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Es et al., 2023 discusses the RAGAS method + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates context relevance + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates faithfulness + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates answer relevance + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator assesses answers based on the directness metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Table 2 shows an example of LLM-generated assessment + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses a head-to-head comparison approach + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator assesses answers based on target metrics + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses a control metric for validity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator accounts for stochasticity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses mean scores from multiple comparisons + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Directness is used to evaluate the straightforwardness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The question asks about public figures mentioned in entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Public figures are repeatedly mentioned across various entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Answer 1 covers a wide range of public figures from different sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Controversies involve public figures and impact public discourse. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Entertainment articles cover topics related to the entertainment industry + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Taylor Swift is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Travis Kelce is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Britney Spears is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Justin Timberlake is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Taylor Swift is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Travis Kelce is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Britney Spears is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Justin Timberlake is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Actors and Directors are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Musicians and Executives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Athletes and Coaches are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Influencers and Entrepreneurs are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Public Figures in Controversy are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Film is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Television is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Music is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Sports is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Digital Media is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Cultural Narratives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Trends are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Social Discussions are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Public Discourse is a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Reference-based metrics require gold standard answers + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Gold standard answers are lacking for sensemaking questions + 322e02986c8724eedbcf3ebfa20b989c + + + 3.0 + End users play a crucial role in the validation process of sensemaking questions and target metrics. Sensemaking questions are specifically validated with end users to ensure their relevance and accuracy. This collaborative approach ensures that the questions and metrics are aligned with the needs and expectations of the end users, thereby enhancing the overall effectiveness and applicability of the sensemaking process. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Target metrics are validated with end users + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The control metric is used as an indicator of validity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Taylor Swift is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Taylor Swift is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Travis Kelce is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Travis Kelce is a public figure in the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Britney Spears is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Britney Spears is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Justin Timberlake is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Justin Timberlake is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the film sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the television sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on public figures primarily from the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on public figures primarily from the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the digital media sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 cites specific data sources from the News article dataset for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 provides insights into controversies involving public figures and their impact on public discourse. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the gaming sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 cites specific data sources for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 was generated using the Naïve RAG method, which directly lists specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d + + + 2.0 + ANSWER 2 is a generated answer for a question in the NEWS ARTICLE DATASET. It relies heavily on a single source from the NEWS ARTICLE DATASET for data. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Answer 2 relies heavily on a single data source. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Naïve RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The global approach to Graph RAG shows improvements over naïve RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + LLM-generated assessments are used to evaluate the answers produced for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Example question is part of the News article dataset used for analysis + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Head-to-head win rate percentages were used to compare different conditions + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Win rate percentages were used to measure the performance of different conditions + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The overall winner per dataset and metric was determined for each condition + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Self-win rates were shown as the expected 50% for each condition + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The indexing process resulted in the creation of graphs + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Map-reduce summarization requires the highest number of context tokens + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Root-level community summaries require dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + + + 2.0 + Queries are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Self-memory is related to generation-augmented retrieval + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + CAiRE-COVID is a system for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + ITRG is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + IR-CoT is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + DSP is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + RAPTOR is a method for generating a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + The paper by Baek et al. discusses the KAPING method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by He et al. discusses the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Zhang discusses the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Kang et al. discusses the SURGE method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Ranade and Joshi discusses the FABULA method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Both LangChain and LlamaIndex support a variety of graph databases + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LangChain supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LangChain supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 1.0 + LangChain developed Langchain graphs + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + LlamaIndex supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LlamaIndex supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + NaLLM is a method that can create and reason over knowledge graphs in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Neo4J developed Project NaLLM + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + GraphRAG is a method that can create and reason over knowledge graphs in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Manakul et al. discusses the SelfCheckGPT method + 92e93fc6449756c0a60200636b297f65 + + + 1.0 + SelfCheckGPT is an approach mentioned in the work by Manakul et al., 2023 + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + SelfCheckGPT is used to compare fabrication rates + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Embedding-based matching is used to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Hybrid RAG schemes combine embedding-based matching against community reports + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The roll-up operation can be extended using map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The drill down mechanism follows the information scent in the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The global approach to Graph RAG achieves competitive performance at a fraction of the token cost + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The open-source implementation of Graph RAG approaches is Python-based + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The drill down mechanism follows the information scent + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Alonso Guevara Fernández and Amber Hoak both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Sarah Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Shane Solomon both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Adler co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Altenschmidt and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Altenschmidt and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Altman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and S. Borgeaud co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Schalkwyk and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Schalkwyk and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + A. M. Dai and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Baek and A. F. Aji co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Baek and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + A. F. Aji and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + T. Ban and L. Chen co-authored the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + 1.0 + Baumel, T. and Eyal, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Baumel, T. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Baumel, T. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Elhadad, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Guillaume, J.-L. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Guillaume, J.-L. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Guillaume, J.-L. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Lambiotte, R. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Mann, B. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Zhao, D. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Es, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + James, J. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Espinosa-Anke, L. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Schockaert, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, Z. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, X. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Yang, M. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Qin, B. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Gao, Y. and Xiong, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Dai, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Dai, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Sun, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Goodwin, T. R. and Savery, M. E. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Goodwin, T. R. and Demner-Fushman, D. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + 2.0 + Khattab, O. and Santhanam, K. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and highlights their collaborative work in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Khattab, O. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This collaboration is mentioned in the text, highlighting their joint contribution to the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Santhanam, K. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Li, X. L. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hall, D. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Liang, P. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + LIANG, P. and ZAHARIA, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Liang, P. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Potts, C. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Potts, C. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Zaharia, M. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Kim, S. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Park, J. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Park, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kang, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. and Moon, B. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Moon, B. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Moon, B. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoffman, R. R. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Gregory, K. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Groth, P. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Groth, P. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Simperl, E. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kuratov, Y. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Bulatov, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Anokhin, P. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Sorokin, D. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Sorokin, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Burtsev, M. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Laskar, M. T. R. and Hoque, E. co-authored two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning with transformer models to improve the effectiveness of query-focused abstractive summarization. Both works contribute to advancing the application of transformer models in specialized summarization tasks. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Laskar, M. T. R. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Hoque, E. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoque, E. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoque, E. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Hoque, E. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Huang, J. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + The 33rd Canadian Conference on Artificial Intelligence is also known as Canadian AI 2020 + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Springer published the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Huang, J. X. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Perez, E. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Karpukhin, V. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Xu, Y. and Lapata, M. co-authored the paper "Text summarization with latent queries" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Huang, M. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Duan, N. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 3.0 + Martin, S. and Brown, W. M. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with large graph structures, making it a valuable resource for researchers and practitioners in the domain of graph theory and network analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Martin, S. and Klavans, R. co-authored the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Martin, S. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the capabilities and applications of the Openord toolbox, emphasizing its utility in handling extensive graph data efficiently. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Brown, W. M. and Klavans, R. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Brown, W. M. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + KLAVANS, R. and BOYACK, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Newman, M. E. published the paper "Modularity and community structure in networks" in the Proceedings of the National Academy of Sciences + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Levine, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Shashua, A. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Shashua, A. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Leyton-Brown, K. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ranade, P. and Joshi, A. co-authored the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Abdullah, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Yang, Z. and Manning, C. D. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Huang, M. and Duan, N. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Xu, Y. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Siddique, F. B. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Siddique, F. B. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Barezi, E. J. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Tang, Y. and Yang, Y. co-authored the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Martin, L. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bhargava, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bhosale, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Liang, Y. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 2.0 + Li, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Li, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Li, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 2.0 + Xu, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Xu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Qu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Khramtsova, E. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wang, Y. and Lipka, N. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, R. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Qi, P. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Zhang, S. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Bengio, Y. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Cohen, W. W. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Salakhutdinov, R. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Chiang, W.-L. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Xing, E. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + \ No newline at end of file diff --git a/graphfleet/output/graphindex/artifacts/embedded_graph.1.graphml b/graphfleet/output/graphindex/artifacts/embedded_graph.1.graphml new file mode 100644 index 000000000..829426858 --- /dev/null +++ b/graphfleet/output/graphindex/artifacts/embedded_graph.1.graphml @@ -0,0 +1,9715 @@ + + + + + + + + + + + PERSON + Darren Edge is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Ha Trinh is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Newman Cheng is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Joshua Bradley is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Alex Chao is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Apurva Mody is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Steven Truitt is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Jonathan Larson is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Research is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Strategic Missions and Technologies is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Office of the CTO is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + METHOD + RAG (Retrieval-Augmented Generation) is a developing research area with multiple established directions, including knowledge graph creation, completion, and extraction of causal graphs. It is a method used for generating responses in text generation tasks by retrieving relevant information from an external knowledge source to enable large language models to answer questions. This approach incorporates the retrieval of relevant data to augment text generation, producing direct responses in various text generation tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY + LLM (Large Language Model) is a type of artificial intelligence model used for a variety of tasks in the field of Natural Language Processing and Information Retrieval. These tasks include generating and assessing text, entity extraction, summarization, understanding relationships in text, and automating human-like sensemaking and reasoning over large collections of documents. LLMs are also employed to generate intermediate answers and scores for text chunks, process these chunks to extract elements of a graph index, and automate the generation of questions for dataset evaluation. Additionally, LLMs can analyze and generate text based on retrieved information and queries, and they possess a context window that can be exceeded by external datasets. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,2c6ed90897310eea2f28e33fff1c32b0,6f33a085ff3304e5994f7fbb86c881a4,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + METHOD + Graph RAG (Retrieval-Augmented Generation) is a sophisticated method that leverages the natural modularity of graphs to partition data for global summarization tasks. This approach integrates multiple stages and concepts, including knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS), to support human sensemaking over text corpora. It is particularly effective in providing comprehensive and structured overviews of public figures across various sectors of the entertainment industry, as well as generating answers for questions in the News article dataset. + +Graph RAG employs a high-level data flow and pipeline for processing and summarizing text, combining both global and local approaches to optimize token usage in text generation tasks. It uses community summaries to improve answer comprehensiveness and diversity while requiring fewer tokens compared to traditional source text summarization methods. This method has been shown to outperform naive RAG in terms of comprehensiveness and diversity in text generation tasks. + +A specific implementation of Graph RAG involves using four levels of graph communities, incorporating concepts from other systems such as self-memory and parallel generation of community answers. This multi-stage mechanism allows for the comparison of multiple conditions, enhancing the overall effectiveness of the summarization process. + +Graph RAG, launched by NebulaGraph, is a retrieval-augmented generation technology based on knowledge graphs. It combines global summarization with graph-based text indexing to answer questions over private text corpora, making it a versatile tool for various text analysis and summarization applications. + 086021a89900a39bcb62036981737bfa,21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,718017a4871c909420f84b85b8ba969d,833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19,fb3c48579608fa28be585ceb6cd2f0fe + + + METHOD + QFS (Query-Focused Summarization) is a method used to generate summaries based on specific user queries + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + Community summaries are generated summaries of data clusters or communities, used to answer queries and tailored to the domain. These summaries are pre-generated for groups of closely-related entities, particularly in the Graph RAG approach, and are derived from community-generated content to compare with source texts. They are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks. Community summaries provide report-like insights into each community within a hierarchical structure, which is useful for understanding the dataset. They are generated from modular communities in the knowledge graph and cover different levels of each graph community hierarchy, including root-level communities in an entity-based graph index. Additionally, these summaries act as a kind of self-memory for generation-augmented retrieval. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + CONCEPT + Global sensemaking questions are questions that require understanding and summarizing large datasets, often beyond the explicit content of the source texts + e8d83e6e7a7c0f57b218cef24976b745 + + + METRIC + 1 million token range refers to the scale of datasets used in the evaluation of the Graph RAG approach + e8d83e6e7a7c0f57b218cef24976b745 + + + TECHNOLOGY + Python is a programming language used for implementing both global and local Graph RAG approaches. Additionally, Python is utilized to implement the open-source version of the Graph RAG approach. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + + + URL + The URL "HTTPS://AKA.MS/GRAPHRAG" is the location where the open-source, Python-based implementation of Graph RAG approaches will be available. This URL serves as the repository for accessing the open-source implementation of the Graph RAG approach. + e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745 + + + METHOD + Query-Focused Summarization (QFS) is a method used to generate summaries that are relevant to specific user queries. This summarization technique focuses on answering specific queries by utilizing the entire corpus of information available. It is designed to provide concise and relevant information based on the specific needs of the user, ensuring that the generated summaries are directly aligned with the queries posed. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + + + CONCEPT + An external knowledge source is a repository of information that can be accessed to retrieve relevant data for answering questions + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + A text corpus is a large collection of written texts used for analysis and research + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + Source documents are the original texts from which information is extracted, retrieved, or summarized. These documents serve as the foundational input texts for various processing tasks, including the Graph RAG (Retrieval-Augmented Generation) approach. In this context, source documents are critical for extracting and analyzing information, ensuring that the data used in computational linguistics and information retrieval tasks is accurate and comprehensive. + bc9e2c9e369c4108cf4f6dd5f60960f4,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + + + CONCEPT + A partial response is an intermediate answer generated from community summaries before being combined into a final response + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + A final response is the comprehensive answer generated after combining all partial responses + e8d83e6e7a7c0f57b218cef24976b745 + + + METRIC + COMPREHENSIVENESS is a metric used to evaluate the quality of generated responses by measuring how much detail an answer provides to cover all aspects and details of a question. It assesses the completeness and thoroughness of answers, ensuring that they encompass all relevant information. This metric is particularly important in evaluating the summarization approach, focusing on the completeness of the summary. In practical applications, such as evaluating Podcast transcripts and News articles, comprehensiveness has shown win rates between 72-83% and 72-80%, respectively. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + + + METRIC + DIVERSITY is a metric used to evaluate the variety and richness of answers generated in response to a question. It measures how varied and rich an answer is in providing different perspectives and insights. This metric is particularly important in assessing the quality of summarization approaches, focusing on the variety of information included in the summary. DIVERSITY is applied to various types of content, including Podcast transcripts, where win rates range from 75-82%, and News articles, with win rates ranging from 62-71%. It is a crucial target quality for evaluating the effectiveness of different methods in generating diverse and informative responses. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + + + ACTIVITY + Human endeavors refer to activities and efforts across various domains that rely on reading and reasoning about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models designed to understand and generate human-like text, playing a crucial role in automating sensemaking in complex domains. Modern language models, such as GPT, Llama, and Gemini, leverage in-context learning to effectively summarize content. These models are integral to the field of Natural Language Processing and Information Retrieval, enabling sophisticated text analysis and generation capabilities. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + + + DOMAIN + Scientific discovery is a complex domain where sensemaking is applied to understand and generate new knowledge from scientific texts + f0306814bf64f5c9e79603fc6a52f4ea + + + DOMAIN + Intelligence analysis is a complex domain where sensemaking is applied to understand and generate insights from intelligence data + f0306814bf64f5c9e79603fc6a52f4ea + + + PROCESS + SENSEMAKING is the process of understanding and making sense of complex information. It involves understanding connections among people, places, and events to anticipate their trajectories and act effectively. This process is crucial for navigating and interpreting intricate data landscapes, enabling individuals and organizations to make informed decisions based on the relationships and patterns identified within the information. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + TEXT CHUNKS are segments of text that are embedded into a vector space for analysis. These segments are extracted from source documents and are used for processing in the Graph RAG (Retrieval-Augmented Generation) approach. By embedding these text chunks into a vector space, they can be analyzed more effectively, facilitating various natural language processing and information retrieval tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + DATA + Element instances are identified and extracted instances of graph nodes and edges from text chunks. They represent individual occurrences of entities, relationships, and claims extracted from source texts. These specific pieces of information are tailored to the domain, providing a structured representation of the underlying data. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Element summaries are concise representations of element instances, tailored to the domain. They are descriptive texts created by the LLM to summarize entities, relationships, and claims extracted from source texts. These summaries provide detailed descriptions of nodes, edges, and covariates within a community, and are used to understand the structure and semantics of the dataset. In essence, element summaries serve as a tool to encapsulate and convey the intricate details of elements within a graph, facilitating a deeper comprehension of the dataset's structural dynamics and semantic relationships. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Graph communities are groups of elements, including nodes, edges, and covariates, detected within a graph index, primarily used for summarization. These communities consist of groups of nodes that exhibit stronger connections to each other than to nodes outside the group. This structural characteristic allows for the identification and analysis of densely connected subgraphs, which can be crucial for understanding the underlying relationships and dynamics within complex networks. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + COMMUNITY ANSWERS are query-focused summaries derived from community summaries. These answers are generated in parallel from the community summaries and are designed to respond to user queries effectively. Essentially, COMMUNITY ANSWERS serve as responses that synthesize information from community summaries to provide concise and relevant answers to specific questions posed by users. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + DATA + GLOBAL ANSWER is a comprehensive response generated from multiple community summaries to answer a user query. It is the final query-focused summary produced from all relevant community summaries. The final answer is generated by combining intermediate community answers based on their helpfulness scores. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + TIME + Indexing time refers to the time when the graph index is created and elements are summarized + f0306814bf64f5c9e79603fc6a52f4ea + + + TIME + Query time refers to the time when a query is made and the relevant summaries are generated + f0306814bf64f5c9e79603fc6a52f4ea + + + PROCESS + Graph RAG pipeline is a process using an LLM-derived graph index to detect, extract, and summarize nodes, edges, and covariates in source documents + f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + NODES are entities detected in the graph index of source documents. They represent the individual elements or points in a graph. For instance, in the Podcast dataset, there are 8,564 nodes, while the News dataset contains 15,754 nodes. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + EDGES are relationships detected in the graph index of source documents. They represent the connections or links between nodes in a graph. For instance, in the Podcast dataset, there are 20,691 edges, while the News dataset contains 19,520 edges. These edges are crucial for understanding the structural dynamics and relationships within the datasets, providing insights into how different nodes (such as topics, entities, or documents) are interconnected. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Covariates are additional attributes associated with extracted node instances in the graph index. They represent claims or additional information detected in the graph index of source documents. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + LEIDEN is a community detection algorithm renowned for its efficiency in recovering hierarchical community structures. It is widely used to partition graphs into modular communities, effectively grouping elements within a graph index. The algorithm's ability to identify and organize these communities makes it a valuable tool in the analysis of complex networks, particularly within the domains of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Retrieval-Augmented Generation (RAG) is an established approach in the field of Natural Language Processing and Information Retrieval, designed to answer user questions over entire datasets. This method involves retrieving relevant text regions to provide grounding for the generation task, thereby enhancing the accuracy and relevance of the generated responses. By combining retrieval and generation processes, RAG effectively synthesizes and presents pertinent information, making it a powerful tool for handling complex queries and large datasets. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + + + ORGANIZATION + Microsoft is a technology company whose Chief Technology Officer, Kevin Scott, actively participates in podcast conversations. The organization is deeply involved in automating sensemaking in scientific discovery through the use of large language models (LLMs). Notably, Microsoft conducted a study examining the impact of large language models, specifically GPT-4, on scientific discovery. + 1d07b4248c2655081c7af0e373bd70c9,833e7d67dcd30790b26b71c9b5306f6b,f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Ranade is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Joshi is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Klein is an author who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Lewis is an author who contributed to the development of the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Traag is an author who contributed to the development of the Leiden community detection method + f0306814bf64f5c9e79603fc6a52f4ea + + + PUBLICATION + arXiv is a preprint repository where several significant papers in the field of Natural Language Processing and Information Retrieval have been published. It serves as a platform for electronic preprints (known as e-prints) that are approved for publication after moderation, but not full peer review. Notable papers published on arXiv include "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models," "Lost in the middle: How language models use long contexts," "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," "Llama 2: Open foundation and fine-tuned chat models," "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy," "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries," "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions," "Enhancing knowledge graph construction using large language models," "Is chatgpt a good nlg evaluator? a preliminary study," "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt," "Causal graph discovery with retrieval-augmented generation based large language models," "Knowledge graph prompting for multi-document question answering," "Text summarization with latent queries," "Retrieval-augmented generation for large language models: A survey," and "Knowledge graph-augmented language models for knowledge-grounded dialogue generation." This repository is a crucial resource for researchers to disseminate their findings rapidly and access the latest advancements in their fields. + 00e8e4e881bd0862022f4dfc913b900b,086021a89900a39bcb62036981737bfa,58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035,f0306814bf64f5c9e79603fc6a52f4ea,fc4b27d64f055b7fc30176ba110dd02e + + + PUBLICATION + Preprint refers to the version of the research paper that is under review and available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + CATEGORY + cs.CL is the category under which the research paper is classified on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + DATE + 24 Apr 2024 is the date when the research paper was submitted to arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + IDENTIFIER + 2404.16130v1 is the identifier for the research paper on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Document collections refer to large sets of documents that are analyzed for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + TECHNOLOGY + LLM PROMPTS are specific instructions given to large language models (LLMs) to tailor their responses to the domain of the dataset. These prompts are also used to extract elements from text chunks, ensuring that the LLMs provide relevant and precise information based on the given context. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Community detection is a method used to identify groups of related elements within a graph. It involves the process of identifying communities within a graph, which are clusters of nodes that are more densely connected internally than with the rest of the network. This technique is crucial in understanding the structural dynamics and relationships within complex networks, such as those found in social networks, biological systems, and information retrieval systems. By uncovering these communities, researchers can gain insights into the underlying structure and function of the network, facilitating more effective analysis and interpretation of the data. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Domain-tailored summarization is a method used to create summaries that are specific to the domain of the dataset + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Klein et al. are authors who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Ranade and Joshi are authors who discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Lewis et al. are authors who developed the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Traag et al. are the authors who developed the Leiden algorithm, a method renowned for its efficiency in recovering hierarchical community structures. This algorithm is widely recognized in the field of Natural Language Processing and Information Retrieval for its ability to accurately detect and map out complex community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + QFS is a task framing that focuses on generating summaries based on specific queries, rather than just concatenating excerpts + fb3c48579608fa28be585ceb6cd2f0fe + + + METHOD + A type of summarization that generates natural language summaries based on specific queries, rather than just extracting text + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A neural network architecture that has shown substantial improvements in various summarization tasks + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + GEMINI is a family of highly capable multimodal models, as described in an arXiv preprint. These models are known for their ability to perform in-context learning and summarization, making them a significant advancement in the field of Natural Language Processing and Information Retrieval. + 086021a89900a39bcb62036981737bfa,fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A knowledge graph is a structured representation of information, utilized in the Graph RAG approach for summarization. This structured representation of knowledge is specifically employed in the Graph RAG approach for global summarization, highlighting its role in organizing and integrating information to facilitate comprehensive and coherent summaries. + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + + + REFERENCE + Authors of a paper on Retrieval-augmented generation (RAG) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Author of a paper on query-focused summarization (QFS) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "BROWN ET AL., 2020" refers to a publication by Brown et al. in 2020, which discusses in-context learning with few-shot examples. The authors of this paper are also known for their work on the GPT series of large language models. + bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "KURATOV ET AL., 2024" refers to a publication by Kuratov and colleagues in 2024. The study discusses the recall degradation and potential for information loss in longer context windows of Large Language Models (LLMs). The authors explore the limitations of these extended context windows, providing insights into how the performance of LLMs can be affected when dealing with longer sequences of text. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "LIU ET AL., 2023" refers to a publication by Liu et al. in 2023, which discusses the recall degradation and potential for information loss in longer context windows of large language models (LLMs). The authors explore the limitations of LLM context windows, highlighting how extended contexts can lead to decreased recall accuracy and information retention. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + TECHNOLOGY + COMMUNITY DETECTION ALGORITHMS are algorithms used to partition a graph into communities of nodes with stronger connections to one another. These algorithms are designed to identify modular communities of closely-related nodes within a graph, thereby revealing the underlying structure and relationships within the network. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + ALGORITHM + Louvain is a community detection algorithm used to partition graphs into modular communities + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + HOTPOTQA is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical resource for evaluating entity extraction prompts, particularly with advanced models like GPT-4-turbo. Additionally, HotPotQA is utilized to observe the behavior of text chunk extraction within the Graph RAG (Retrieval-Augmented Generation) approach, making it a versatile tool in the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNOLOGY + GPT-4-Turbo is a version of the GPT-4 model characterized by its large context size of 128k tokens, which is utilized in various analytical tasks. Specifically, GPT-4-Turbo is employed for entity extraction in evaluations, leveraging its extensive context capacity to enhance the accuracy and comprehensiveness of the analysis. This model is particularly suited for tasks within the Natural Language Processing and Information Retrieval domain, where handling large volumes of text and extracting relevant entities are critical. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + + + DATASET + The "PODCAST TRANSCRIPTS" dataset is a comprehensive collection of compiled transcripts from podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders. This dataset is used for analysis and consists of 1669 text chunks, each containing 600 tokens with 100-token overlaps between chunks, amounting to approximately 1 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620,ebf5249c888e07fedce6572a4c03f88c + + + DATASET + The "NEWS ARTICLES" dataset is a comprehensive collection of news articles used for analysis. It serves as a benchmark dataset comprising news articles published from September 2013 to December 2023. The dataset spans a range of categories, including entertainment, business, sports, technology, health, and science. It consists of 3197 text chunks, each containing 600 tokens, with a 100-token overlap between chunks, amounting to approximately 1.7 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620 + + + METHOD + MAP-REDUCE is a method employed for text summarization by applying a map-reduce approach directly to source texts. It is particularly utilized for query-focused summarization of an entire corpus, enabling efficient processing and extraction of relevant information from large datasets. This technique leverages the map-reduce paradigm to distribute the computational workload, making it suitable for handling extensive text collections in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,973164fa90bf2b4ee267f4fd795916bf + + + METRIC + "EMPOWERMENT" is a concept and metric used in the evaluation of various methods, with an average win rate of 51.3%. It measures how well an answer helps the reader understand and make informed judgments about a topic. Specifically, it evaluates the effectiveness of generated answers in empowering users by developing their understanding of broad issues and themes. Empowerment is a target quality in summarization approaches, focusing on the ability to help users reach an informed understanding. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + Naive RAG is a basic retrieval-augmented generation (RAG) method used as a baseline for comparison in text generation tasks. It converts documents to text, splits them into chunks, and embeds these chunks into a vector space for query matching. While it produces the most direct responses, it is outperformed by global approaches in terms of comprehensiveness and diversity. Naive RAG is also noted for listing public figures mentioned in entertainment articles, focusing on their professional achievements and personal lives. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c + + + METHOD + A method for summarizing source texts using a map-reduce approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Questions generated to evaluate the summarization approach, focusing on understanding activities + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + METRIC + TOKEN COSTS refer to the computational cost measured in tokens used in the summarization process. Specifically, in the context of the Graph RAG (Retrieval-Augmented Generation) approach, token costs denote the number of tokens required for processing text. This metric is crucial for evaluating the efficiency and scalability of text processing methods within the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS + The high-level process of the Graph RAG approach and pipeline + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + Design parameters are key settings and configurations in the Graph RAG approach. These parameters are crucial as they influence the design of the Graph RAG approach and pipeline, determining the effectiveness and efficiency of the overall system. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + PARAMETER + + + METHOD + GLOBAL SUMMARIZATION is a method for summarizing information on a global scale. It aims to encapsulate the overall structure and semantics of a dataset, providing a comprehensive overview of information from large datasets or corpora. This technique is particularly useful in the field of Natural Language Processing and Information Retrieval, where it helps in distilling vast amounts of data into coherent and concise summaries, facilitating better understanding and analysis of the underlying information. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e4d9b12cf2b4c691c74019eefff4fb39 + + + ATTRIBUTE + Modularity is an inherent quality of graphs that allows them to be partitioned into communities of closely-related nodes + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Descriptions generated from modular communities in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + + + INPUT + A specific question or request for information that the summarization methods aim to answer + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + A large collection of texts or documents used for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Intermediate answers generated from community summaries before being combined into a final global answer + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + The comprehensive answer generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + + + METHOD + A method that focuses on generating questions to understand activities from datasets + 21e52bc06a82796b1f4bcd73edda1f2a + + + INPUT + Brief descriptions of datasets used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + Datasets that represent real-world information, such as podcast transcripts and news articles + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + METHOD + A method that summarizes the original source texts directly + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + LOW-LEVEL COMMUNITY SUMMARIES are a type of community summary used in the News dataset for analysis. These summaries provide a detailed overview of the source text and are generated from lower hierarchical levels of the community in the knowledge graph. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + OUTPUT + INTERMEDIATE-LEVEL COMMUNITY SUMMARIES are summaries that provide a mid-level overview of the source text. These summaries are generated from intermediate hierarchical levels of the community in the knowledge graph, offering a balanced perspective that captures essential details without overwhelming the reader with excessive information. This approach ensures that the summaries are both informative and concise, making them valuable for understanding the structural dynamics and relationships within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + OUTPUT + Summaries generated from higher hierarchical levels of the community in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + + + PROCESS, SYSTEM + The entity "PIPELINE" refers to a series of processes or steps used to analyze and summarize a dataset. Specifically, in the context of the Graph RAG approach, the pipeline denotes the sequence of steps and processes involved. This structured sequence is essential for systematically handling data, ensuring that each stage of the analysis is methodically executed to achieve accurate and comprehensive results. + 7fb7d9ce2da9c940a32afdd87d1d9e56,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA STRUCTURE, OUTPUT + The "GRAPH INDEX" is a data structure used in Retrieval-Augmented Generation (RAG) systems to organize and retrieve information. It is a self-generated index that enables Graph RAG by utilizing a graph structure to organize and retrieve data. This index is created from a graph structure and is employed for tasks such as query-focused summarization. The graph index includes various elements extracted from text chunks using Large Language Model (LLM) prompts. Additionally, it supports conditions C0-C3 and is created using generic prompts for entity and relationship extraction. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + + + DATA, UNIT + Entity references are mentions of entities within text chunks, extracted during the processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC + Recall is a metric used to measure the completeness of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC + Precision is a metric used to measure the accuracy of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + FEW-SHOT EXAMPLES are specialized instances provided to the Large Language Model (LLM) to improve its performance in domains with specialized knowledge such as science, medicine, and law. These examples are tailored to the domain of the data used in the graph indexing process and serve as sample inputs for in-context learning. By tailoring the extraction prompt to the document corpus domain, few-shot examples enhance the LLM's ability to understand and process domain-specific information effectively. + 2c6ed90897310eea2f28e33fff1c32b0,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA, UNIT + Named entities are specific types of entities such as people, places, and organizations, extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + REFERENCE, PUBLICATION + A reference to a publication by Yang et al. in 2018, introducing the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METHOD, APPROACH + Techniques refer to the specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Implementation details are specific configurations and settings used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS, METHOD + A single extraction round refers to one complete cycle of extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC, ISSUE + Recall degradation refers to the decrease in recall performance when using longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS, METHOD + The extraction process involves identifying and extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Domain refers to the specific area of knowledge or field to which the document corpus belongs + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA, INPUT + Document corpus refers to the collection of documents being processed in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + Default prompt is the standard set of instructions given to the LLM for extracting named entities + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + Secondary extraction prompt is an additional set of instructions given to the LLM for extracting covariates + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METHOD + A covariate prompt is used to extract additional attributes associated with detected entities, including claims linked to the entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + Claims are statements or assertions linked to detected entities, including details such as subject, object, type, description, source text span, and start and end dates + 2c6ed90897310eea2f28e33fff1c32b0 + + + METHOD + Gleanings refer to multiple rounds of entity extraction to ensure that no entities are missed in the process + 2c6ed90897310eea2f28e33fff1c32b0 + + + TECHNIQUE + Logit bias is a technique used to force a yes/no decision from the LLM during the entity extraction process + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + An entity node is a representation of an entity in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A relationship edge is a representation of a relationship between entities in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A claim covariate is an additional attribute or variable associated with a claim in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + Communities of entities are groups of closely-related entities detected and summarized by the LLM + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + The "NOISY GRAPH STRUCTURE" refers to a graph structure that may contain inconsistencies or errors, making it challenging to analyze. This type of graph often includes duplicate or inconsistent entity elements due to variations in text format. These inconsistencies can arise from various sources, such as data entry errors, differing data formats, or incomplete information, which complicate the process of extracting meaningful insights and relationships from the graph. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + DOMAIN + Science is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + DOMAIN + Medicine is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + DOMAIN + Law is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Source text span is an attribute of a claim that indicates the specific portion of text from which the claim was extracted + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Start date is an attribute of a claim that indicates when the event or fact described in the claim began + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + End date is an attribute of a claim that indicates when the event or fact described in the claim ended + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Description is an attribute of a claim that provides a detailed explanation of the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Subject is an attribute of a claim that indicates the main entity involved in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Object is an attribute of a claim that indicates the entity that is affected by the subject in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A concept referring to an entity that has multiple name variations but is resilient to such variations due to sufficient connectivity to closely-related entities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models capable of understanding and generating human-like text. They are utilized for a variety of tasks, including the creation and completion of knowledge graphs, which are essential for structuring and interlinking information in a meaningful way. Additionally, LLMs serve as evaluators of natural language generation, assessing the quality and coherence of text produced by other AI systems. These models play a crucial role in the field of Natural Language Processing and Information Retrieval, contributing significantly to advancements in how machines comprehend and interact with human language. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf + + + TECHNOLOGY + Structured representations of knowledge in the form of triples (subject, predicate, object) used for reasoning tasks + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + Nodes in a graph that are of the same type and are described using rich descriptive text + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + Edges in a graph that represent relationships between entity nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + METRIC + Weights assigned to edges in a graph, representing the normalized counts of detected relationship instances + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The "HIERARCHICAL COMMUNITY STRUCTURE" is a framework in which communities are organized in a hierarchy, with each level providing a partition of the graph nodes. This structure organizes data into a hierarchy of communities, facilitating a multi-level clustering approach. Hierarchical community structure is utilized to generate community summaries, offering a comprehensive method for understanding the relationships and structural dynamics within specialized communities. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39 + + + CONCEPT + A division of graph nodes into mutually-exclusive, collectively-exhaustive communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + TECHNOLOGY + MULTIHOP-RAG is a benchmark dataset comprising news articles published from September 2013 to December 2023, covering a range of categories including entertainment, business, sports, technology, health, and science. It is specifically designed for open-domain question answering, targeting explicit fact retrieval. Additionally, MULTIHOP-RAG represents a specific implementation or variant of Retrieval-Augmented Generation (RAG) used in the context of graph communities. This dataset is also utilized for community detection and analysis, making it a versatile tool in the field of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author who has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + PERSON + Authors who have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The entity "DATASET" refers to a collection of data used for various purposes such as analysis, summarization, and evaluation. This can include diverse types of data like podcast transcripts and news articles. Specifically, the term encompasses datasets used for evaluation purposes, including notable examples like the Podcast and News datasets. + 1d07b4248c2655081c7af0e373bd70c9,7fb7d9ce2da9c940a32afdd87d1d9e56,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + GLOBAL QUERIES refer to questions or inquiries that require comprehensive answers derived from multiple sources or datasets. These queries aim to retrieve information from a global perspective, covering the entire dataset. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + ROOT COMMUNITIES are the top-level clusters in a hierarchical community structure. These communities represent the highest level of organization within the hierarchy, serving as the primary divisions from which more specific sub-communities branch out. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + SUB-COMMUNITIES are lower-level clusters within root communities in a hierarchical community structure, providing more detailed information. These sub-communities play a crucial role in breaking down the larger, more general root communities into more specific and focused groups, thereby facilitating a deeper and more granular understanding of the overall community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + TECHNOLOGY + Detailed documents that provide information about specific subtopics within a community + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The division of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + A system in which elements are ranked or organized in levels + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + LEVEL 0 represents the root-level communities in the hierarchical clustering with maximum modularity. It serves as the foundational layer in a hierarchical community structure, indicating the initial and most significant division of the dataset into distinct groups. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + LEVEL 1 is a sub-level in a hierarchical community structure, providing more detailed information about the internal organization. Specifically, Level 1 represents sub-communities within the root-level communities, thereby revealing the internal structure and dynamics of these larger groups. This level of granularity helps in understanding the intricate relationships and specialized interactions that occur within the broader community framework. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + A visual representation of graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + METHOD + The Leiden algorithm is a method used for detecting communities in large networks + 843fc5421e086120ffa1c75856ecf6cd + + + TOOL + OpenORD is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + + + TOOL + Force Atlas 2 is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Nodes represent entities in a graph, with size proportional to their degree + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Edges represent connections between nodes in a graph + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Covariates are variables that are linked to nodes and edges in a graph + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + The LLM context window is the token limit within which summaries are added for processing by a language model + 843fc5421e086120ffa1c75856ecf6cd + + + METHOD + Hierarchical clustering is a method of clustering data into a tree-like structure with multiple levels + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + The token limit is the maximum number of tokens that can be processed in a single context window by a language model + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Summary detail refers to the level of detail provided in a summary + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Scope refers to the range or extent of information covered in a summary + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + A "USER QUERY" is a question or inquiry posed by a user seeking information, which the system aims to answer. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd + CONCEPT + + + ELEMENT + Chunks are segments of community summaries divided into pre-specified token sizes + 843fc5421e086120ffa1c75856ecf6cd + ELEMENT + + + METRIC + Prominence is a metric used to prioritize community edges based on the combined degree of source and target nodes + 843fc5421e086120ffa1c75856ecf6cd + + + METRIC + Combined source and target node degree is a metric used to measure the overall prominence of community edges + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Community edges are connections between nodes within a community, prioritized based on prominence + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Sub-community summaries are shorter summaries of sub-communities used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + + + CATEGORY + Community level refers to the different levels in the hierarchical community structure used to generate summaries + 843fc5421e086120ffa1c75856ecf6cd + + + DATA + Chunks are segments of community summaries divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + + + METRIC + A score between 0-100 generated by the LLM to indicate how helpful an answer is in addressing the target question + 1d07b4248c2655081c7af0e373bd70c9 + + + USER + A user looking for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + + + USER + A user incorporating current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic addressing the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + The importance of health literacy highlighted through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + + + OUTPUT + Answers generated for each chunk of community summaries + 1d07b4248c2655081c7af0e373bd70c9 + + + METRIC + The pre-specified size of tokens used to divide community summaries into chunks + 1d07b4248c2655081c7af0e373bd70c9 + + + TECHNOLOGY + The "CONTEXT WINDOW" refers to a window of text used to generate answers, constrained by token size. The size of the context window is consistent across all conditions, ensuring uniformity in answer generation processes. + 1d07b4248c2655081c7af0e373bd70c9,973164fa90bf2b4ee267f4fd795916bf + + + PERSON + Kevin Scott is the Chief Technology Officer (CTO) of Microsoft and actively participates in podcast conversations. His involvement in these discussions is documented and compiled in the dataset, highlighting his contributions to the field of technology and his role in shaping Microsoft's strategic direction. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + Individuals who are leaders in the technology industry and participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + + + INPUT + A specific activity or goal that the user aims to achieve using the datasets + 1d07b4248c2655081c7af0e373bd70c9 + + + INPUT + QUESTIONS refer to specific inquiries generated by the Large Language Model (LLM) based on the user's task and the target datasets. These questions are utilized in the analysis to evaluate the performance of different methods within the domain of Natural Language Processing and Information Retrieval. The generation and subsequent use of these questions are crucial for assessing the effectiveness and accuracy of various computational techniques and models. + 1d07b4248c2655081c7af0e373bd70c9,4c855404ee3d3c94aa2136f1513c666f + + + + + 1d07b4248c2655081c7af0e373bd70c9 + + + DATASET + MT-BENCH is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical tool in evaluating the performance of models in retrieving accurate and relevant information from a broad range of topics. MT-Bench is prominently featured in the academic paper titled "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," where it is utilized to assess the capabilities of large language models in a structured and rigorous manner. This dataset plays a significant role in advancing the field of Natural Language Processing and Information Retrieval by providing a standardized metric for model comparison and evaluation. + 922778ce1cb2fdd6dbab1746c8795620,b1bbda43309e8e0e2175ea034aa88e13 + DATASET + + + PROCESS + The process through which people inspect, engage with, and contextualize data within the broader scope of real-world activities + 922778ce1cb2fdd6dbab1746c8795620 + PROCESS + + + TECHNOLOGY + Retrieval-Augmented Generation systems used for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + TECHNOLOGY + + + AUTHORS + Authors of a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors of a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + AUTHORS + Authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + + + 922778ce1cb2fdd6dbab1746c8795620 + + + PODCAST + "BEHIND THE TECH" is a podcast series featuring conversations between Kevin Scott and other technology leaders. It serves as a media platform associated with Kevin Scott, providing insights and discussions on various technological advancements and industry trends. + 833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + Kevin Scott, Microsoft CTO, who participates in the podcast conversations compiled in the dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + DATASET + A benchmark dataset for open-domain question answering, targeting explicit fact retrieval + 922778ce1cb2fdd6dbab1746c8795620 + + + METRIC + N represents the number of test questions per dataset used in the evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + METHOD + A method applying a map-reduce approach directly to source texts for summarization + 973164fa90bf2b4ee267f4fd795916bf + + + METHOD + A na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached + 973164fa90bf2b4ee267f4fd795916bf + + + CATEGORY + C0 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a root-level community summary, which is utilized to answer user queries by providing the fewest number of summaries. This category is essential for understanding the structural dynamics within the community, particularly in the domain of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C1 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a high-level community summary used to answer user queries, effectively representing sub-communities of C0. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C2 is a category or condition used in the analysis, representing a specific subset of the data. It functions as an intermediate-level community summary used to answer user queries, representing sub-communities of C1. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C3 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a category or condition that encapsulates low-level community summaries, which are instrumental in answering user queries. These summaries represent sub-communities of C2, providing detailed insights into the structural dynamics and relationships within the broader community. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + TS, or "Text Summarization," is a category or condition used in the analysis, representing a specific subset of the data. It is particularly focused on source text summarization within the analysis. TS employs a text summarization method that applies a map-reduce approach directly to source texts, facilitating efficient and scalable summarization processes. This category is integral to understanding and processing large volumes of text data, making it a crucial component in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + "SS" is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a baseline condition and is associated with a na¨ıve RAG (Retrieval-Augmented Generation) approach. In this context, text chunks are retrieved and added to the context window until the token limit is reached. + 4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CONCEPT + The prompts used for answer generation, which are the same across all conditions with minor modifications + 973164fa90bf2b4ee267f4fd795916bf + + + DATASET + The "PODCAST DATASET" is a collection of podcast transcripts utilized for both analysis and evaluation purposes. This dataset is specifically designed to support various analytical tasks, providing a rich source of textual data for researchers and practitioners in the field of Natural Language Processing and Information Retrieval. The transcripts within the dataset offer valuable insights and serve as a critical resource for evaluating different computational models and techniques. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + DATASET + The "NEWS DATASET" is a collection of news articles utilized for both analysis and evaluation purposes. This dataset serves as a valuable resource for examining and assessing various aspects of news content, making it an essential tool in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + METRICS in the context of Natural Language Processing and Information Retrieval are essential tools used to evaluate the performance of natural language generation. These metrics include both reference-based metrics, which compare generated texts to a set of reference texts, and qualities of the generated texts themselves. They are crucial in the analysis to assess the effectiveness of different methods in generating natural language, ensuring that the outputs are both accurate and of high quality. + 4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + "WANG ET AL., 2023A" refers to a study conducted by Wang and colleagues in 2023, which highlights the effectiveness of Large Language Models (LLMs) in evaluation. This study is a significant contribution to the field, providing insights into the capabilities and performance of LLMs in various evaluative tasks. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + "ZHENG ET AL., 2024" refers to a study conducted by Zheng and colleagues in 2024. This study highlights the effectiveness of Large Language Models (LLMs) in evaluation processes. The research, authored by Zheng et al., provides significant insights into the capabilities and applications of LLMs within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + The entity "CONDITIONS" refers to the different scenarios or variables that are compared in an experiment. Specifically, in the context of the analysis, these conditions include Graph RAG, text summarization, and semantic search RAG. These conditions are used to evaluate and compare various aspects of performance and effectiveness within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + USER QUERIES refer to the inquiries made by users to retrieve information. These queries are answered using different methods and conditions, depending on the context and the specific requirements of the information retrieval process. + 973164fa90bf2b4ee267f4fd795916bf,e4d9b12cf2b4c691c74019eefff4fb39 + + + CONCEPT + Types of entities extracted during the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + + + METRIC + The "CONTEXT WINDOW SIZE" refers to the fixed size of the context window used in various stages of natural language processing and information retrieval tasks. For the final evaluation, the context window size is set to 8k tokens. During the analysis phase, different context window sizes are tested, including 8k, 16k, 32k, and 64k tokens. Additionally, in the graph indexing process, the context window size is set to 600 tokens. This variability in context window sizes highlights the importance of adapting the window size to the specific requirements of different tasks within the domain. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + The process of extracting information, with 1 gleaning for the Podcast dataset and 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + TECHNOLOGY + Natural Language Generation (NLG) is a subfield of artificial intelligence that focuses on generating human-like text from data + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method where a Large Language Model (LLM) is used to compare and evaluate competing outputs + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method for evaluating the performance of Retrieval-Augmented Generation (RAG) systems, focusing on context relevance, faithfulness, and answer relevance + 322e02986c8724eedbcf3ebfa20b989c + + + PUBLICATION + A reference to a study or paper authored by Es and others in 2023 + 322e02986c8724eedbcf3ebfa20b989c + + + TOOL + A Large Language Model (LLM) used to evaluate and compare generated texts based on specific metrics + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + DIRECTNESS is a metric that measures how specifically and clearly an answer addresses a question. It is used to evaluate the straightforwardness of the generated answers. Additionally, it serves as a validity test metric to measure the directness of responses, with naive RAG (Retrieval-Augmented Generation) producing the most direct responses. + 322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + DATA + An example of LLM-generated assessment shown in a table format + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The entity "QUESTION" refers to a specific query used in the evaluation process, particularly as a metric to evaluate the generated responses by asking specific questions. This approach is commonly employed in the domain of Natural Language Processing and Information Retrieval to assess the quality and relevance of responses generated by various models or systems. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + ENTITY + Public figures are individuals who have gained fame or notoriety in various sectors such as entertainment, sports, and digital media. These individuals are well-known in the entertainment industry and are frequently mentioned across various articles. Their prominence in public discourse spans multiple domains, reflecting their influence and recognition in society. + 322e02986c8724eedbcf3ebfa20b989c,718017a4871c909420f84b85b8ba969d + + + DATASET + ENTERTAINMENT ARTICLES is a collection of articles focused on the entertainment industry. This dataset consists of articles related to various aspects of the entertainment sector, providing a comprehensive resource for understanding trends, developments, and key topics within this field. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + DOMAIN + The **ENTERTAINMENT INDUSTRY** is a multifaceted sector that encompasses various forms of entertainment, including movies, music, television, sports, and digital media. This industry is characterized by its diverse range of content and mediums, which collectively contribute to its broad appeal and significant cultural impact. The entertainment industry plays a crucial role in shaping public opinion, trends, and cultural norms through its extensive reach and influence across different platforms and genres. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric indicating the highest level of development or achievement in a particular field + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric indicating results that are comparable to or better than those of others in the same field + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric based on evaluations made by humans + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + Metrics that require a gold standard or reference answers for evaluation + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + An evaluation method that does not require reference answers + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how relevant the generated text is to the given context + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how accurately the generated text reflects the source information + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how relevant the generated answer is to the question + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method involving multiple stages or steps + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The correct or ideal answers used as a benchmark in evaluations + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + "SENSEMAKING QUESTIONS" are a class of questions used to evaluate the performance of Retrieval-Augmented Generation (RAG) systems. These questions are specifically designed to help users understand and make sense of complex information, as well as to validate the understanding and interpretation of data. By employing sensemaking questions, researchers and practitioners can assess how effectively a RAG system can retrieve and generate relevant information, thereby ensuring that the system aids in the comprehension and accurate interpretation of intricate datasets. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method where two items are directly compared against each other + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + TARGET METRICS are specific measures used to evaluate the performance of RAG systems. These metrics are aimed to be achieved or measured in the analysis and are the focus of an evaluation. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A metric used as a baseline or standard for comparison + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures the accuracy and reliability of a method or result + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures the randomness or variability in a process + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The average scores obtained from multiple evaluations + 322e02986c8724eedbcf3ebfa20b989c + + + PERSON + Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Britney Spears is a public figure frequently mentioned in entertainment articles, known for her significant contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his significant contributions to the music industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in film and television + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in music + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in sports + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in digital media and business + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry who are involved in controversies + e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric used to determine the winner in the comparison of generated responses + e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric used to evaluate the quality of LLM-generated responses + e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "FILM" refers to a sector within the entertainment industry that encompasses movies and cinema. This sector includes public figures involved in the movie industry, such as actors, directors, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "TELEVISION" refers to a sector within the entertainment industry that encompasses TV shows and series. This sector includes public figures involved in TV shows, such as actors, hosts, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + MUSIC is a sector within the entertainment industry that encompasses musical performances and recordings. This sector includes public figures involved in the music industry, such as singers, musicians, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "SPORTS" refers to a sector within the entertainment industry that encompasses athletic events and competitions. This sector includes public figures involved in sports, such as athletes, coaches, and sports commentators. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + DIGITAL MEDIA is a sector within the entertainment industry that encompasses online content and social media. This sector includes public figures involved in online platforms, such as influencers, content creators, and digital marketers. These individuals play a significant role in shaping digital landscapes through their engagement with audiences and their ability to leverage various online tools and platforms for content dissemination and marketing purposes. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes stories and themes that shape culture + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes popular movements and styles + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes public conversations and debates + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes formal discussions and communications + e8c8f911135faf3ff35f24107eb3f99c + + + RESPONSE + Answer 1 provides a varied and rich response by covering a wide range of public figures from different sectors of the entertainment industry, including film, television, music, sports, gaming, and digital media. It offers insights into the contributions and influence of these figures, as well as controversies and their impact on public discourse. The answer also cites specific data sources for each mentioned figure, indicating a diverse range of evidence to support the claims. + 718017a4871c909420f84b85b8ba969d + + + RESPONSE + "ANSWER 2" is a generated answer for the example question in the News article dataset. It focuses on a smaller group of public figures, primarily from the music industry and sports, and relies heavily on a single source for data. "ANSWER 2" provides concise explanations for the frequent mentions of specific public figures such as Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + Naïve RAG is a baseline method for retrieval-augmented generation (RAG) that does not use advanced techniques. It is a basic form of RAG with certain drawbacks that advanced RAG systems aim to overcome. Naïve RAG is used to generate answers for questions in the News article dataset and to generate responses that directly list specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d,e4d9b12cf2b4c691c74019eefff4fb39,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19 + + + DATASET + The "NEWS ARTICLE DATASET" is a collection of news articles utilized for various analytical purposes. This dataset is specifically employed for generating responses to questions about public figures in the entertainment industry, making it a valuable resource for both analysis and information retrieval tasks within this domain. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + TOPIC + Controversies are events or issues involving public figures that generate public debate and impact public discourse. + 718017a4871c909420f84b85b8ba969d + + + SECTOR + The gaming sector includes public figures involved in the gaming industry, including gamers, developers, and streamers. + 718017a4871c909420f84b85b8ba969d + + + RESOURCE + Data sources are references or reports used to support claims about public figures and their influence. + 718017a4871c909420f84b85b8ba969d + + + METHOD + Assessments generated by large language models (LLMs) to evaluate the answers produced by different methods + ebf5249c888e07fedce6572a4c03f88c + + + DATASET + An example question used in the News article dataset for analysis + ebf5249c888e07fedce6572a4c03f88c + + + DATA + The datasets used in the analysis, consisting of various text sources + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + A metric used to compare the performance of different conditions in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + CATEGORY + A specific setup or scenario used in the analysis, such as C0, C1, C2, C3, TS, and SS + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + WIN RATE is a measure used to evaluate the success rate of different methods in providing comprehensive and diverse answers. It represents the percentage of times a particular approach or method achieves a win in a given context. Specifically, it quantifies the percentage of times a condition outperformed another in the analysis. This metric is crucial in assessing the effectiveness of various strategies within the domain of Natural Language Processing and Information Retrieval, offering insights into the comparative performance of different techniques. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4 + + + METRIC + The condition that performed the best across all comparisons in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + The expected win rate of a condition when compared to itself, shown as 50% for reference + 4c855404ee3d3c94aa2136f1513c666f + + + METHOD + The use of large language models (LLMs) at the time of querying, evaluated in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + METHOD + The "FINAL EVALUATION" is the last stage of the analysis where the best performing context window size was used. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + + + PROCESS + The process that resulted in the creation of graphs for the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + + + STRUCTURE + A data structure consisting of nodes and edges, used to represent the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + + + METHOD + Approaches that consistently outperformed the naive RAG in comprehensiveness and diversity metrics + 36db32c37e1987e2c5863898ad882190 + + + METRIC + The number of context units, such as community summaries or text chunks, used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + + + METRIC + The term "TOKENS" refers to the number of individual words used in the analysis. The evaluation typically focuses on corpora in the region of 1 million tokens. This metric is crucial for understanding the scope and scale of the text data being analyzed, particularly in the fields of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,92e93fc6449756c0a60200636b297f65 + METRIC + + + METRIC + The percentage of the maximum token count used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + + + METHOD + MAP-REDUCE SUMMARIZATION is a method for summarizing source texts using a map-reduce approach. This summarization technique is notably resource-intensive, necessitating the highest number of context tokens compared to other methods. The map-reduce framework, originally popularized for its efficiency in processing large-scale data, is adapted here to handle the complexities of text summarization, ensuring comprehensive and accurate extraction of key information from extensive source texts. + 36db32c37e1987e2c5863898ad882190,e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Summaries at the root level of the community hierarchy, requiring dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + + + DATASET + SOURCE TEXTS are the original texts from which summaries or analyses are derived. These texts serve as the foundational material used for comparison with community summaries in the analysis. + 6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39 + + + REFERENCE + A reference to a paper by Ram et al. in 2023 discussing RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + REFERENCE + "GAO ET AL., 2023" is a paper published in 2023 by Gao et al. that delves into advanced Retrieval-Augmented Generation (RAG) techniques, specifically where the index is a knowledge graph. The publication also touches upon naive RAG approaches, providing a comprehensive examination of both advanced and basic methodologies within the domain of Natural Language Processing and Information Retrieval. + 6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + + + CATEGORY + Intermediate-level summaries are a type of community summary used in the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + CATEGORY + Root-level summaries are a type of community summary used in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + METRIC + Answer comprehensiveness is a measure used to evaluate the completeness of answers provided by different methods + 6f33a085ff3304e5994f7fbb86c881a4 + + + TECHNOLOGY + Ad-hoc LLM use refers to the spontaneous use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + + + TECHNOLOGY + Element extraction prompts are used to extract specific details in the Graph RAG index + 6f33a085ff3304e5994f7fbb86c881a4 + + + CONCEPT, TECHNOLOGY + A mathematical space in which text chunks and queries are embedded to represent similar semantics + f35de4d9fb65f1d5a392064b20545c19 + + + CONCEPT, DATA + Search inputs that are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A type of RAG system that includes patterns for iterative and dynamic cycles of interleaved retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, CONCEPT + A concept related to generation-augmented retrieval that facilitates future generation cycles + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method that facilitates future generation cycles by using self-memory + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A strategy for iterative retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A federated strategy for retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method that combines multiple concepts for summarizing multiple documents + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method for answering questions that require multiple steps or "hops" to gather information + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + An approach that involves generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A method for generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A method for answering multiple interpretations of ambiguous questions by generating a hierarchical structure + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A process that involves using LLMs to create knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A process that involves using LLMs to complete existing knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + Graphs that represent causal relationships, which can be extracted using LLMs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + REFERENCE, PUBLICATION + A reference to a publication by Cheng et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Mao et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Shao et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Wang et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Su et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Feng et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Trivedi et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Khattab et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Sarthi et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Kim et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + "TRAJANOSKA ET AL., 2023" refers to a paper by Trajanoska et al. published in 2023, which focuses on using Large Language Models (LLMs) for knowledge graph creation. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting innovative methodologies for leveraging advanced language models to construct and enhance knowledge graphs. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + "Yao et al., 2023" refers to a paper published by Yao and colleagues in 2023. The study focuses on the application of large language models (LLMs) for the task of knowledge graph completion. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting the potential of advanced LLMs to enhance the accuracy and efficiency of knowledge graph completion processes. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + TECHNOLOGY, METHOD + A system that combines multiple concepts for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + Strategies used before the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Strategies used during the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Strategies used after the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A pattern in Modular RAG systems for iterative and dynamic cycles of retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Cycles of generation that are facilitated by self-memory in Graph RAG + f35de4d9fb65f1d5a392064b20545c19 + + + PUBLICATION + A paper by Ban et al. published in 2023, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + PUBLICATION + A paper by Zhang et al. published in 2024, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where the index is a knowledge graph, developed by Baek et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Baek et al. published in 2023, focusing on the KAPING method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where subsets of the graph structure are the objects of enquiry, developed by He et al. in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by He et al. published in 2024, focusing on the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where derived graph metrics are the objects of enquiry, developed by Zhang in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Zhang published in 2023, focusing on the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, developed by Kang et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Kang et al. published in 2023, focusing on the SURGE method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where retrieved event-plot subgraphs are serialized using narrative templates, developed by Ranade and Joshi in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Ranade and Joshi published in 2023, focusing on the FABULA method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + PUBLICATION + A paper by Wang et al. published in 2023, focusing on a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + ORGANIZATION + LangChain is an organization that developed Langchain graphs and supports a variety of graph databases. + 71f6daf11e64e5273a3847d46bf228e1,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + + + ORGANIZATION + LlamaIndex is an organization that developed the LlamaIndex Knowledge Graph Index and supports a variety of graph databases. + 6cd82819982879bd164547d2773ba5c7,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + + + TECHNOLOGY + Neo4J is both a graph database format supported by various Retrieval-Augmented Generation (RAG) applications and an organization that developed Project NaLLM. The graph database format of Neo4J is widely recognized for its efficiency in handling complex relationships and structures, making it a valuable tool in the field of Natural Language Processing and Information Retrieval. As an organization, Neo4J has contributed significantly to the advancement of these domains through innovative projects like NaLLM, which further underscores its pivotal role in the community. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + + + METHOD + A method that can create and reason over knowledge graphs in Neo4J format, developed by Neo4J in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + TECHNOLOGY + NebulaGraph is both a graph database format and an organization that has made significant contributions to the field of graph databases and retrieval-augmented generation (RAG) applications. As a graph database format, NebulaGraph is supported by various RAG applications, facilitating the efficient handling and querying of complex graph data structures. Additionally, NebulaGraph, as an organization, has pioneered the industry-first graph RAG, which integrates retrieval-augmented generation with large language models (LLMs) based on knowledge graphs. This innovation underscores NebulaGraph's role in advancing the capabilities of knowledge graph-based applications and enhancing the performance of LLMs in generating contextually relevant information. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + + + METHOD + A method that can create and reason over knowledge graphs in NebulaGraph format, developed by NebulaGraph in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + METHOD + A method for comparing fabrication rates, developed by Manakul et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + "MANAKUL ET AL., 2023" refers to a paper by Manakul et al. published in 2023, which focuses on the SelfCheckGPT method. This work by Manakul and colleagues is centered around the development and application of SelfCheckGPT, a technique likely aimed at enhancing the performance and reliability of GPT models. The paper contributes to the field of Natural Language Processing and Information Retrieval by addressing specific challenges and proposing innovative solutions through the SelfCheckGPT method. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + PUBLICATION + + + STAKEHOLDER + END USERS are individuals who are the final users of the system or analysis. They play a crucial role in validating sensemaking questions and target metrics, ensuring that the system or analysis meets the intended objectives and provides meaningful insights. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + STAKEHOLDER + + + CONCEPT + Considerations and compromises involved in building a graph index + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METRIC + The effectiveness of RAG systems, which varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + METRIC + + + CONCEPT + Various forms of data used in RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METRIC + The scale of datasets used in RAG systems, which affects performance + 92e93fc6449756c0a60200636b297f65 + METRIC + + + PROCESS + The process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + PROCESS + + + DATASET + Collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + DATASET + + + CONCEPT + Different categories of questions used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METHOD + SelfCheckGPT is an approach used to compare fabrication rates in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method for global summarization of source texts that does not use a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + RESOURCE + The amount of computational resources allocated for a task + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The expected number of queries over the lifetime of a dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Annotations that provide detailed information about the text + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method that uses embeddings to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + RAG schemes that combine embedding-based matching with other approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Mechanisms used in map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A hierarchical organization of communities + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) for global text summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The cost associated with the number of tokens used in a text generation task + e4d9b12cf2b4c691c74019eefff4fb39 + + + TECHNOLOGY + An implementation of Graph RAG approaches using the Python programming language + e4d9b12cf2b4c691c74019eefff4fb39 + + + PERSON + A person who contributed to the work mentioned in the acknowledgements + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The rates at which fabrications occur in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The expected number of queries over the lifetime of a specific dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The benefits or value obtained from using a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Different methods related to retrieval-augmented generation that utilize graph structures + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Graph RAG approaches that operate in a more localized manner + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Annotations made on the graph to provide additional information + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Reports generated from community summaries + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + An operation that aggregates information across multiple levels of a hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A mechanism that allows for exploring detailed information by following higher-level summaries + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + The trail of information that guides users to more detailed data + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + The top-level communities in a hierarchical structure + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A graph index organized around entities + e4d9b12cf2b4c691c74019eefff4fb39 + + + TECHNOLOGY + A publicly available implementation of a technology + e4d9b12cf2b4c691c74019eefff4fb39 + + + PERSON + Alonso Guevara Fernández is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Amber Hoak is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Andrés Morales Esquivel is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Ben Cutler is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Billie Rinaldi is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Chris Sanchez is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Chris Trevino is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Christine Caggiano is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + David Tittsworth is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Dayenne de Souza is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Douglas Orbaker is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Ed Clark is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Gabriel Nieves-Ponce is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Gaudy Blanco Meneses is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Kate Lytvynets is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Katy Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Mónica Carvajal is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Nathan Evans is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Richard Ortega is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Rodrigo Racanicci is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Sarah Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Shane Solomon is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + A technical report on GPT-4 published as an arXiv preprint + 086021a89900a39bcb62036981737bfa + + + METHOD + A method for zero-shot knowledge graph question answering described in an arXiv preprint + 086021a89900a39bcb62036981737bfa + + + METHOD + A method for harnessing large language models for advanced causal discovery from data + 086021a89900a39bcb62036981737bfa + + + METHOD + A method incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Achiam is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Adler is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Agarwal is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + L. Ahmad is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + I. Akkaya is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + F. L. Aleman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + D. Almeida is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Altenschmidt is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Altman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Anadkat is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + R. Anil is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Borgeaud is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + Y. Wu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J.-B. Alayrac is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Yu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + R. Soricut is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Schalkwyk is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + A. M. Dai is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + A. Hauth is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Baek is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + A. F. Aji is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + A. Saffari is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + T. Ban is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + L. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + X. Wang is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + H. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + T. Baumel is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + M. Eyal is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + M. Elhadad is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + Baumel, T. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Eyal, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Elhadad, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Blondel, V. D. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Guillaume, J.-L. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Lambiotte, R. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Lefebvre, E. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The journal where the paper "Fast unfolding of communities in large networks" was published + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Brown, T. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Mann, B. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Ryder, N. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Subbiah, M. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Kaplan, J. D. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dhariwal, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Neelakantan, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Shyam, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Sastry, G. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Askell, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + "ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS" is a prominent conference where significant papers in the field of Natural Language Processing and Information Retrieval are presented. Notable papers presented at this conference include "Language models are few-shot learners" and "Retrieval-augmented generation for knowledge-intensive NLP tasks." Additionally, it is also the journal where the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" was published. This conference and journal serve as key platforms for disseminating cutting-edge research in neural information processing systems. + 58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,b1bbda43309e8e0e2175ea034aa88e13 + + + PERSON + Cheng, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Luo, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Chen, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Liu, L. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Zhao, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory"Zhao, D. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + PERSON + + + PERSON + Yan, R. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dang, H. T. is an author of the paper "Duc 2005: Evaluation of question-focused summarization systems" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The conference where the paper "Duc 2005: Evaluation of question-focused summarization systems" was presented + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Es, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + James, J. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Espinosa-Anke, L. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Schockaert, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Feng, Z. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Feng, X. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Yang, M. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Qin, B. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Fortunato, S. is an author of the paper "Community detection in graphs" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The journal where the paper "Community detection in graphs" was published + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Gao, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Xiong, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models. The paper provides a comprehensive survey of the methodologies and applications of retrieval-augmented generation, highlighting its significance in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Gao, X. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Jia, K. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant area of research within the domains of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Pan, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Bi, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dai, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance and capabilities of large language models, a significant area of research within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Sun, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Wang, H. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Goodwin, T. R. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Savery, M. E. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Demner-Fushman, D. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + CONFERENCE + COLING (International Conference on Computational Linguistics) is the conference where the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" was presented + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + He, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Tian, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Sun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Chawla, N. V. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Laurent, T. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + LeCun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Bresson, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Hooi, B. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jacomy, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Venturini, T. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Heymann, S. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Bastian, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PUBLICATION + PLOS ONE is the journal where the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" was published + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jin, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Yu, Z. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jiao, P. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Pan, S. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + He, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Wu, J. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Philip, S. Y. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Zhang, W. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PUBLICATION + IEEE Transactions on Knowledge and Data Engineering is the journal where the paper "A survey of community detection approaches: From statistical modeling to deep learning" was published + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Kang, M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Kwak, J. M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Baek, J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Hwang, S. J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Khattab, O. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Santhanam, K. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Li, X. L. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hall, D. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text, indicating its relevance within the domain of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Liang, P. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Liang, P. contributed to the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP," which explores the integration of retrieval and language models to enhance knowledge-intensive tasks in NLP. Additionally, Liang, P. authored the paper "Lost in the middle: How language models use long contexts," which investigates the utilization of extended contexts by language models. These contributions highlight Liang, P.'s significant role in advancing the understanding and application of language models in complex NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Potts, C. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Zaharia, M. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kim, G. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kim, S. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Jeon, B. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Park, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kang, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Klein, G. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Moon, B. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hoffman, R. R. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The journal where the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" were published + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Koesten, L. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Gregory, K. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Groth, P. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Simperl, E. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The journal where the paper "Talking datasets–understanding data sensemaking behaviours" was published + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kuratov, Y. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Bulatov, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Anokhin, P. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Sorokin, D. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Sorokin, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Burtsev, M. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + TECHNOLOGY + Langchain graphs is a technology developed by LangChain + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Laskar, M. T. R. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" and also contributed to the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models." Both works highlight Laskar's expertise in leveraging transformer models and transfer learning techniques to enhance the performance of query-focused abstractive text summarization, demonstrating a significant contribution to the field of Natural Language Processing and Information Retrieval. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hoque, E. is an author of two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning to improve the effectiveness of transformer models in query-focused abstractive summarization tasks. Both works contribute to advancing the understanding and application of transformer models in specialized summarization contexts. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Huang, J. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The conference where the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" was presented + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + arXiv preprint refers to a preprint of a paper that is available on the arXiv repository + 71f6daf11e64e5273a3847d46bf228e1 + + + EVENT + The 33rd Canadian Conference on Artificial Intelligence, held in Ottawa, ON, Canada, from May 13–15, 2020 + 6cd82819982879bd164547d2773ba5c7 + + + EVENT + The 2020 edition of the Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + PUBLISHER + Springer is the publisher of the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Huang, J. X. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + PUBLICATION + The journal where the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" was published + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lewis, P. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Perez, E. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Piktus, A. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Petroni, F. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks"Petroni, F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + PERSON + + + PERSON + Karpukhin, V. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Goyal, N. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Küttler, H. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lewis, M. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Yih, W.-T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Rocktäschel, T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, N. F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lin, K. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Hewitt, J. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Paranjape, A. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Bevilacqua, M. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, Y. is an author of the paper "Hierarchical transformers for multi-document summarization" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lapata, M. is an author known for significant contributions to the field of Natural Language Processing and Information Retrieval. Notably, Lapata, M. has authored the paper "Hierarchical transformers for multi-document summarization," which explores advanced techniques in summarizing information from multiple documents using hierarchical transformer models. Additionally, Lapata, M. has contributed to the paper "Text summarization with latent queries," which delves into innovative methods for summarizing text by leveraging latent query representations. These works highlight Lapata, M.'s expertise and active research in the domain of text summarization, showcasing a focus on developing sophisticated models and methodologies to enhance the efficiency and accuracy of summarization tasks. + 6cd82819982879bd164547d2773ba5c7,fc4b27d64f055b7fc30176ba110dd02e + + + TECHNOLOGY + LlamaIndex Knowledge Graph Index is a technology developed by LlamaIndex + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Manakul, P. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liusie, A. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Gales, M. J. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Mao, Y. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + He, P. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, X. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Shen, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Shen, Y.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Gao, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Han, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Chen, W. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Chen, W.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Martin, S. is an author of the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing a comprehensive, open-source solution for the layout of large graphs, which is a critical task in the visualization and analysis of complex networks. The toolbox aims to facilitate the understanding and interpretation of large-scale graph data, making it a valuable resource for researchers and practitioners in fields such as computational linguistics, information retrieval, and data science. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Brown, W. M. is an author of the paper "Openord: An open-source toolbox for large graph layout." + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + KLAVANS, R. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Boyack, K. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on the development and application of Openord, a comprehensive open-source toolbox designed for the layout of large graphs. The paper likely discusses the methodologies, algorithms, and practical implementations of the toolbox, contributing to the fields of graph theory and data visualization. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + EVENT + The conference where the paper "Openord: An open-source toolbox for large graph layout" was presented + 833e7d67dcd30790b26b71c9b5306f6b + EVENT + + + TECHNOLOGY + GPT-4 is a large language model used in Microsoft's study on scientific discovery + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + + + TECHNOLOGY + Project NaLLM is a project developed by Neo4J + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + + + PERSON + Newman, M. E. is the author of the paper "Modularity and community structure in networks" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PUBLICATION + The journal where the paper "Modularity and community structure in networks" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + + + PERSON + Ram, O. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Levine, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Dalmedigos, I. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Muhlgay, D. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shashua, A. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Leyton-Brown, K. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shoham, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PUBLICATION + The journal where the paper "In-context retrieval-augmented language models" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + + + PERSON + Ranade, P. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Joshi, A. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Sarthi, P. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Abdullah, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Tuli, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Khanna, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Goldie, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Manning, C. D. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" and the paper "Raptor: Recursive abstractive processing for tree-organized retrieval". These contributions highlight Manning's involvement in advancing the fields of Natural Language Processing and Information Retrieval, particularly in the areas of multi-hop question answering and recursive abstractive processing. + 833e7d67dcd30790b26b71c9b5306f6b,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Scott, K. is associated with "Behind the Tech" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shao, Z. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Gong, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Huang, M. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + + + PERSON + Duan, N. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + + + PERSON + Su, D. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Xu, Y. is an author of multiple academic papers in the field of Natural Language Processing and Information Retrieval. Notably, Xu, Y. contributed to the paper titled "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," which addresses the management of scholarly information related to COVID-19 through advanced question answering and summarization techniques. Additionally, Xu, Y. co-authored the paper "Text summarization with latent queries," which explores innovative methods for text summarization by leveraging latent queries. These contributions highlight Xu, Y.'s expertise and active involvement in developing sophisticated systems for information retrieval and summarization. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yu, T. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Siddique, F. B. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Barezi, E. J. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Fung, P. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Tang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Yang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Touvron, H. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Martin, L. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Stone, K. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Albert, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Almahairi, A. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Babaei, Y. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bashlykov, N. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Batra, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bhargava, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bhosale, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Traag, V. A. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Waltman, L. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Van Eck, N. J. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PUBLICATION + Scientific Reports is the journal where the paper "From Louvain to Leiden: guaranteeing well-connected communities" was published + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trajanoska, M. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Stojanov, R. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trajanov, D. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trivedi, H. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Balasubramanian, N. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Khot, T. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Sabharwal, A. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Wang, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Liang, Y. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Meng, F. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Sun, Z. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Shi, H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Li, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through their work on evaluating language models. Specifically, Li, Z. has co-authored the paper titled "Is ChatGPT a Good NLG Evaluator? A Preliminary Study," which explores the effectiveness of ChatGPT as a natural language generation evaluator. Additionally, Li, Z. has co-authored another paper, "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which examines the performance of large language models in evaluative roles using specific benchmarking tools. These contributions highlight Li, Z.'s active involvement in advancing the understanding and assessment of language models within the academic community. + 8d87efac8c50cf20cdf26bf61e5e2035,b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Xu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Qu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhou, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wang, S. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" and also contributed to the paper "Is chatgpt a good nlg evaluator? a preliminary study." These works indicate Wang, S.'s involvement in cutting-edge research within the fields of federated search, retrieval augmented generation, and natural language generation evaluation, showcasing a focus on both the technical and evaluative aspects of Natural Language Processing and Information Retrieval. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Khramtsova is an author mentioned in the text + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Khramtsova, E. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhuang, S. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through multiple academic papers. Notably, Zhuang, S. co-authored the paper titled "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," which explores the evaluation of federated search systems within the framework of retrieval-augmented generation. Additionally, Zhuang, S. co-authored another significant paper, "Judging llm-as-a-judge with mt-bench and chatbot arena," which delves into the assessment of large language models (LLMs) using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Zhuang, S.'s active involvement in advancing research in federated search and the evaluation of LLMs. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zuccon, G. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wang, Y. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Lipka, N. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Rossi, R. A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Siu, A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhang, R. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Derr, T. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yang, Z. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Qi, P. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhang, S. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Bengio, Y. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Cohen, W. W. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Salakhutdinov, R. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + CONFERENCE + The conference where the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" was presented + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yao, J.-g. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wan, X. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Xiao, J. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PUBLICATION + The journal where the paper "Recent advances in document summarization" was published + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yao, L. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models"Yao, L. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Peng, J. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Mao, C. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Luo, Y. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhang, J. is an author of the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhang, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Gan, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Wang, C. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zheng, L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zheng, L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Zheng, L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools like MT-Bench and Chatbot Arena. These contributions highlight Zheng, L.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR domains. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Chiang, W.-L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Chiang, W.-L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Chiang, W.-L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Chiang, W.-L.'s active involvement in advancing the understanding and capabilities of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Sheng, Y. is an author known for contributing to the field of Natural Language Processing and Information Retrieval. Notably, Sheng, Y. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Sheng, Y. has contributed to the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Sheng, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic and technical community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Wu, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Wu, Z. co-authored the paper titled "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Wu, Z. is also credited with co-authoring the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Wu, Z.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhuang, Y. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zhuang, Y. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness of knowledge graphs. Additionally, Zhuang, Y. has also authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Zhuang, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the domain. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Lin, Z. is an author of the paper "Exploring large language models for knowledge graph completion" and also contributed to the paper "Judging LLM-as-a-judge with MT-Bench and Chatbot Arena." These works indicate Lin, Z.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the application of large language models for tasks such as knowledge graph completion and the evaluation of language models in judgment tasks. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Li, D. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant research. Notably, Li, D. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Li, D. has also co-authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Li, D.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Xing, E. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Xing, E. contributed to the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Xing, E.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + TECHNOLOGY + Chatbot Arena is a platform or tool used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Darren Edge and Ha Trinh co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Steven Truitt and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Steven Truitt is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Jonathan Larson is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 4.0 + RAG (Retrieval-Augmented Generation) and LLM (Large Language Models) are closely intertwined in the domain of Natural Language Processing and Information Retrieval. RAG is employed to enhance the capabilities of LLMs by enabling them to retrieve pertinent information from external knowledge sources. This symbiotic relationship allows LLMs to generate and assess text more effectively. Specifically, RAG leverages the power of LLMs to access and utilize relevant data, thereby augmenting the overall performance and accuracy of text generation tasks. + e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 7.0 + Graph RAG is a specific implementation of RAG that combines the strengths of RAG with graph-based text indexing. This method leverages the natural modularity of graphs to partition data, facilitating global summarization. As a specialized approach within the RAG framework, Graph RAG enhances the capabilities of RAG by integrating graph structures to improve the efficiency and effectiveness of text data processing and summarization. + 21e52bc06a82796b1f4bcd73edda1f2a,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Query-Focused Summarization is a task that RAG fails to address effectively + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + RAG retrieves relevant information from an external knowledge source + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Naive RAG is a specific implementation of RAG + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Ram et al., 2023 discusses RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Naïve RAG is a basic form of RAG + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Modular RAG is an advanced form of RAG + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used in various RAG tasks such as knowledge graph creation and completion + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Trajanoska et al. discusses using LLMs for knowledge graph creation, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Yao et al. discusses using LLMs for knowledge graph completion, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Ban et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Zhang et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Gao et al. discusses advanced RAG where the index is a knowledge graph + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + KAPING is a method where the index is a knowledge graph, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + G-Retriever is a method where subsets of the graph structure are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Graph-ToolFormer is a method where derived graph metrics are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + SURGE is a method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + FABULA is a method where retrieved event-plot subgraphs are serialized using narrative templates, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Wang et al. discusses a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Sensemaking questions are used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The evaluation of RAG systems focuses on corpora in the region of 1 million tokens + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Trade-offs are considerations involved in building a graph index for RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + A graph index is a data structure used in RAG systems to organize and retrieve information + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Performance of RAG systems varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Different data types are used in RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Dataset sizes affect the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Evaluation is the process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Corpora are collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Different question types are used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Target metrics are specific measures used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 4.0 + Graph RAG utilizes Large Language Models (LLMs) to construct a graph-based text index, enabling the generation of summaries and the answering of queries. In this approach, LLMs play a crucial role in analyzing and generating text based on the information retrieved through the graph structure. Additionally, LLMs leverage the Graph RAG framework to provide comprehensive overviews of public figures in the entertainment industry. This integration of LLMs within Graph RAG enhances the system's ability to process and synthesize complex text data, making it a powerful tool for information retrieval and natural language processing tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Text chunks are processed using LLM to extract elements of a graph index + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM is used to extract elements of a graph index from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + LLM (Large Language Model) and Few-Shot Examples are closely related in the context of Natural Language Processing and Information Retrieval. Few-shot examples are provided to the LLM for in-context learning, which helps tailor the extraction prompt. This technique is particularly useful for improving the performance of the LLM in specialized domains. By leveraging a small number of examples, the LLM can better understand and adapt to specific tasks, thereby enhancing its overall effectiveness in extracting and processing information within those specialized areas. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM extracts named entities from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Kuratov et al. (2024) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Liu et al. (2023) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM prompts are instructions given to the LLM for extracting elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Recall degradation occurs with longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + The extraction process involves using LLM to identify and extract elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Default prompt is the standard set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Secondary extraction prompt is an additional set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + The LLM uses covariate prompts to extract additional attributes associated with detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM uses multiple rounds of gleanings to ensure no entities are missed + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Logit bias is used to force a yes/no decision from the LLM during entity extraction + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM extracts element instances from source texts + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM detects and summarizes communities of entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + LLM generates intermediate answers and scores for each chunk + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + LLM generates a helpfulness score for each answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + LLM is used to generate questions for evaluating the Podcast Transcripts dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + LLM is used to generate questions for evaluating the News Articles dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + LLM uses Naive RAG to list public figures mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + LLM-generated responses are evaluated using assessment metrics + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + LLM-generated responses are evaluated using specific questions + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Ad-hoc LLM use involves the use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + LLMs are used for knowledge graph creation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph completion + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for the extraction of causal graphs + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph creation as per Trajanoska et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph completion as per Yao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for the extraction of causal graphs as per Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is proposed as a method to combine the strengths of RAG and QFS + e8d83e6e7a7c0f57b218cef24976b745 + + + 7.0 + Graph RAG is an approach that utilizes community summaries in various capacities to enhance its functionality. Specifically, community summaries are generated within the Graph RAG framework and are employed to generate partial responses. These summaries are also used to compare against source texts, serving as a form of self-memory for the system. By leveraging community summaries, Graph RAG aims to improve the comprehensiveness and diversity of answers. Additionally, Graph RAG incorporates summaries of root-level communities within an entity-based graph index, further enriching its analytical capabilities. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is designed to handle global sensemaking questions over large datasets + e8d83e6e7a7c0f57b218cef24976b745 + + + 2.0 + Graph RAG is implemented in Python. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The open-source implementation of Graph RAG will be available at this URL + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Graph RAG uses an entity knowledge graph to index text + e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG is an approach that is evaluated for its comprehensiveness, a target quality used to assess its effectiveness. The method aims to improve the comprehensiveness of generated answers, ensuring that the information provided is thorough and complete. This evaluation metric is crucial in determining the success of Graph RAG in producing detailed and accurate responses. + 21e52bc06a82796b1f4bcd73edda1f2a,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG is an approach in the domain of Natural Language Processing and Information Retrieval that focuses on improving the diversity of generated answers. Diversity, in this context, is a target quality used to evaluate the performance of the Graph RAG approach. By enhancing the diversity of responses, Graph RAG aims to provide a broader range of answers, thereby improving the overall effectiveness and robustness of the system. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG uses a knowledge graph for global summarization + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Community detection algorithms are used in the Graph RAG approach to partition graphs + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Podcast transcripts are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + News articles are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 2.0 + Graph RAG is evaluated using the target quality of Empowerment. Empowerment is specifically utilized to assess Graph RAG's capability in aiding users to achieve an informed understanding. This evaluation metric underscores the importance of user comprehension and the effectiveness of the Graph RAG approach in facilitating informed decision-making processes. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Graph RAG is compared to naive RAG in the evaluation. In this comparison, Graph RAG outperformed naive RAG on comprehensiveness and diversity. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Graph RAG is compared to global map-reduce summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Query-focused summarization is a method used in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Activity-centered sensemaking questions are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 2.0 + The "Graph RAG" approach is evaluated in terms of its performance by considering "Token Costs." Token costs are measured to assess the efficiency of the Graph RAG method, indicating that the computational expense associated with processing tokens is a critical factor in determining the overall effectiveness of this approach. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Data flow describes the high-level process of the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 3.0 + Design parameters are key settings in the Graph RAG approach and significantly influence the Graph RAG approach and pipeline. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Graph RAG uses global summarization to summarize information from a large dataset + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG aims to answer specific queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG uses a corpus for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Activity-centered sensemaking is used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Real-world datasets are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG is compared to source text summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Low-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Intermediate-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + High-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + The Graph RAG approach involves a specific pipeline for processing and summarizing text + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Techniques are specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Implementation details are specific configurations used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Graph RAG is a specific implementation of RAG systems + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + Graph RAG is a system that utilizes root-level community summaries, denoted as C0, to answer user queries. C0 represents these root-level community summaries within the Graph RAG analysis, serving as a foundational element in the system's ability to map out relationships and understand the structural dynamics within specialized communities. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses high-level community summaries (C1) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses intermediate-level community summaries (C2) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 2.0 + Graph RAG utilizes low-level community summaries, represented by C3, to answer user queries. C3 plays a crucial role in the Graph RAG analysis by providing detailed summaries of community structures, which are essential for effectively addressing user inquiries. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + 2.0 + Graph RAG is a key entity in the analysis, serving both as a condition being compared and as a tool for comparing multiple conditions. This dual role highlights its significance in the study, where it functions not only as a subject of comparison but also as a methodological framework for evaluating other conditions. The analysis likely involves a detailed examination of various conditions, with Graph RAG playing a central role in facilitating these comparisons. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses different levels of graph communities to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The Graph RAG mechanism uses an LLM evaluator for head-to-head comparison + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Graph RAG is a multi-stage mechanism + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Graph RAG mentions Taylor Swift as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Travis Kelce as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Britney Spears as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Justin Timberlake as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG is determined to be the winner based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Answer 1 was generated using the Graph RAG method, which provides a comprehensive and structured overview of public figures across various sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Graph RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Graph RAG is compared with source texts for answer comprehensiveness and diversity + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + TS represents source text summarization in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Root-level summaries are used in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Answer comprehensiveness is used to evaluate the performance of Graph RAG + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Win rate is used to measure the success rate of Graph RAG in providing comprehensive and diverse answers + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Element extraction prompts are used in Graph RAG to retain specific details in the index + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Graph RAG incorporates the concept of self-memory + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of iterative retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of federated retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts used in multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts used in multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG uses a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of a tree of clarifications + f35de4d9fb65f1d5a392064b20545c19 + + + 3.0 + Graph RAG utilizes a self-generated graph index. This self-generated graph index is a crucial component of Graph RAG, enabling it to efficiently manage and retrieve information within its graph-based framework. The use of a self-generated graph index suggests that Graph RAG has an inherent capability to construct and maintain its indexing structure, which likely enhances its performance and adaptability in handling complex data relationships. + e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Gao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Cheng et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Mao et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Shao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Wang et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Su et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Feng et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Trivedi et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Khattab et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Sarthi et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Kim et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG generates community answers in parallel + f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is compared to a graph-free approach for global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG is compared to map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses rich text annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses a hierarchical community structure + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can operate using embedding-based matching + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can be part of hybrid RAG schemes + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can employ map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can extend operations across the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Alonso contributed to the work on Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG includes local graph RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses an entity-based graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + 2.0 + NebulaGraph launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Community summaries are used to generate partial responses + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Community summaries are created from graph communities + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Community answers are created from community summaries. These answers are generated by synthesizing information from the summaries provided by the community, ensuring that the responses are comprehensive and reflective of the collective knowledge and insights within the community. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Domain-tailored summarization is used to create community summaries + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Community descriptions are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Partial answers are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Community summaries are created for each level in the hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Community summaries are useful for understanding the global structure and semantics of the dataset + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Community summaries are used to answer global queries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are generated from root communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are generated from sub-communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are added to the LLM context window until the token limit is reached + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Global answers are generated from community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The level of summary detail affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The scope of information affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are used for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Community summaries are divided into chunks of pre-specified token size + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Summary detail and scope affect the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are divided into chunks + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Community summaries are prepared to answer user queries + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Intermediate answers are generated from community summaries + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Community summaries are part of the graph community hierarchy + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Community summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Summaries of root-level communities are used in Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Global sensemaking questions are evaluated over datasets in the 1 million token range + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Global sensemaking questions are directed at an entire text corpus + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The Python-based implementation of Graph RAG approaches will be available at this URL + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Query-focused summarization is used to produce the global answer + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Map-reduce is used for query-focused summarization of an entire corpus + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Query-focused summarization is used for answering global queries + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + An entity knowledge graph is derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + + + 2.0 + In the domain of Natural Language Processing and Information Retrieval, "SOURCE DOCUMENTS" and "TEXT CHUNKS" are closely related entities. Text chunks are extracted from source documents, specifically for processing in the Graph RAG (Retrieval-Augmented Generation) approach. This method involves breaking down the source documents into smaller, manageable pieces, or text chunks, which can then be effectively utilized in various computational processes, enhancing the efficiency and accuracy of information retrieval and generation tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Intermediate-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Low-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Document corpus consists of source documents being processed + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Partial responses are summarized to generate a final response + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The LLM evaluator assesses answers based on the comprehensiveness metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Naive RAG is evaluated for comprehensiveness + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Comprehensiveness is a metric used to determine the decision + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) was universally better for comprehensiveness + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The final evaluation prioritized comprehensiveness in answers + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Global approaches achieved higher comprehensiveness win rates + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The LLM evaluator assesses answers based on the diversity metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on diversity + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The final evaluation prioritized diversity in answers + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Global approaches achieved higher diversity win rates + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Human endeavors rely on sensemaking to understand and reason about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Human endeavors rely on analyzing document collections for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + LLMs are used to automate sensemaking in complex domains + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Microsoft uses LLMs for automating sensemaking in scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Ranade uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Joshi uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + LLM prompts are used to tailor the responses of large language models + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Ranade and Joshi discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + GPT is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Llama is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Gemini is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Kuratov et al., 2024, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Liu et al., 2023, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Sensemaking is applied in the domain of scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Sensemaking is applied in the domain of intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Klein defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Klein et al. defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Element instances are extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Entity references are extracted from text chunks during processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Element instances are converted into element summaries by the LLM (Large Language Model). Element summaries are created from element instances, indicating a transformation process facilitated by the LLM. This process involves the LLM taking detailed element instances and generating concise element summaries, which encapsulate the essential information from the instances. + 2c6ed90897310eea2f28e33fff1c32b0,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Covariates are additional attributes associated with extracted element instances + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Domain-tailored summarization is used to create element summaries + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Element summaries include descriptions of entity nodes + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries include descriptions of relationship edges + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries include descriptions of claim covariates + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries are used to understand the structure and semantics of graph communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Element summaries include descriptions of nodes + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Element summaries include descriptions of edges + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Element summaries include descriptions of covariates + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Sub-community summaries are used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Community detection is a technique used to identify graph communities. Graph communities are groups of nodes within a graph that are more densely connected to each other than to the rest of the graph. This process of identifying such communities is crucial for understanding the structural dynamics and relationships within complex networks, particularly in the domain of Natural Language Processing and Information Retrieval. By leveraging community detection algorithms, researchers can uncover hidden patterns and insights within large datasets, facilitating more effective data analysis and interpretation. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Global answer is created from community answers + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Global answers are generated in response to user queries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Global answer is generated by sorting intermediate answers based on helpfulness scores + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Intermediate answers are combined to form the global answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + The final context window is used to generate the global answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Graph RAG pipeline operates at indexing time + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Graph RAG pipeline operates at query time + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Nodes are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Edges are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Covariates are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Leiden method is used in the graph RAG pipeline for community detection + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Graph RAG pipeline uses the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + The Podcast dataset graph consists of 8564 nodes + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The News dataset graph consists of 15754 nodes + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The Podcast dataset graph consists of 20691 edges + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The News dataset graph consists of 19520 edges + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Traag contributed to the development of the Leiden method + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Traag et al. are the authors of the Leiden algorithm and developed the Leiden method. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Leiden is a specific type of community detection algorithm used in various analytical pipelines. It is designed to identify and map out the structural dynamics within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. The algorithm is known for its efficiency and accuracy in detecting community structures, making it a valuable tool for researchers and practitioners in the field. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Leiden is known for its ability to recover hierarchical community structures efficiently + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The Leiden algorithm is used to detect graph communities in the MultiHop-RAG + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Figure 3 shows graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Lewis contributed to the development of the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Lewis et al. developed the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Lewis et al., 2020, are the authors who established the RAG approach + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Kevin Scott is the CTO of Microsoft + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + Microsoft conducted a study on the impact of large language models on scientific discovery using GPT-4 + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Preprint is available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Baumel, T. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Elhadad, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Es, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + James, J. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Espinosa-Anke, L. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Schockaert, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, Z. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, X. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Zhao, D. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Yang, M. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Qin, B. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + LangChain is an organization that has published on arXiv + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Wang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zuccon, G. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, R. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Derr, T. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Xu, Y. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lapata, M. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, J. published the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Gan, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yao, L. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, C. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Chiang, W.-L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Sheng, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wu, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lin, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Li, D. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Xing, E. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Preprint is classified under cs.CL on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Preprint was submitted on 24 Apr 2024 + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Preprint has the identifier 2404.16130v1 on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Community detection results in the partition of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The pipeline includes a step for community detection + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 2.0 + Dang, 2006, is the author who established the QFS approach + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Baumel et al., 2018, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Laskar et al., 2020, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Yao et al., 2017, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Goodwin et al., 2020, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Laskar et al., 2022, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Liu and Lapata, 2019, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Achiam et al., 2023, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Brown et al., 2020, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Touvron et al., 2023, are the authors who worked on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Anil et al., 2023, are the authors who worked on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Modularity is an inherent quality of knowledge graphs + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Brown et al. (2020) discuss in-context learning with few-shot examples + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Kuratov et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Liu et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Louvain is a type of community detection algorithm + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Community detection algorithms are used to partition the graph index into communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Fortunato has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Jin et al. have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + HotPotQA dataset is used to evaluate the entity extraction prompt with gpt-4-turbo + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Yang et al. (2018) introduced the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Yang et al. are the authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + GPT-4-Turbo was tested with varying context window sizes + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Tech journalist uses podcast transcripts to look for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + + + 3.0 + Kevin Scott is a participant in the podcast conversations compiled in the Podcast Transcripts dataset. His conversations are included as part of the podcast transcripts, contributing to the overall content and discussions captured within this dataset. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Technology leaders participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + RAG systems are used to evaluate the Podcast Transcripts dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + C0 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C1 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C2 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C3 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + TS is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + SS is a category used in the analysis of podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Units are used to measure the context in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Both are datasets used in the analysis + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Educator uses news articles to incorporate current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + RAG systems are used to evaluate the News Articles dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + C0 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Map-reduce is the method used in the text summarization condition + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The LLM evaluator assesses answers based on the empowerment metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on empowerment + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Empowerment has an average win rate of 51.3% + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Naive RAG mentions Taylor Swift as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Travis Kelce as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Britney Spears as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Justin Timberlake as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG is determined to be the loser based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Global approaches consistently outperformed the naive RAG + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Naive RAG produces the most direct responses + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + SS represents naive RAG in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Gao et al., 2023 discusses naive RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Community partitions enable divide-and-conquer global summarization + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Global summarization can be performed using a graph-free approach + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Source texts are used in global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Final global answer is generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Short descriptions are used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Low-level community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + The use of rich descriptive text for homogeneous nodes in a graph index aligns with the capabilities of LLMs + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Graph indexes differ from typical knowledge graphs in their use of rich descriptive text instead of concise knowledge triples + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The graph index supports condition C0 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C1 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C2 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C3 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index was created using generic prompts for entity and relationship extraction + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Few-shot examples tailored to the domain of the data were used in the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The decision to build a graph index depends on the expected number of lifetime queries per dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The decision to build a graph index depends on the value obtained from it + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The decision to build a graph index depends on the value obtained from other graph-related RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Recall measures the completeness of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Precision measures the accuracy of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to tailor the default prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to tailor the secondary extraction prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of science + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of medicine + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of law + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + A single extraction round is part of the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Domain refers to the specific area of knowledge of the document corpus + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Covariate prompts are used to extract claims linked to detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Source text span is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Start date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + End date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Description is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Subject is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Object is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Communities of entities help manage variations in a noisy graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Common entities are described using rich descriptive text for homogeneous nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + LLMs are used to generate metrics for evaluating natural language generation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Wang et al. (2023) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Zheng et al. (2024) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Relationship edges connect homogeneous nodes in a graph + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Edge weights represent the normalized counts of detected relationship instances on relationship edges + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Each level of the hierarchical community structure provides a community partition + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 2.0 + The hierarchical community structure is a framework used to organize and understand the relationships and dynamics within specialized communities. Root communities are an integral part of this structure, serving as the top-level communities. These root communities form the foundational layer in the hierarchical community structure, providing a basis for further subdivision and organization of more specific sub-communities. This hierarchical approach allows for a systematic analysis of complex networks, facilitating a deeper understanding of the interconnections and dependencies within the domain of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + The hierarchical community structure is a framework that organizes communities into different levels, with sub-communities representing the lower-level communities within this structure. Sub-communities are integral components of the hierarchical community structure, indicating that they are nested within larger communities and contribute to the overall organization and dynamics of the community. This hierarchical arrangement allows for a more detailed and nuanced understanding of the relationships and interactions within the community, facilitating more effective analysis and mapping of complex text data, particularly in specialized domains such as Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community levels are part of the hierarchical community structure + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The Leiden algorithm is used to detect communities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + OpenORD is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Force Atlas 2 is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Nodes represent entities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Edges represent connections between nodes in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Covariates are variables linked to nodes and edges in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Tang and Yang are the authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Questions are generated based on the target datasets + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + N represents the number of test questions per dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Root communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Level 0 represents the root-level communities in the hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Reports provide detailed information about specific subtopics within sub-communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Sub-communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Level 1 represents sub-communities within the root-level communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Partitions can be organized into a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Level 0 is the root level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Level 1 is a sub-level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The token limit defines the maximum number of tokens in the LLM context window + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Prominence is used to prioritize community edges + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Combined source and target node degree is used to measure prominence + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Chunks are divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Helpfulness scores are assigned to intermediate answers + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in episodes dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in how guests perceive the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in discussions about the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in discussions about collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in how news articles address the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in highlighting the importance of health literacy through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + The size of the context window and the prompts used for answer generation are the same across all conditions + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The task is an activity or goal that the user aims to achieve + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Questions are generated based on the user's task + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Datasets were used in combination with questions for the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Questions were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + + + 2.0 + Zheng et al. are the authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Zheng, L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Xing, E. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + MT-Bench and Chatbot Arena are both tools used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 2.0 + Koesten et al. authored a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + Xu and Lapata authored a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Text summarization method applies a map-reduce approach directly to source texts (TS) + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Text summarization is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Semantic search RAG is a na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached (SS) + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Semantic search RAG is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C0 uses root-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C0 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C0 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C0 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 uses high-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C1 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C1 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C1 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 uses intermediate-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C2 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C2 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C2 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 uses low-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C3 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C3 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C3 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + TS is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + SS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The graph indexing process used 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + A graph was created for the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Intermediate-level summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + The graph indexing process used 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + A graph was created for the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Datasets were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Wang et al., 2023a discusses the state-of-the-art results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Zheng et al., 2024 discusses the competitive results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Zheng et al., 2024 discusses the LLM-as-a-judge method + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Embedding-based matching is used to match user queries + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Query-time LLM use was evaluated with different context window sizes + 4c855404ee3d3c94aa2136f1513c666f + + + 2.0 + The **CONTEXT WINDOW SIZE** and **FINAL EVALUATION** are closely related in the given data. A fixed context window size of 8k tokens was used for the final evaluation. This indicates that during the final evaluation phase, the system or model was configured to process and analyze text data within a predefined window of 8,000 tokens, ensuring consistency and standardization in the evaluation process. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Natural Language Generation achieves state-of-the-art results + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation achieves competitive results + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation is compared against human judgements + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation can generate reference-based metrics + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation can measure qualities in a reference-free style + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Es et al., 2023 discusses the RAGAS method + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates context relevance + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates faithfulness + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates answer relevance + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator assesses answers based on the directness metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Table 2 shows an example of LLM-generated assessment + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses a head-to-head comparison approach + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator assesses answers based on target metrics + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses a control metric for validity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator accounts for stochasticity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses mean scores from multiple comparisons + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Directness is used to evaluate the straightforwardness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The question asks about public figures mentioned in entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Public figures are repeatedly mentioned across various entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Answer 1 covers a wide range of public figures from different sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Controversies involve public figures and impact public discourse. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Entertainment articles cover topics related to the entertainment industry + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Taylor Swift is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Travis Kelce is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Britney Spears is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Justin Timberlake is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Taylor Swift is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Travis Kelce is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Britney Spears is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Justin Timberlake is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Actors and Directors are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Musicians and Executives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Athletes and Coaches are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Influencers and Entrepreneurs are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Public Figures in Controversy are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Film is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Television is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Music is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Sports is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Digital Media is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Cultural Narratives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Trends are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Social Discussions are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Public Discourse is a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Reference-based metrics require gold standard answers + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Gold standard answers are lacking for sensemaking questions + 322e02986c8724eedbcf3ebfa20b989c + + + 3.0 + End users play a crucial role in the validation process of sensemaking questions and target metrics. Sensemaking questions are specifically validated with end users to ensure their relevance and accuracy. This collaborative approach ensures that the questions and metrics are aligned with the needs and expectations of the end users, thereby enhancing the overall effectiveness and applicability of the sensemaking process. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Target metrics are validated with end users + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The control metric is used as an indicator of validity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Taylor Swift is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Taylor Swift is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Travis Kelce is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Travis Kelce is a public figure in the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Britney Spears is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Britney Spears is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Justin Timberlake is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Justin Timberlake is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the film sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the television sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on public figures primarily from the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on public figures primarily from the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the digital media sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 cites specific data sources from the News article dataset for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 provides insights into controversies involving public figures and their impact on public discourse. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the gaming sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 cites specific data sources for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 was generated using the Naïve RAG method, which directly lists specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d + + + 2.0 + ANSWER 2 is a generated answer for a question in the NEWS ARTICLE DATASET. It relies heavily on a single source from the NEWS ARTICLE DATASET for data. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Answer 2 relies heavily on a single data source. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Naïve RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The global approach to Graph RAG shows improvements over naïve RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + LLM-generated assessments are used to evaluate the answers produced for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Example question is part of the News article dataset used for analysis + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Head-to-head win rate percentages were used to compare different conditions + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Win rate percentages were used to measure the performance of different conditions + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The overall winner per dataset and metric was determined for each condition + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Self-win rates were shown as the expected 50% for each condition + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The indexing process resulted in the creation of graphs + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Map-reduce summarization requires the highest number of context tokens + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Root-level community summaries require dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + + + 2.0 + Queries are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Self-memory is related to generation-augmented retrieval + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + CAiRE-COVID is a system for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + ITRG is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + IR-CoT is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + DSP is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + RAPTOR is a method for generating a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + The paper by Baek et al. discusses the KAPING method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by He et al. discusses the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Zhang discusses the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Kang et al. discusses the SURGE method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Ranade and Joshi discusses the FABULA method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Both LangChain and LlamaIndex support a variety of graph databases + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LangChain supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LangChain supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 1.0 + LangChain developed Langchain graphs + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + LlamaIndex supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LlamaIndex supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + NaLLM is a method that can create and reason over knowledge graphs in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Neo4J developed Project NaLLM + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + GraphRAG is a method that can create and reason over knowledge graphs in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Manakul et al. discusses the SelfCheckGPT method + 92e93fc6449756c0a60200636b297f65 + + + 1.0 + SelfCheckGPT is an approach mentioned in the work by Manakul et al., 2023 + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + SelfCheckGPT is used to compare fabrication rates + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Embedding-based matching is used to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Hybrid RAG schemes combine embedding-based matching against community reports + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The roll-up operation can be extended using map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The drill down mechanism follows the information scent in the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The global approach to Graph RAG achieves competitive performance at a fraction of the token cost + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The open-source implementation of Graph RAG approaches is Python-based + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The drill down mechanism follows the information scent + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Alonso Guevara Fernández and Amber Hoak both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Sarah Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Shane Solomon both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Adler co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Altenschmidt and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Altenschmidt and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Altman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and S. Borgeaud co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Schalkwyk and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Schalkwyk and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + A. M. Dai and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Baek and A. F. Aji co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Baek and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + A. F. Aji and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + T. Ban and L. Chen co-authored the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + 1.0 + Baumel, T. and Eyal, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Baumel, T. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Baumel, T. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Elhadad, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Guillaume, J.-L. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Guillaume, J.-L. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Guillaume, J.-L. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Lambiotte, R. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Mann, B. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Zhao, D. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Es, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + James, J. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Espinosa-Anke, L. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Schockaert, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, Z. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, X. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Yang, M. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Qin, B. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Gao, Y. and Xiong, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Dai, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Dai, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Sun, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Goodwin, T. R. and Savery, M. E. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Goodwin, T. R. and Demner-Fushman, D. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + 2.0 + Khattab, O. and Santhanam, K. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and highlights their collaborative work in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Khattab, O. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This collaboration is mentioned in the text, highlighting their joint contribution to the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Santhanam, K. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Li, X. L. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hall, D. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Liang, P. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + LIANG, P. and ZAHARIA, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Liang, P. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Potts, C. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Potts, C. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Zaharia, M. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Kim, S. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Park, J. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Park, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kang, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. and Moon, B. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Moon, B. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Moon, B. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoffman, R. R. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Gregory, K. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Groth, P. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Groth, P. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Simperl, E. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kuratov, Y. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Bulatov, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Anokhin, P. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Sorokin, D. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Sorokin, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Burtsev, M. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Laskar, M. T. R. and Hoque, E. co-authored two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning with transformer models to improve the effectiveness of query-focused abstractive summarization. Both works contribute to advancing the application of transformer models in specialized summarization tasks. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Laskar, M. T. R. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Hoque, E. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoque, E. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoque, E. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Hoque, E. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Huang, J. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + The 33rd Canadian Conference on Artificial Intelligence is also known as Canadian AI 2020 + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Springer published the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Huang, J. X. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Perez, E. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Karpukhin, V. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Xu, Y. and Lapata, M. co-authored the paper "Text summarization with latent queries" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Huang, M. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Duan, N. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 3.0 + Martin, S. and Brown, W. M. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with large graph structures, making it a valuable resource for researchers and practitioners in the domain of graph theory and network analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Martin, S. and Klavans, R. co-authored the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Martin, S. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the capabilities and applications of the Openord toolbox, emphasizing its utility in handling extensive graph data efficiently. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Brown, W. M. and Klavans, R. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Brown, W. M. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + KLAVANS, R. and BOYACK, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Newman, M. E. published the paper "Modularity and community structure in networks" in the Proceedings of the National Academy of Sciences + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Levine, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Shashua, A. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Shashua, A. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Leyton-Brown, K. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ranade, P. and Joshi, A. co-authored the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Abdullah, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Yang, Z. and Manning, C. D. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Huang, M. and Duan, N. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Xu, Y. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Siddique, F. B. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Siddique, F. B. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Barezi, E. J. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Tang, Y. and Yang, Y. co-authored the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Martin, L. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bhargava, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bhosale, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Liang, Y. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 2.0 + Li, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Li, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Li, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 2.0 + Xu, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Xu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Qu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Khramtsova, E. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wang, Y. and Lipka, N. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, R. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Qi, P. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Zhang, S. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Bengio, Y. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Cohen, W. W. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Salakhutdinov, R. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Chiang, W.-L. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Xing, E. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + \ No newline at end of file diff --git a/graphfleet/output/graphindex/artifacts/embedded_graph.2.graphml b/graphfleet/output/graphindex/artifacts/embedded_graph.2.graphml new file mode 100644 index 000000000..829426858 --- /dev/null +++ b/graphfleet/output/graphindex/artifacts/embedded_graph.2.graphml @@ -0,0 +1,9715 @@ + + + + + + + + + + + PERSON + Darren Edge is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Ha Trinh is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Newman Cheng is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Joshua Bradley is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Alex Chao is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Apurva Mody is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Steven Truitt is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Jonathan Larson is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Research is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Strategic Missions and Technologies is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Office of the CTO is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + METHOD + RAG (Retrieval-Augmented Generation) is a developing research area with multiple established directions, including knowledge graph creation, completion, and extraction of causal graphs. It is a method used for generating responses in text generation tasks by retrieving relevant information from an external knowledge source to enable large language models to answer questions. This approach incorporates the retrieval of relevant data to augment text generation, producing direct responses in various text generation tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY + LLM (Large Language Model) is a type of artificial intelligence model used for a variety of tasks in the field of Natural Language Processing and Information Retrieval. These tasks include generating and assessing text, entity extraction, summarization, understanding relationships in text, and automating human-like sensemaking and reasoning over large collections of documents. LLMs are also employed to generate intermediate answers and scores for text chunks, process these chunks to extract elements of a graph index, and automate the generation of questions for dataset evaluation. Additionally, LLMs can analyze and generate text based on retrieved information and queries, and they possess a context window that can be exceeded by external datasets. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,2c6ed90897310eea2f28e33fff1c32b0,6f33a085ff3304e5994f7fbb86c881a4,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + METHOD + Graph RAG (Retrieval-Augmented Generation) is a sophisticated method that leverages the natural modularity of graphs to partition data for global summarization tasks. This approach integrates multiple stages and concepts, including knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS), to support human sensemaking over text corpora. It is particularly effective in providing comprehensive and structured overviews of public figures across various sectors of the entertainment industry, as well as generating answers for questions in the News article dataset. + +Graph RAG employs a high-level data flow and pipeline for processing and summarizing text, combining both global and local approaches to optimize token usage in text generation tasks. It uses community summaries to improve answer comprehensiveness and diversity while requiring fewer tokens compared to traditional source text summarization methods. This method has been shown to outperform naive RAG in terms of comprehensiveness and diversity in text generation tasks. + +A specific implementation of Graph RAG involves using four levels of graph communities, incorporating concepts from other systems such as self-memory and parallel generation of community answers. This multi-stage mechanism allows for the comparison of multiple conditions, enhancing the overall effectiveness of the summarization process. + +Graph RAG, launched by NebulaGraph, is a retrieval-augmented generation technology based on knowledge graphs. It combines global summarization with graph-based text indexing to answer questions over private text corpora, making it a versatile tool for various text analysis and summarization applications. + 086021a89900a39bcb62036981737bfa,21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,718017a4871c909420f84b85b8ba969d,833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19,fb3c48579608fa28be585ceb6cd2f0fe + + + METHOD + QFS (Query-Focused Summarization) is a method used to generate summaries based on specific user queries + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + Community summaries are generated summaries of data clusters or communities, used to answer queries and tailored to the domain. These summaries are pre-generated for groups of closely-related entities, particularly in the Graph RAG approach, and are derived from community-generated content to compare with source texts. They are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks. Community summaries provide report-like insights into each community within a hierarchical structure, which is useful for understanding the dataset. They are generated from modular communities in the knowledge graph and cover different levels of each graph community hierarchy, including root-level communities in an entity-based graph index. Additionally, these summaries act as a kind of self-memory for generation-augmented retrieval. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + CONCEPT + Global sensemaking questions are questions that require understanding and summarizing large datasets, often beyond the explicit content of the source texts + e8d83e6e7a7c0f57b218cef24976b745 + + + METRIC + 1 million token range refers to the scale of datasets used in the evaluation of the Graph RAG approach + e8d83e6e7a7c0f57b218cef24976b745 + + + TECHNOLOGY + Python is a programming language used for implementing both global and local Graph RAG approaches. Additionally, Python is utilized to implement the open-source version of the Graph RAG approach. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + + + URL + The URL "HTTPS://AKA.MS/GRAPHRAG" is the location where the open-source, Python-based implementation of Graph RAG approaches will be available. This URL serves as the repository for accessing the open-source implementation of the Graph RAG approach. + e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745 + + + METHOD + Query-Focused Summarization (QFS) is a method used to generate summaries that are relevant to specific user queries. This summarization technique focuses on answering specific queries by utilizing the entire corpus of information available. It is designed to provide concise and relevant information based on the specific needs of the user, ensuring that the generated summaries are directly aligned with the queries posed. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + + + CONCEPT + An external knowledge source is a repository of information that can be accessed to retrieve relevant data for answering questions + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + A text corpus is a large collection of written texts used for analysis and research + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + Source documents are the original texts from which information is extracted, retrieved, or summarized. These documents serve as the foundational input texts for various processing tasks, including the Graph RAG (Retrieval-Augmented Generation) approach. In this context, source documents are critical for extracting and analyzing information, ensuring that the data used in computational linguistics and information retrieval tasks is accurate and comprehensive. + bc9e2c9e369c4108cf4f6dd5f60960f4,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + + + CONCEPT + A partial response is an intermediate answer generated from community summaries before being combined into a final response + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + A final response is the comprehensive answer generated after combining all partial responses + e8d83e6e7a7c0f57b218cef24976b745 + + + METRIC + COMPREHENSIVENESS is a metric used to evaluate the quality of generated responses by measuring how much detail an answer provides to cover all aspects and details of a question. It assesses the completeness and thoroughness of answers, ensuring that they encompass all relevant information. This metric is particularly important in evaluating the summarization approach, focusing on the completeness of the summary. In practical applications, such as evaluating Podcast transcripts and News articles, comprehensiveness has shown win rates between 72-83% and 72-80%, respectively. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + + + METRIC + DIVERSITY is a metric used to evaluate the variety and richness of answers generated in response to a question. It measures how varied and rich an answer is in providing different perspectives and insights. This metric is particularly important in assessing the quality of summarization approaches, focusing on the variety of information included in the summary. DIVERSITY is applied to various types of content, including Podcast transcripts, where win rates range from 75-82%, and News articles, with win rates ranging from 62-71%. It is a crucial target quality for evaluating the effectiveness of different methods in generating diverse and informative responses. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + + + ACTIVITY + Human endeavors refer to activities and efforts across various domains that rely on reading and reasoning about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models designed to understand and generate human-like text, playing a crucial role in automating sensemaking in complex domains. Modern language models, such as GPT, Llama, and Gemini, leverage in-context learning to effectively summarize content. These models are integral to the field of Natural Language Processing and Information Retrieval, enabling sophisticated text analysis and generation capabilities. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + + + DOMAIN + Scientific discovery is a complex domain where sensemaking is applied to understand and generate new knowledge from scientific texts + f0306814bf64f5c9e79603fc6a52f4ea + + + DOMAIN + Intelligence analysis is a complex domain where sensemaking is applied to understand and generate insights from intelligence data + f0306814bf64f5c9e79603fc6a52f4ea + + + PROCESS + SENSEMAKING is the process of understanding and making sense of complex information. It involves understanding connections among people, places, and events to anticipate their trajectories and act effectively. This process is crucial for navigating and interpreting intricate data landscapes, enabling individuals and organizations to make informed decisions based on the relationships and patterns identified within the information. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + TEXT CHUNKS are segments of text that are embedded into a vector space for analysis. These segments are extracted from source documents and are used for processing in the Graph RAG (Retrieval-Augmented Generation) approach. By embedding these text chunks into a vector space, they can be analyzed more effectively, facilitating various natural language processing and information retrieval tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + DATA + Element instances are identified and extracted instances of graph nodes and edges from text chunks. They represent individual occurrences of entities, relationships, and claims extracted from source texts. These specific pieces of information are tailored to the domain, providing a structured representation of the underlying data. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Element summaries are concise representations of element instances, tailored to the domain. They are descriptive texts created by the LLM to summarize entities, relationships, and claims extracted from source texts. These summaries provide detailed descriptions of nodes, edges, and covariates within a community, and are used to understand the structure and semantics of the dataset. In essence, element summaries serve as a tool to encapsulate and convey the intricate details of elements within a graph, facilitating a deeper comprehension of the dataset's structural dynamics and semantic relationships. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Graph communities are groups of elements, including nodes, edges, and covariates, detected within a graph index, primarily used for summarization. These communities consist of groups of nodes that exhibit stronger connections to each other than to nodes outside the group. This structural characteristic allows for the identification and analysis of densely connected subgraphs, which can be crucial for understanding the underlying relationships and dynamics within complex networks. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + COMMUNITY ANSWERS are query-focused summaries derived from community summaries. These answers are generated in parallel from the community summaries and are designed to respond to user queries effectively. Essentially, COMMUNITY ANSWERS serve as responses that synthesize information from community summaries to provide concise and relevant answers to specific questions posed by users. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + DATA + GLOBAL ANSWER is a comprehensive response generated from multiple community summaries to answer a user query. It is the final query-focused summary produced from all relevant community summaries. The final answer is generated by combining intermediate community answers based on their helpfulness scores. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + TIME + Indexing time refers to the time when the graph index is created and elements are summarized + f0306814bf64f5c9e79603fc6a52f4ea + + + TIME + Query time refers to the time when a query is made and the relevant summaries are generated + f0306814bf64f5c9e79603fc6a52f4ea + + + PROCESS + Graph RAG pipeline is a process using an LLM-derived graph index to detect, extract, and summarize nodes, edges, and covariates in source documents + f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + NODES are entities detected in the graph index of source documents. They represent the individual elements or points in a graph. For instance, in the Podcast dataset, there are 8,564 nodes, while the News dataset contains 15,754 nodes. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + EDGES are relationships detected in the graph index of source documents. They represent the connections or links between nodes in a graph. For instance, in the Podcast dataset, there are 20,691 edges, while the News dataset contains 19,520 edges. These edges are crucial for understanding the structural dynamics and relationships within the datasets, providing insights into how different nodes (such as topics, entities, or documents) are interconnected. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Covariates are additional attributes associated with extracted node instances in the graph index. They represent claims or additional information detected in the graph index of source documents. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + LEIDEN is a community detection algorithm renowned for its efficiency in recovering hierarchical community structures. It is widely used to partition graphs into modular communities, effectively grouping elements within a graph index. The algorithm's ability to identify and organize these communities makes it a valuable tool in the analysis of complex networks, particularly within the domains of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Retrieval-Augmented Generation (RAG) is an established approach in the field of Natural Language Processing and Information Retrieval, designed to answer user questions over entire datasets. This method involves retrieving relevant text regions to provide grounding for the generation task, thereby enhancing the accuracy and relevance of the generated responses. By combining retrieval and generation processes, RAG effectively synthesizes and presents pertinent information, making it a powerful tool for handling complex queries and large datasets. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + + + ORGANIZATION + Microsoft is a technology company whose Chief Technology Officer, Kevin Scott, actively participates in podcast conversations. The organization is deeply involved in automating sensemaking in scientific discovery through the use of large language models (LLMs). Notably, Microsoft conducted a study examining the impact of large language models, specifically GPT-4, on scientific discovery. + 1d07b4248c2655081c7af0e373bd70c9,833e7d67dcd30790b26b71c9b5306f6b,f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Ranade is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Joshi is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Klein is an author who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Lewis is an author who contributed to the development of the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Traag is an author who contributed to the development of the Leiden community detection method + f0306814bf64f5c9e79603fc6a52f4ea + + + PUBLICATION + arXiv is a preprint repository where several significant papers in the field of Natural Language Processing and Information Retrieval have been published. It serves as a platform for electronic preprints (known as e-prints) that are approved for publication after moderation, but not full peer review. Notable papers published on arXiv include "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models," "Lost in the middle: How language models use long contexts," "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," "Llama 2: Open foundation and fine-tuned chat models," "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy," "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries," "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions," "Enhancing knowledge graph construction using large language models," "Is chatgpt a good nlg evaluator? a preliminary study," "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt," "Causal graph discovery with retrieval-augmented generation based large language models," "Knowledge graph prompting for multi-document question answering," "Text summarization with latent queries," "Retrieval-augmented generation for large language models: A survey," and "Knowledge graph-augmented language models for knowledge-grounded dialogue generation." This repository is a crucial resource for researchers to disseminate their findings rapidly and access the latest advancements in their fields. + 00e8e4e881bd0862022f4dfc913b900b,086021a89900a39bcb62036981737bfa,58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035,f0306814bf64f5c9e79603fc6a52f4ea,fc4b27d64f055b7fc30176ba110dd02e + + + PUBLICATION + Preprint refers to the version of the research paper that is under review and available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + CATEGORY + cs.CL is the category under which the research paper is classified on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + DATE + 24 Apr 2024 is the date when the research paper was submitted to arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + IDENTIFIER + 2404.16130v1 is the identifier for the research paper on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Document collections refer to large sets of documents that are analyzed for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + TECHNOLOGY + LLM PROMPTS are specific instructions given to large language models (LLMs) to tailor their responses to the domain of the dataset. These prompts are also used to extract elements from text chunks, ensuring that the LLMs provide relevant and precise information based on the given context. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Community detection is a method used to identify groups of related elements within a graph. It involves the process of identifying communities within a graph, which are clusters of nodes that are more densely connected internally than with the rest of the network. This technique is crucial in understanding the structural dynamics and relationships within complex networks, such as those found in social networks, biological systems, and information retrieval systems. By uncovering these communities, researchers can gain insights into the underlying structure and function of the network, facilitating more effective analysis and interpretation of the data. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Domain-tailored summarization is a method used to create summaries that are specific to the domain of the dataset + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Klein et al. are authors who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Ranade and Joshi are authors who discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Lewis et al. are authors who developed the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Traag et al. are the authors who developed the Leiden algorithm, a method renowned for its efficiency in recovering hierarchical community structures. This algorithm is widely recognized in the field of Natural Language Processing and Information Retrieval for its ability to accurately detect and map out complex community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + QFS is a task framing that focuses on generating summaries based on specific queries, rather than just concatenating excerpts + fb3c48579608fa28be585ceb6cd2f0fe + + + METHOD + A type of summarization that generates natural language summaries based on specific queries, rather than just extracting text + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A neural network architecture that has shown substantial improvements in various summarization tasks + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + GEMINI is a family of highly capable multimodal models, as described in an arXiv preprint. These models are known for their ability to perform in-context learning and summarization, making them a significant advancement in the field of Natural Language Processing and Information Retrieval. + 086021a89900a39bcb62036981737bfa,fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A knowledge graph is a structured representation of information, utilized in the Graph RAG approach for summarization. This structured representation of knowledge is specifically employed in the Graph RAG approach for global summarization, highlighting its role in organizing and integrating information to facilitate comprehensive and coherent summaries. + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + + + REFERENCE + Authors of a paper on Retrieval-augmented generation (RAG) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Author of a paper on query-focused summarization (QFS) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "BROWN ET AL., 2020" refers to a publication by Brown et al. in 2020, which discusses in-context learning with few-shot examples. The authors of this paper are also known for their work on the GPT series of large language models. + bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "KURATOV ET AL., 2024" refers to a publication by Kuratov and colleagues in 2024. The study discusses the recall degradation and potential for information loss in longer context windows of Large Language Models (LLMs). The authors explore the limitations of these extended context windows, providing insights into how the performance of LLMs can be affected when dealing with longer sequences of text. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "LIU ET AL., 2023" refers to a publication by Liu et al. in 2023, which discusses the recall degradation and potential for information loss in longer context windows of large language models (LLMs). The authors explore the limitations of LLM context windows, highlighting how extended contexts can lead to decreased recall accuracy and information retention. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + TECHNOLOGY + COMMUNITY DETECTION ALGORITHMS are algorithms used to partition a graph into communities of nodes with stronger connections to one another. These algorithms are designed to identify modular communities of closely-related nodes within a graph, thereby revealing the underlying structure and relationships within the network. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + ALGORITHM + Louvain is a community detection algorithm used to partition graphs into modular communities + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + HOTPOTQA is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical resource for evaluating entity extraction prompts, particularly with advanced models like GPT-4-turbo. Additionally, HotPotQA is utilized to observe the behavior of text chunk extraction within the Graph RAG (Retrieval-Augmented Generation) approach, making it a versatile tool in the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNOLOGY + GPT-4-Turbo is a version of the GPT-4 model characterized by its large context size of 128k tokens, which is utilized in various analytical tasks. Specifically, GPT-4-Turbo is employed for entity extraction in evaluations, leveraging its extensive context capacity to enhance the accuracy and comprehensiveness of the analysis. This model is particularly suited for tasks within the Natural Language Processing and Information Retrieval domain, where handling large volumes of text and extracting relevant entities are critical. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + + + DATASET + The "PODCAST TRANSCRIPTS" dataset is a comprehensive collection of compiled transcripts from podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders. This dataset is used for analysis and consists of 1669 text chunks, each containing 600 tokens with 100-token overlaps between chunks, amounting to approximately 1 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620,ebf5249c888e07fedce6572a4c03f88c + + + DATASET + The "NEWS ARTICLES" dataset is a comprehensive collection of news articles used for analysis. It serves as a benchmark dataset comprising news articles published from September 2013 to December 2023. The dataset spans a range of categories, including entertainment, business, sports, technology, health, and science. It consists of 3197 text chunks, each containing 600 tokens, with a 100-token overlap between chunks, amounting to approximately 1.7 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620 + + + METHOD + MAP-REDUCE is a method employed for text summarization by applying a map-reduce approach directly to source texts. It is particularly utilized for query-focused summarization of an entire corpus, enabling efficient processing and extraction of relevant information from large datasets. This technique leverages the map-reduce paradigm to distribute the computational workload, making it suitable for handling extensive text collections in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,973164fa90bf2b4ee267f4fd795916bf + + + METRIC + "EMPOWERMENT" is a concept and metric used in the evaluation of various methods, with an average win rate of 51.3%. It measures how well an answer helps the reader understand and make informed judgments about a topic. Specifically, it evaluates the effectiveness of generated answers in empowering users by developing their understanding of broad issues and themes. Empowerment is a target quality in summarization approaches, focusing on the ability to help users reach an informed understanding. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + Naive RAG is a basic retrieval-augmented generation (RAG) method used as a baseline for comparison in text generation tasks. It converts documents to text, splits them into chunks, and embeds these chunks into a vector space for query matching. While it produces the most direct responses, it is outperformed by global approaches in terms of comprehensiveness and diversity. Naive RAG is also noted for listing public figures mentioned in entertainment articles, focusing on their professional achievements and personal lives. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c + + + METHOD + A method for summarizing source texts using a map-reduce approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Questions generated to evaluate the summarization approach, focusing on understanding activities + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + METRIC + TOKEN COSTS refer to the computational cost measured in tokens used in the summarization process. Specifically, in the context of the Graph RAG (Retrieval-Augmented Generation) approach, token costs denote the number of tokens required for processing text. This metric is crucial for evaluating the efficiency and scalability of text processing methods within the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS + The high-level process of the Graph RAG approach and pipeline + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + Design parameters are key settings and configurations in the Graph RAG approach. These parameters are crucial as they influence the design of the Graph RAG approach and pipeline, determining the effectiveness and efficiency of the overall system. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + PARAMETER + + + METHOD + GLOBAL SUMMARIZATION is a method for summarizing information on a global scale. It aims to encapsulate the overall structure and semantics of a dataset, providing a comprehensive overview of information from large datasets or corpora. This technique is particularly useful in the field of Natural Language Processing and Information Retrieval, where it helps in distilling vast amounts of data into coherent and concise summaries, facilitating better understanding and analysis of the underlying information. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e4d9b12cf2b4c691c74019eefff4fb39 + + + ATTRIBUTE + Modularity is an inherent quality of graphs that allows them to be partitioned into communities of closely-related nodes + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Descriptions generated from modular communities in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + + + INPUT + A specific question or request for information that the summarization methods aim to answer + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + A large collection of texts or documents used for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Intermediate answers generated from community summaries before being combined into a final global answer + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + The comprehensive answer generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + + + METHOD + A method that focuses on generating questions to understand activities from datasets + 21e52bc06a82796b1f4bcd73edda1f2a + + + INPUT + Brief descriptions of datasets used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + Datasets that represent real-world information, such as podcast transcripts and news articles + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + METHOD + A method that summarizes the original source texts directly + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + LOW-LEVEL COMMUNITY SUMMARIES are a type of community summary used in the News dataset for analysis. These summaries provide a detailed overview of the source text and are generated from lower hierarchical levels of the community in the knowledge graph. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + OUTPUT + INTERMEDIATE-LEVEL COMMUNITY SUMMARIES are summaries that provide a mid-level overview of the source text. These summaries are generated from intermediate hierarchical levels of the community in the knowledge graph, offering a balanced perspective that captures essential details without overwhelming the reader with excessive information. This approach ensures that the summaries are both informative and concise, making them valuable for understanding the structural dynamics and relationships within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + OUTPUT + Summaries generated from higher hierarchical levels of the community in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + + + PROCESS, SYSTEM + The entity "PIPELINE" refers to a series of processes or steps used to analyze and summarize a dataset. Specifically, in the context of the Graph RAG approach, the pipeline denotes the sequence of steps and processes involved. This structured sequence is essential for systematically handling data, ensuring that each stage of the analysis is methodically executed to achieve accurate and comprehensive results. + 7fb7d9ce2da9c940a32afdd87d1d9e56,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA STRUCTURE, OUTPUT + The "GRAPH INDEX" is a data structure used in Retrieval-Augmented Generation (RAG) systems to organize and retrieve information. It is a self-generated index that enables Graph RAG by utilizing a graph structure to organize and retrieve data. This index is created from a graph structure and is employed for tasks such as query-focused summarization. The graph index includes various elements extracted from text chunks using Large Language Model (LLM) prompts. Additionally, it supports conditions C0-C3 and is created using generic prompts for entity and relationship extraction. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + + + DATA, UNIT + Entity references are mentions of entities within text chunks, extracted during the processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC + Recall is a metric used to measure the completeness of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC + Precision is a metric used to measure the accuracy of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + FEW-SHOT EXAMPLES are specialized instances provided to the Large Language Model (LLM) to improve its performance in domains with specialized knowledge such as science, medicine, and law. These examples are tailored to the domain of the data used in the graph indexing process and serve as sample inputs for in-context learning. By tailoring the extraction prompt to the document corpus domain, few-shot examples enhance the LLM's ability to understand and process domain-specific information effectively. + 2c6ed90897310eea2f28e33fff1c32b0,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA, UNIT + Named entities are specific types of entities such as people, places, and organizations, extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + REFERENCE, PUBLICATION + A reference to a publication by Yang et al. in 2018, introducing the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METHOD, APPROACH + Techniques refer to the specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Implementation details are specific configurations and settings used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS, METHOD + A single extraction round refers to one complete cycle of extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC, ISSUE + Recall degradation refers to the decrease in recall performance when using longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS, METHOD + The extraction process involves identifying and extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Domain refers to the specific area of knowledge or field to which the document corpus belongs + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA, INPUT + Document corpus refers to the collection of documents being processed in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + Default prompt is the standard set of instructions given to the LLM for extracting named entities + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + Secondary extraction prompt is an additional set of instructions given to the LLM for extracting covariates + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METHOD + A covariate prompt is used to extract additional attributes associated with detected entities, including claims linked to the entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + Claims are statements or assertions linked to detected entities, including details such as subject, object, type, description, source text span, and start and end dates + 2c6ed90897310eea2f28e33fff1c32b0 + + + METHOD + Gleanings refer to multiple rounds of entity extraction to ensure that no entities are missed in the process + 2c6ed90897310eea2f28e33fff1c32b0 + + + TECHNIQUE + Logit bias is a technique used to force a yes/no decision from the LLM during the entity extraction process + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + An entity node is a representation of an entity in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A relationship edge is a representation of a relationship between entities in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A claim covariate is an additional attribute or variable associated with a claim in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + Communities of entities are groups of closely-related entities detected and summarized by the LLM + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + The "NOISY GRAPH STRUCTURE" refers to a graph structure that may contain inconsistencies or errors, making it challenging to analyze. This type of graph often includes duplicate or inconsistent entity elements due to variations in text format. These inconsistencies can arise from various sources, such as data entry errors, differing data formats, or incomplete information, which complicate the process of extracting meaningful insights and relationships from the graph. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + DOMAIN + Science is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + DOMAIN + Medicine is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + DOMAIN + Law is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Source text span is an attribute of a claim that indicates the specific portion of text from which the claim was extracted + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Start date is an attribute of a claim that indicates when the event or fact described in the claim began + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + End date is an attribute of a claim that indicates when the event or fact described in the claim ended + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Description is an attribute of a claim that provides a detailed explanation of the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Subject is an attribute of a claim that indicates the main entity involved in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Object is an attribute of a claim that indicates the entity that is affected by the subject in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A concept referring to an entity that has multiple name variations but is resilient to such variations due to sufficient connectivity to closely-related entities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models capable of understanding and generating human-like text. They are utilized for a variety of tasks, including the creation and completion of knowledge graphs, which are essential for structuring and interlinking information in a meaningful way. Additionally, LLMs serve as evaluators of natural language generation, assessing the quality and coherence of text produced by other AI systems. These models play a crucial role in the field of Natural Language Processing and Information Retrieval, contributing significantly to advancements in how machines comprehend and interact with human language. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf + + + TECHNOLOGY + Structured representations of knowledge in the form of triples (subject, predicate, object) used for reasoning tasks + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + Nodes in a graph that are of the same type and are described using rich descriptive text + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + Edges in a graph that represent relationships between entity nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + METRIC + Weights assigned to edges in a graph, representing the normalized counts of detected relationship instances + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The "HIERARCHICAL COMMUNITY STRUCTURE" is a framework in which communities are organized in a hierarchy, with each level providing a partition of the graph nodes. This structure organizes data into a hierarchy of communities, facilitating a multi-level clustering approach. Hierarchical community structure is utilized to generate community summaries, offering a comprehensive method for understanding the relationships and structural dynamics within specialized communities. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39 + + + CONCEPT + A division of graph nodes into mutually-exclusive, collectively-exhaustive communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + TECHNOLOGY + MULTIHOP-RAG is a benchmark dataset comprising news articles published from September 2013 to December 2023, covering a range of categories including entertainment, business, sports, technology, health, and science. It is specifically designed for open-domain question answering, targeting explicit fact retrieval. Additionally, MULTIHOP-RAG represents a specific implementation or variant of Retrieval-Augmented Generation (RAG) used in the context of graph communities. This dataset is also utilized for community detection and analysis, making it a versatile tool in the field of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author who has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + PERSON + Authors who have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The entity "DATASET" refers to a collection of data used for various purposes such as analysis, summarization, and evaluation. This can include diverse types of data like podcast transcripts and news articles. Specifically, the term encompasses datasets used for evaluation purposes, including notable examples like the Podcast and News datasets. + 1d07b4248c2655081c7af0e373bd70c9,7fb7d9ce2da9c940a32afdd87d1d9e56,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + GLOBAL QUERIES refer to questions or inquiries that require comprehensive answers derived from multiple sources or datasets. These queries aim to retrieve information from a global perspective, covering the entire dataset. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + ROOT COMMUNITIES are the top-level clusters in a hierarchical community structure. These communities represent the highest level of organization within the hierarchy, serving as the primary divisions from which more specific sub-communities branch out. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + SUB-COMMUNITIES are lower-level clusters within root communities in a hierarchical community structure, providing more detailed information. These sub-communities play a crucial role in breaking down the larger, more general root communities into more specific and focused groups, thereby facilitating a deeper and more granular understanding of the overall community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + TECHNOLOGY + Detailed documents that provide information about specific subtopics within a community + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The division of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + A system in which elements are ranked or organized in levels + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + LEVEL 0 represents the root-level communities in the hierarchical clustering with maximum modularity. It serves as the foundational layer in a hierarchical community structure, indicating the initial and most significant division of the dataset into distinct groups. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + LEVEL 1 is a sub-level in a hierarchical community structure, providing more detailed information about the internal organization. Specifically, Level 1 represents sub-communities within the root-level communities, thereby revealing the internal structure and dynamics of these larger groups. This level of granularity helps in understanding the intricate relationships and specialized interactions that occur within the broader community framework. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + A visual representation of graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + METHOD + The Leiden algorithm is a method used for detecting communities in large networks + 843fc5421e086120ffa1c75856ecf6cd + + + TOOL + OpenORD is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + + + TOOL + Force Atlas 2 is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Nodes represent entities in a graph, with size proportional to their degree + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Edges represent connections between nodes in a graph + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Covariates are variables that are linked to nodes and edges in a graph + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + The LLM context window is the token limit within which summaries are added for processing by a language model + 843fc5421e086120ffa1c75856ecf6cd + + + METHOD + Hierarchical clustering is a method of clustering data into a tree-like structure with multiple levels + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + The token limit is the maximum number of tokens that can be processed in a single context window by a language model + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Summary detail refers to the level of detail provided in a summary + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Scope refers to the range or extent of information covered in a summary + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + A "USER QUERY" is a question or inquiry posed by a user seeking information, which the system aims to answer. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd + CONCEPT + + + ELEMENT + Chunks are segments of community summaries divided into pre-specified token sizes + 843fc5421e086120ffa1c75856ecf6cd + ELEMENT + + + METRIC + Prominence is a metric used to prioritize community edges based on the combined degree of source and target nodes + 843fc5421e086120ffa1c75856ecf6cd + + + METRIC + Combined source and target node degree is a metric used to measure the overall prominence of community edges + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Community edges are connections between nodes within a community, prioritized based on prominence + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Sub-community summaries are shorter summaries of sub-communities used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + + + CATEGORY + Community level refers to the different levels in the hierarchical community structure used to generate summaries + 843fc5421e086120ffa1c75856ecf6cd + + + DATA + Chunks are segments of community summaries divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + + + METRIC + A score between 0-100 generated by the LLM to indicate how helpful an answer is in addressing the target question + 1d07b4248c2655081c7af0e373bd70c9 + + + USER + A user looking for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + + + USER + A user incorporating current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic addressing the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + The importance of health literacy highlighted through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + + + OUTPUT + Answers generated for each chunk of community summaries + 1d07b4248c2655081c7af0e373bd70c9 + + + METRIC + The pre-specified size of tokens used to divide community summaries into chunks + 1d07b4248c2655081c7af0e373bd70c9 + + + TECHNOLOGY + The "CONTEXT WINDOW" refers to a window of text used to generate answers, constrained by token size. The size of the context window is consistent across all conditions, ensuring uniformity in answer generation processes. + 1d07b4248c2655081c7af0e373bd70c9,973164fa90bf2b4ee267f4fd795916bf + + + PERSON + Kevin Scott is the Chief Technology Officer (CTO) of Microsoft and actively participates in podcast conversations. His involvement in these discussions is documented and compiled in the dataset, highlighting his contributions to the field of technology and his role in shaping Microsoft's strategic direction. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + Individuals who are leaders in the technology industry and participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + + + INPUT + A specific activity or goal that the user aims to achieve using the datasets + 1d07b4248c2655081c7af0e373bd70c9 + + + INPUT + QUESTIONS refer to specific inquiries generated by the Large Language Model (LLM) based on the user's task and the target datasets. These questions are utilized in the analysis to evaluate the performance of different methods within the domain of Natural Language Processing and Information Retrieval. The generation and subsequent use of these questions are crucial for assessing the effectiveness and accuracy of various computational techniques and models. + 1d07b4248c2655081c7af0e373bd70c9,4c855404ee3d3c94aa2136f1513c666f + + + + + 1d07b4248c2655081c7af0e373bd70c9 + + + DATASET + MT-BENCH is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical tool in evaluating the performance of models in retrieving accurate and relevant information from a broad range of topics. MT-Bench is prominently featured in the academic paper titled "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," where it is utilized to assess the capabilities of large language models in a structured and rigorous manner. This dataset plays a significant role in advancing the field of Natural Language Processing and Information Retrieval by providing a standardized metric for model comparison and evaluation. + 922778ce1cb2fdd6dbab1746c8795620,b1bbda43309e8e0e2175ea034aa88e13 + DATASET + + + PROCESS + The process through which people inspect, engage with, and contextualize data within the broader scope of real-world activities + 922778ce1cb2fdd6dbab1746c8795620 + PROCESS + + + TECHNOLOGY + Retrieval-Augmented Generation systems used for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + TECHNOLOGY + + + AUTHORS + Authors of a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors of a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + AUTHORS + Authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + + + 922778ce1cb2fdd6dbab1746c8795620 + + + PODCAST + "BEHIND THE TECH" is a podcast series featuring conversations between Kevin Scott and other technology leaders. It serves as a media platform associated with Kevin Scott, providing insights and discussions on various technological advancements and industry trends. + 833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + Kevin Scott, Microsoft CTO, who participates in the podcast conversations compiled in the dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + DATASET + A benchmark dataset for open-domain question answering, targeting explicit fact retrieval + 922778ce1cb2fdd6dbab1746c8795620 + + + METRIC + N represents the number of test questions per dataset used in the evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + METHOD + A method applying a map-reduce approach directly to source texts for summarization + 973164fa90bf2b4ee267f4fd795916bf + + + METHOD + A na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached + 973164fa90bf2b4ee267f4fd795916bf + + + CATEGORY + C0 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a root-level community summary, which is utilized to answer user queries by providing the fewest number of summaries. This category is essential for understanding the structural dynamics within the community, particularly in the domain of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C1 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a high-level community summary used to answer user queries, effectively representing sub-communities of C0. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C2 is a category or condition used in the analysis, representing a specific subset of the data. It functions as an intermediate-level community summary used to answer user queries, representing sub-communities of C1. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C3 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a category or condition that encapsulates low-level community summaries, which are instrumental in answering user queries. These summaries represent sub-communities of C2, providing detailed insights into the structural dynamics and relationships within the broader community. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + TS, or "Text Summarization," is a category or condition used in the analysis, representing a specific subset of the data. It is particularly focused on source text summarization within the analysis. TS employs a text summarization method that applies a map-reduce approach directly to source texts, facilitating efficient and scalable summarization processes. This category is integral to understanding and processing large volumes of text data, making it a crucial component in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + "SS" is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a baseline condition and is associated with a na¨ıve RAG (Retrieval-Augmented Generation) approach. In this context, text chunks are retrieved and added to the context window until the token limit is reached. + 4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CONCEPT + The prompts used for answer generation, which are the same across all conditions with minor modifications + 973164fa90bf2b4ee267f4fd795916bf + + + DATASET + The "PODCAST DATASET" is a collection of podcast transcripts utilized for both analysis and evaluation purposes. This dataset is specifically designed to support various analytical tasks, providing a rich source of textual data for researchers and practitioners in the field of Natural Language Processing and Information Retrieval. The transcripts within the dataset offer valuable insights and serve as a critical resource for evaluating different computational models and techniques. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + DATASET + The "NEWS DATASET" is a collection of news articles utilized for both analysis and evaluation purposes. This dataset serves as a valuable resource for examining and assessing various aspects of news content, making it an essential tool in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + METRICS in the context of Natural Language Processing and Information Retrieval are essential tools used to evaluate the performance of natural language generation. These metrics include both reference-based metrics, which compare generated texts to a set of reference texts, and qualities of the generated texts themselves. They are crucial in the analysis to assess the effectiveness of different methods in generating natural language, ensuring that the outputs are both accurate and of high quality. + 4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + "WANG ET AL., 2023A" refers to a study conducted by Wang and colleagues in 2023, which highlights the effectiveness of Large Language Models (LLMs) in evaluation. This study is a significant contribution to the field, providing insights into the capabilities and performance of LLMs in various evaluative tasks. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + "ZHENG ET AL., 2024" refers to a study conducted by Zheng and colleagues in 2024. This study highlights the effectiveness of Large Language Models (LLMs) in evaluation processes. The research, authored by Zheng et al., provides significant insights into the capabilities and applications of LLMs within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + The entity "CONDITIONS" refers to the different scenarios or variables that are compared in an experiment. Specifically, in the context of the analysis, these conditions include Graph RAG, text summarization, and semantic search RAG. These conditions are used to evaluate and compare various aspects of performance and effectiveness within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + USER QUERIES refer to the inquiries made by users to retrieve information. These queries are answered using different methods and conditions, depending on the context and the specific requirements of the information retrieval process. + 973164fa90bf2b4ee267f4fd795916bf,e4d9b12cf2b4c691c74019eefff4fb39 + + + CONCEPT + Types of entities extracted during the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + + + METRIC + The "CONTEXT WINDOW SIZE" refers to the fixed size of the context window used in various stages of natural language processing and information retrieval tasks. For the final evaluation, the context window size is set to 8k tokens. During the analysis phase, different context window sizes are tested, including 8k, 16k, 32k, and 64k tokens. Additionally, in the graph indexing process, the context window size is set to 600 tokens. This variability in context window sizes highlights the importance of adapting the window size to the specific requirements of different tasks within the domain. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + The process of extracting information, with 1 gleaning for the Podcast dataset and 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + TECHNOLOGY + Natural Language Generation (NLG) is a subfield of artificial intelligence that focuses on generating human-like text from data + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method where a Large Language Model (LLM) is used to compare and evaluate competing outputs + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method for evaluating the performance of Retrieval-Augmented Generation (RAG) systems, focusing on context relevance, faithfulness, and answer relevance + 322e02986c8724eedbcf3ebfa20b989c + + + PUBLICATION + A reference to a study or paper authored by Es and others in 2023 + 322e02986c8724eedbcf3ebfa20b989c + + + TOOL + A Large Language Model (LLM) used to evaluate and compare generated texts based on specific metrics + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + DIRECTNESS is a metric that measures how specifically and clearly an answer addresses a question. It is used to evaluate the straightforwardness of the generated answers. Additionally, it serves as a validity test metric to measure the directness of responses, with naive RAG (Retrieval-Augmented Generation) producing the most direct responses. + 322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + DATA + An example of LLM-generated assessment shown in a table format + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The entity "QUESTION" refers to a specific query used in the evaluation process, particularly as a metric to evaluate the generated responses by asking specific questions. This approach is commonly employed in the domain of Natural Language Processing and Information Retrieval to assess the quality and relevance of responses generated by various models or systems. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + ENTITY + Public figures are individuals who have gained fame or notoriety in various sectors such as entertainment, sports, and digital media. These individuals are well-known in the entertainment industry and are frequently mentioned across various articles. Their prominence in public discourse spans multiple domains, reflecting their influence and recognition in society. + 322e02986c8724eedbcf3ebfa20b989c,718017a4871c909420f84b85b8ba969d + + + DATASET + ENTERTAINMENT ARTICLES is a collection of articles focused on the entertainment industry. This dataset consists of articles related to various aspects of the entertainment sector, providing a comprehensive resource for understanding trends, developments, and key topics within this field. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + DOMAIN + The **ENTERTAINMENT INDUSTRY** is a multifaceted sector that encompasses various forms of entertainment, including movies, music, television, sports, and digital media. This industry is characterized by its diverse range of content and mediums, which collectively contribute to its broad appeal and significant cultural impact. The entertainment industry plays a crucial role in shaping public opinion, trends, and cultural norms through its extensive reach and influence across different platforms and genres. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric indicating the highest level of development or achievement in a particular field + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric indicating results that are comparable to or better than those of others in the same field + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric based on evaluations made by humans + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + Metrics that require a gold standard or reference answers for evaluation + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + An evaluation method that does not require reference answers + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how relevant the generated text is to the given context + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how accurately the generated text reflects the source information + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how relevant the generated answer is to the question + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method involving multiple stages or steps + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The correct or ideal answers used as a benchmark in evaluations + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + "SENSEMAKING QUESTIONS" are a class of questions used to evaluate the performance of Retrieval-Augmented Generation (RAG) systems. These questions are specifically designed to help users understand and make sense of complex information, as well as to validate the understanding and interpretation of data. By employing sensemaking questions, researchers and practitioners can assess how effectively a RAG system can retrieve and generate relevant information, thereby ensuring that the system aids in the comprehension and accurate interpretation of intricate datasets. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method where two items are directly compared against each other + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + TARGET METRICS are specific measures used to evaluate the performance of RAG systems. These metrics are aimed to be achieved or measured in the analysis and are the focus of an evaluation. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A metric used as a baseline or standard for comparison + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures the accuracy and reliability of a method or result + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures the randomness or variability in a process + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The average scores obtained from multiple evaluations + 322e02986c8724eedbcf3ebfa20b989c + + + PERSON + Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Britney Spears is a public figure frequently mentioned in entertainment articles, known for her significant contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his significant contributions to the music industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in film and television + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in music + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in sports + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in digital media and business + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry who are involved in controversies + e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric used to determine the winner in the comparison of generated responses + e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric used to evaluate the quality of LLM-generated responses + e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "FILM" refers to a sector within the entertainment industry that encompasses movies and cinema. This sector includes public figures involved in the movie industry, such as actors, directors, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "TELEVISION" refers to a sector within the entertainment industry that encompasses TV shows and series. This sector includes public figures involved in TV shows, such as actors, hosts, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + MUSIC is a sector within the entertainment industry that encompasses musical performances and recordings. This sector includes public figures involved in the music industry, such as singers, musicians, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "SPORTS" refers to a sector within the entertainment industry that encompasses athletic events and competitions. This sector includes public figures involved in sports, such as athletes, coaches, and sports commentators. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + DIGITAL MEDIA is a sector within the entertainment industry that encompasses online content and social media. This sector includes public figures involved in online platforms, such as influencers, content creators, and digital marketers. These individuals play a significant role in shaping digital landscapes through their engagement with audiences and their ability to leverage various online tools and platforms for content dissemination and marketing purposes. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes stories and themes that shape culture + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes popular movements and styles + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes public conversations and debates + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes formal discussions and communications + e8c8f911135faf3ff35f24107eb3f99c + + + RESPONSE + Answer 1 provides a varied and rich response by covering a wide range of public figures from different sectors of the entertainment industry, including film, television, music, sports, gaming, and digital media. It offers insights into the contributions and influence of these figures, as well as controversies and their impact on public discourse. The answer also cites specific data sources for each mentioned figure, indicating a diverse range of evidence to support the claims. + 718017a4871c909420f84b85b8ba969d + + + RESPONSE + "ANSWER 2" is a generated answer for the example question in the News article dataset. It focuses on a smaller group of public figures, primarily from the music industry and sports, and relies heavily on a single source for data. "ANSWER 2" provides concise explanations for the frequent mentions of specific public figures such as Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + Naïve RAG is a baseline method for retrieval-augmented generation (RAG) that does not use advanced techniques. It is a basic form of RAG with certain drawbacks that advanced RAG systems aim to overcome. Naïve RAG is used to generate answers for questions in the News article dataset and to generate responses that directly list specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d,e4d9b12cf2b4c691c74019eefff4fb39,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19 + + + DATASET + The "NEWS ARTICLE DATASET" is a collection of news articles utilized for various analytical purposes. This dataset is specifically employed for generating responses to questions about public figures in the entertainment industry, making it a valuable resource for both analysis and information retrieval tasks within this domain. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + TOPIC + Controversies are events or issues involving public figures that generate public debate and impact public discourse. + 718017a4871c909420f84b85b8ba969d + + + SECTOR + The gaming sector includes public figures involved in the gaming industry, including gamers, developers, and streamers. + 718017a4871c909420f84b85b8ba969d + + + RESOURCE + Data sources are references or reports used to support claims about public figures and their influence. + 718017a4871c909420f84b85b8ba969d + + + METHOD + Assessments generated by large language models (LLMs) to evaluate the answers produced by different methods + ebf5249c888e07fedce6572a4c03f88c + + + DATASET + An example question used in the News article dataset for analysis + ebf5249c888e07fedce6572a4c03f88c + + + DATA + The datasets used in the analysis, consisting of various text sources + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + A metric used to compare the performance of different conditions in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + CATEGORY + A specific setup or scenario used in the analysis, such as C0, C1, C2, C3, TS, and SS + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + WIN RATE is a measure used to evaluate the success rate of different methods in providing comprehensive and diverse answers. It represents the percentage of times a particular approach or method achieves a win in a given context. Specifically, it quantifies the percentage of times a condition outperformed another in the analysis. This metric is crucial in assessing the effectiveness of various strategies within the domain of Natural Language Processing and Information Retrieval, offering insights into the comparative performance of different techniques. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4 + + + METRIC + The condition that performed the best across all comparisons in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + The expected win rate of a condition when compared to itself, shown as 50% for reference + 4c855404ee3d3c94aa2136f1513c666f + + + METHOD + The use of large language models (LLMs) at the time of querying, evaluated in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + METHOD + The "FINAL EVALUATION" is the last stage of the analysis where the best performing context window size was used. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + + + PROCESS + The process that resulted in the creation of graphs for the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + + + STRUCTURE + A data structure consisting of nodes and edges, used to represent the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + + + METHOD + Approaches that consistently outperformed the naive RAG in comprehensiveness and diversity metrics + 36db32c37e1987e2c5863898ad882190 + + + METRIC + The number of context units, such as community summaries or text chunks, used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + + + METRIC + The term "TOKENS" refers to the number of individual words used in the analysis. The evaluation typically focuses on corpora in the region of 1 million tokens. This metric is crucial for understanding the scope and scale of the text data being analyzed, particularly in the fields of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,92e93fc6449756c0a60200636b297f65 + METRIC + + + METRIC + The percentage of the maximum token count used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + + + METHOD + MAP-REDUCE SUMMARIZATION is a method for summarizing source texts using a map-reduce approach. This summarization technique is notably resource-intensive, necessitating the highest number of context tokens compared to other methods. The map-reduce framework, originally popularized for its efficiency in processing large-scale data, is adapted here to handle the complexities of text summarization, ensuring comprehensive and accurate extraction of key information from extensive source texts. + 36db32c37e1987e2c5863898ad882190,e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Summaries at the root level of the community hierarchy, requiring dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + + + DATASET + SOURCE TEXTS are the original texts from which summaries or analyses are derived. These texts serve as the foundational material used for comparison with community summaries in the analysis. + 6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39 + + + REFERENCE + A reference to a paper by Ram et al. in 2023 discussing RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + REFERENCE + "GAO ET AL., 2023" is a paper published in 2023 by Gao et al. that delves into advanced Retrieval-Augmented Generation (RAG) techniques, specifically where the index is a knowledge graph. The publication also touches upon naive RAG approaches, providing a comprehensive examination of both advanced and basic methodologies within the domain of Natural Language Processing and Information Retrieval. + 6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + + + CATEGORY + Intermediate-level summaries are a type of community summary used in the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + CATEGORY + Root-level summaries are a type of community summary used in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + METRIC + Answer comprehensiveness is a measure used to evaluate the completeness of answers provided by different methods + 6f33a085ff3304e5994f7fbb86c881a4 + + + TECHNOLOGY + Ad-hoc LLM use refers to the spontaneous use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + + + TECHNOLOGY + Element extraction prompts are used to extract specific details in the Graph RAG index + 6f33a085ff3304e5994f7fbb86c881a4 + + + CONCEPT, TECHNOLOGY + A mathematical space in which text chunks and queries are embedded to represent similar semantics + f35de4d9fb65f1d5a392064b20545c19 + + + CONCEPT, DATA + Search inputs that are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A type of RAG system that includes patterns for iterative and dynamic cycles of interleaved retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, CONCEPT + A concept related to generation-augmented retrieval that facilitates future generation cycles + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method that facilitates future generation cycles by using self-memory + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A strategy for iterative retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A federated strategy for retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method that combines multiple concepts for summarizing multiple documents + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method for answering questions that require multiple steps or "hops" to gather information + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + An approach that involves generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A method for generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A method for answering multiple interpretations of ambiguous questions by generating a hierarchical structure + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A process that involves using LLMs to create knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A process that involves using LLMs to complete existing knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + Graphs that represent causal relationships, which can be extracted using LLMs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + REFERENCE, PUBLICATION + A reference to a publication by Cheng et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Mao et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Shao et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Wang et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Su et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Feng et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Trivedi et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Khattab et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Sarthi et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Kim et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + "TRAJANOSKA ET AL., 2023" refers to a paper by Trajanoska et al. published in 2023, which focuses on using Large Language Models (LLMs) for knowledge graph creation. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting innovative methodologies for leveraging advanced language models to construct and enhance knowledge graphs. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + "Yao et al., 2023" refers to a paper published by Yao and colleagues in 2023. The study focuses on the application of large language models (LLMs) for the task of knowledge graph completion. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting the potential of advanced LLMs to enhance the accuracy and efficiency of knowledge graph completion processes. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + TECHNOLOGY, METHOD + A system that combines multiple concepts for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + Strategies used before the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Strategies used during the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Strategies used after the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A pattern in Modular RAG systems for iterative and dynamic cycles of retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Cycles of generation that are facilitated by self-memory in Graph RAG + f35de4d9fb65f1d5a392064b20545c19 + + + PUBLICATION + A paper by Ban et al. published in 2023, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + PUBLICATION + A paper by Zhang et al. published in 2024, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where the index is a knowledge graph, developed by Baek et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Baek et al. published in 2023, focusing on the KAPING method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where subsets of the graph structure are the objects of enquiry, developed by He et al. in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by He et al. published in 2024, focusing on the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where derived graph metrics are the objects of enquiry, developed by Zhang in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Zhang published in 2023, focusing on the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, developed by Kang et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Kang et al. published in 2023, focusing on the SURGE method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where retrieved event-plot subgraphs are serialized using narrative templates, developed by Ranade and Joshi in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Ranade and Joshi published in 2023, focusing on the FABULA method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + PUBLICATION + A paper by Wang et al. published in 2023, focusing on a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + ORGANIZATION + LangChain is an organization that developed Langchain graphs and supports a variety of graph databases. + 71f6daf11e64e5273a3847d46bf228e1,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + + + ORGANIZATION + LlamaIndex is an organization that developed the LlamaIndex Knowledge Graph Index and supports a variety of graph databases. + 6cd82819982879bd164547d2773ba5c7,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + + + TECHNOLOGY + Neo4J is both a graph database format supported by various Retrieval-Augmented Generation (RAG) applications and an organization that developed Project NaLLM. The graph database format of Neo4J is widely recognized for its efficiency in handling complex relationships and structures, making it a valuable tool in the field of Natural Language Processing and Information Retrieval. As an organization, Neo4J has contributed significantly to the advancement of these domains through innovative projects like NaLLM, which further underscores its pivotal role in the community. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + + + METHOD + A method that can create and reason over knowledge graphs in Neo4J format, developed by Neo4J in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + TECHNOLOGY + NebulaGraph is both a graph database format and an organization that has made significant contributions to the field of graph databases and retrieval-augmented generation (RAG) applications. As a graph database format, NebulaGraph is supported by various RAG applications, facilitating the efficient handling and querying of complex graph data structures. Additionally, NebulaGraph, as an organization, has pioneered the industry-first graph RAG, which integrates retrieval-augmented generation with large language models (LLMs) based on knowledge graphs. This innovation underscores NebulaGraph's role in advancing the capabilities of knowledge graph-based applications and enhancing the performance of LLMs in generating contextually relevant information. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + + + METHOD + A method that can create and reason over knowledge graphs in NebulaGraph format, developed by NebulaGraph in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + METHOD + A method for comparing fabrication rates, developed by Manakul et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + "MANAKUL ET AL., 2023" refers to a paper by Manakul et al. published in 2023, which focuses on the SelfCheckGPT method. This work by Manakul and colleagues is centered around the development and application of SelfCheckGPT, a technique likely aimed at enhancing the performance and reliability of GPT models. The paper contributes to the field of Natural Language Processing and Information Retrieval by addressing specific challenges and proposing innovative solutions through the SelfCheckGPT method. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + PUBLICATION + + + STAKEHOLDER + END USERS are individuals who are the final users of the system or analysis. They play a crucial role in validating sensemaking questions and target metrics, ensuring that the system or analysis meets the intended objectives and provides meaningful insights. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + STAKEHOLDER + + + CONCEPT + Considerations and compromises involved in building a graph index + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METRIC + The effectiveness of RAG systems, which varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + METRIC + + + CONCEPT + Various forms of data used in RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METRIC + The scale of datasets used in RAG systems, which affects performance + 92e93fc6449756c0a60200636b297f65 + METRIC + + + PROCESS + The process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + PROCESS + + + DATASET + Collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + DATASET + + + CONCEPT + Different categories of questions used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METHOD + SelfCheckGPT is an approach used to compare fabrication rates in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method for global summarization of source texts that does not use a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + RESOURCE + The amount of computational resources allocated for a task + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The expected number of queries over the lifetime of a dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Annotations that provide detailed information about the text + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method that uses embeddings to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + RAG schemes that combine embedding-based matching with other approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Mechanisms used in map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A hierarchical organization of communities + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) for global text summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The cost associated with the number of tokens used in a text generation task + e4d9b12cf2b4c691c74019eefff4fb39 + + + TECHNOLOGY + An implementation of Graph RAG approaches using the Python programming language + e4d9b12cf2b4c691c74019eefff4fb39 + + + PERSON + A person who contributed to the work mentioned in the acknowledgements + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The rates at which fabrications occur in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The expected number of queries over the lifetime of a specific dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The benefits or value obtained from using a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Different methods related to retrieval-augmented generation that utilize graph structures + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Graph RAG approaches that operate in a more localized manner + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Annotations made on the graph to provide additional information + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Reports generated from community summaries + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + An operation that aggregates information across multiple levels of a hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A mechanism that allows for exploring detailed information by following higher-level summaries + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + The trail of information that guides users to more detailed data + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + The top-level communities in a hierarchical structure + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A graph index organized around entities + e4d9b12cf2b4c691c74019eefff4fb39 + + + TECHNOLOGY + A publicly available implementation of a technology + e4d9b12cf2b4c691c74019eefff4fb39 + + + PERSON + Alonso Guevara Fernández is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Amber Hoak is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Andrés Morales Esquivel is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Ben Cutler is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Billie Rinaldi is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Chris Sanchez is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Chris Trevino is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Christine Caggiano is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + David Tittsworth is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Dayenne de Souza is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Douglas Orbaker is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Ed Clark is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Gabriel Nieves-Ponce is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Gaudy Blanco Meneses is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Kate Lytvynets is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Katy Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Mónica Carvajal is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Nathan Evans is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Richard Ortega is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Rodrigo Racanicci is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Sarah Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Shane Solomon is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + A technical report on GPT-4 published as an arXiv preprint + 086021a89900a39bcb62036981737bfa + + + METHOD + A method for zero-shot knowledge graph question answering described in an arXiv preprint + 086021a89900a39bcb62036981737bfa + + + METHOD + A method for harnessing large language models for advanced causal discovery from data + 086021a89900a39bcb62036981737bfa + + + METHOD + A method incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Achiam is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Adler is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Agarwal is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + L. Ahmad is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + I. Akkaya is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + F. L. Aleman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + D. Almeida is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Altenschmidt is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Altman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Anadkat is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + R. Anil is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Borgeaud is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + Y. Wu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J.-B. Alayrac is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Yu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + R. Soricut is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Schalkwyk is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + A. M. Dai is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + A. Hauth is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Baek is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + A. F. Aji is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + A. Saffari is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + T. Ban is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + L. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + X. Wang is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + H. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + T. Baumel is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + M. Eyal is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + M. Elhadad is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + Baumel, T. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Eyal, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Elhadad, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Blondel, V. D. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Guillaume, J.-L. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Lambiotte, R. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Lefebvre, E. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The journal where the paper "Fast unfolding of communities in large networks" was published + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Brown, T. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Mann, B. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Ryder, N. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Subbiah, M. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Kaplan, J. D. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dhariwal, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Neelakantan, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Shyam, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Sastry, G. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Askell, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + "ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS" is a prominent conference where significant papers in the field of Natural Language Processing and Information Retrieval are presented. Notable papers presented at this conference include "Language models are few-shot learners" and "Retrieval-augmented generation for knowledge-intensive NLP tasks." Additionally, it is also the journal where the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" was published. This conference and journal serve as key platforms for disseminating cutting-edge research in neural information processing systems. + 58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,b1bbda43309e8e0e2175ea034aa88e13 + + + PERSON + Cheng, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Luo, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Chen, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Liu, L. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Zhao, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory"Zhao, D. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + PERSON + + + PERSON + Yan, R. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dang, H. T. is an author of the paper "Duc 2005: Evaluation of question-focused summarization systems" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The conference where the paper "Duc 2005: Evaluation of question-focused summarization systems" was presented + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Es, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + James, J. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Espinosa-Anke, L. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Schockaert, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Feng, Z. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Feng, X. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Yang, M. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Qin, B. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Fortunato, S. is an author of the paper "Community detection in graphs" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The journal where the paper "Community detection in graphs" was published + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Gao, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Xiong, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models. The paper provides a comprehensive survey of the methodologies and applications of retrieval-augmented generation, highlighting its significance in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Gao, X. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Jia, K. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant area of research within the domains of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Pan, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Bi, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dai, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance and capabilities of large language models, a significant area of research within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Sun, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Wang, H. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Goodwin, T. R. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Savery, M. E. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Demner-Fushman, D. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + CONFERENCE + COLING (International Conference on Computational Linguistics) is the conference where the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" was presented + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + He, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Tian, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Sun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Chawla, N. V. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Laurent, T. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + LeCun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Bresson, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Hooi, B. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jacomy, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Venturini, T. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Heymann, S. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Bastian, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PUBLICATION + PLOS ONE is the journal where the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" was published + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jin, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Yu, Z. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jiao, P. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Pan, S. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + He, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Wu, J. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Philip, S. Y. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Zhang, W. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PUBLICATION + IEEE Transactions on Knowledge and Data Engineering is the journal where the paper "A survey of community detection approaches: From statistical modeling to deep learning" was published + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Kang, M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Kwak, J. M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Baek, J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Hwang, S. J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Khattab, O. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Santhanam, K. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Li, X. L. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hall, D. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text, indicating its relevance within the domain of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Liang, P. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Liang, P. contributed to the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP," which explores the integration of retrieval and language models to enhance knowledge-intensive tasks in NLP. Additionally, Liang, P. authored the paper "Lost in the middle: How language models use long contexts," which investigates the utilization of extended contexts by language models. These contributions highlight Liang, P.'s significant role in advancing the understanding and application of language models in complex NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Potts, C. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Zaharia, M. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kim, G. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kim, S. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Jeon, B. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Park, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kang, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Klein, G. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Moon, B. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hoffman, R. R. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The journal where the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" were published + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Koesten, L. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Gregory, K. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Groth, P. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Simperl, E. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The journal where the paper "Talking datasets–understanding data sensemaking behaviours" was published + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kuratov, Y. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Bulatov, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Anokhin, P. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Sorokin, D. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Sorokin, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Burtsev, M. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + TECHNOLOGY + Langchain graphs is a technology developed by LangChain + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Laskar, M. T. R. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" and also contributed to the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models." Both works highlight Laskar's expertise in leveraging transformer models and transfer learning techniques to enhance the performance of query-focused abstractive text summarization, demonstrating a significant contribution to the field of Natural Language Processing and Information Retrieval. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hoque, E. is an author of two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning to improve the effectiveness of transformer models in query-focused abstractive summarization tasks. Both works contribute to advancing the understanding and application of transformer models in specialized summarization contexts. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Huang, J. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The conference where the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" was presented + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + arXiv preprint refers to a preprint of a paper that is available on the arXiv repository + 71f6daf11e64e5273a3847d46bf228e1 + + + EVENT + The 33rd Canadian Conference on Artificial Intelligence, held in Ottawa, ON, Canada, from May 13–15, 2020 + 6cd82819982879bd164547d2773ba5c7 + + + EVENT + The 2020 edition of the Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + PUBLISHER + Springer is the publisher of the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Huang, J. X. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + PUBLICATION + The journal where the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" was published + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lewis, P. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Perez, E. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Piktus, A. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Petroni, F. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks"Petroni, F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + PERSON + + + PERSON + Karpukhin, V. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Goyal, N. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Küttler, H. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lewis, M. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Yih, W.-T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Rocktäschel, T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, N. F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lin, K. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Hewitt, J. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Paranjape, A. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Bevilacqua, M. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, Y. is an author of the paper "Hierarchical transformers for multi-document summarization" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lapata, M. is an author known for significant contributions to the field of Natural Language Processing and Information Retrieval. Notably, Lapata, M. has authored the paper "Hierarchical transformers for multi-document summarization," which explores advanced techniques in summarizing information from multiple documents using hierarchical transformer models. Additionally, Lapata, M. has contributed to the paper "Text summarization with latent queries," which delves into innovative methods for summarizing text by leveraging latent query representations. These works highlight Lapata, M.'s expertise and active research in the domain of text summarization, showcasing a focus on developing sophisticated models and methodologies to enhance the efficiency and accuracy of summarization tasks. + 6cd82819982879bd164547d2773ba5c7,fc4b27d64f055b7fc30176ba110dd02e + + + TECHNOLOGY + LlamaIndex Knowledge Graph Index is a technology developed by LlamaIndex + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Manakul, P. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liusie, A. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Gales, M. J. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Mao, Y. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + He, P. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, X. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Shen, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Shen, Y.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Gao, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Han, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Chen, W. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Chen, W.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Martin, S. is an author of the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing a comprehensive, open-source solution for the layout of large graphs, which is a critical task in the visualization and analysis of complex networks. The toolbox aims to facilitate the understanding and interpretation of large-scale graph data, making it a valuable resource for researchers and practitioners in fields such as computational linguistics, information retrieval, and data science. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Brown, W. M. is an author of the paper "Openord: An open-source toolbox for large graph layout." + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + KLAVANS, R. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Boyack, K. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on the development and application of Openord, a comprehensive open-source toolbox designed for the layout of large graphs. The paper likely discusses the methodologies, algorithms, and practical implementations of the toolbox, contributing to the fields of graph theory and data visualization. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + EVENT + The conference where the paper "Openord: An open-source toolbox for large graph layout" was presented + 833e7d67dcd30790b26b71c9b5306f6b + EVENT + + + TECHNOLOGY + GPT-4 is a large language model used in Microsoft's study on scientific discovery + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + + + TECHNOLOGY + Project NaLLM is a project developed by Neo4J + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + + + PERSON + Newman, M. E. is the author of the paper "Modularity and community structure in networks" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PUBLICATION + The journal where the paper "Modularity and community structure in networks" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + + + PERSON + Ram, O. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Levine, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Dalmedigos, I. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Muhlgay, D. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shashua, A. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Leyton-Brown, K. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shoham, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PUBLICATION + The journal where the paper "In-context retrieval-augmented language models" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + + + PERSON + Ranade, P. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Joshi, A. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Sarthi, P. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Abdullah, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Tuli, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Khanna, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Goldie, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Manning, C. D. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" and the paper "Raptor: Recursive abstractive processing for tree-organized retrieval". These contributions highlight Manning's involvement in advancing the fields of Natural Language Processing and Information Retrieval, particularly in the areas of multi-hop question answering and recursive abstractive processing. + 833e7d67dcd30790b26b71c9b5306f6b,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Scott, K. is associated with "Behind the Tech" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shao, Z. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Gong, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Huang, M. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + + + PERSON + Duan, N. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + + + PERSON + Su, D. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Xu, Y. is an author of multiple academic papers in the field of Natural Language Processing and Information Retrieval. Notably, Xu, Y. contributed to the paper titled "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," which addresses the management of scholarly information related to COVID-19 through advanced question answering and summarization techniques. Additionally, Xu, Y. co-authored the paper "Text summarization with latent queries," which explores innovative methods for text summarization by leveraging latent queries. These contributions highlight Xu, Y.'s expertise and active involvement in developing sophisticated systems for information retrieval and summarization. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yu, T. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Siddique, F. B. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Barezi, E. J. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Fung, P. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Tang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Yang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Touvron, H. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Martin, L. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Stone, K. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Albert, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Almahairi, A. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Babaei, Y. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bashlykov, N. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Batra, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bhargava, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bhosale, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Traag, V. A. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Waltman, L. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Van Eck, N. J. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PUBLICATION + Scientific Reports is the journal where the paper "From Louvain to Leiden: guaranteeing well-connected communities" was published + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trajanoska, M. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Stojanov, R. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trajanov, D. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trivedi, H. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Balasubramanian, N. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Khot, T. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Sabharwal, A. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Wang, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Liang, Y. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Meng, F. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Sun, Z. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Shi, H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Li, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through their work on evaluating language models. Specifically, Li, Z. has co-authored the paper titled "Is ChatGPT a Good NLG Evaluator? A Preliminary Study," which explores the effectiveness of ChatGPT as a natural language generation evaluator. Additionally, Li, Z. has co-authored another paper, "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which examines the performance of large language models in evaluative roles using specific benchmarking tools. These contributions highlight Li, Z.'s active involvement in advancing the understanding and assessment of language models within the academic community. + 8d87efac8c50cf20cdf26bf61e5e2035,b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Xu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Qu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhou, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wang, S. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" and also contributed to the paper "Is chatgpt a good nlg evaluator? a preliminary study." These works indicate Wang, S.'s involvement in cutting-edge research within the fields of federated search, retrieval augmented generation, and natural language generation evaluation, showcasing a focus on both the technical and evaluative aspects of Natural Language Processing and Information Retrieval. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Khramtsova is an author mentioned in the text + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Khramtsova, E. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhuang, S. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through multiple academic papers. Notably, Zhuang, S. co-authored the paper titled "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," which explores the evaluation of federated search systems within the framework of retrieval-augmented generation. Additionally, Zhuang, S. co-authored another significant paper, "Judging llm-as-a-judge with mt-bench and chatbot arena," which delves into the assessment of large language models (LLMs) using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Zhuang, S.'s active involvement in advancing research in federated search and the evaluation of LLMs. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zuccon, G. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wang, Y. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Lipka, N. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Rossi, R. A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Siu, A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhang, R. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Derr, T. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yang, Z. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Qi, P. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhang, S. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Bengio, Y. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Cohen, W. W. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Salakhutdinov, R. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + CONFERENCE + The conference where the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" was presented + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yao, J.-g. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wan, X. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Xiao, J. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PUBLICATION + The journal where the paper "Recent advances in document summarization" was published + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yao, L. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models"Yao, L. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Peng, J. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Mao, C. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Luo, Y. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhang, J. is an author of the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhang, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Gan, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Wang, C. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zheng, L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zheng, L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Zheng, L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools like MT-Bench and Chatbot Arena. These contributions highlight Zheng, L.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR domains. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Chiang, W.-L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Chiang, W.-L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Chiang, W.-L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Chiang, W.-L.'s active involvement in advancing the understanding and capabilities of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Sheng, Y. is an author known for contributing to the field of Natural Language Processing and Information Retrieval. Notably, Sheng, Y. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Sheng, Y. has contributed to the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Sheng, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic and technical community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Wu, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Wu, Z. co-authored the paper titled "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Wu, Z. is also credited with co-authoring the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Wu, Z.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhuang, Y. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zhuang, Y. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness of knowledge graphs. Additionally, Zhuang, Y. has also authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Zhuang, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the domain. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Lin, Z. is an author of the paper "Exploring large language models for knowledge graph completion" and also contributed to the paper "Judging LLM-as-a-judge with MT-Bench and Chatbot Arena." These works indicate Lin, Z.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the application of large language models for tasks such as knowledge graph completion and the evaluation of language models in judgment tasks. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Li, D. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant research. Notably, Li, D. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Li, D. has also co-authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Li, D.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Xing, E. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Xing, E. contributed to the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Xing, E.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + TECHNOLOGY + Chatbot Arena is a platform or tool used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Darren Edge and Ha Trinh co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Steven Truitt and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Steven Truitt is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Jonathan Larson is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 4.0 + RAG (Retrieval-Augmented Generation) and LLM (Large Language Models) are closely intertwined in the domain of Natural Language Processing and Information Retrieval. RAG is employed to enhance the capabilities of LLMs by enabling them to retrieve pertinent information from external knowledge sources. This symbiotic relationship allows LLMs to generate and assess text more effectively. Specifically, RAG leverages the power of LLMs to access and utilize relevant data, thereby augmenting the overall performance and accuracy of text generation tasks. + e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 7.0 + Graph RAG is a specific implementation of RAG that combines the strengths of RAG with graph-based text indexing. This method leverages the natural modularity of graphs to partition data, facilitating global summarization. As a specialized approach within the RAG framework, Graph RAG enhances the capabilities of RAG by integrating graph structures to improve the efficiency and effectiveness of text data processing and summarization. + 21e52bc06a82796b1f4bcd73edda1f2a,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Query-Focused Summarization is a task that RAG fails to address effectively + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + RAG retrieves relevant information from an external knowledge source + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Naive RAG is a specific implementation of RAG + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Ram et al., 2023 discusses RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Naïve RAG is a basic form of RAG + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Modular RAG is an advanced form of RAG + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used in various RAG tasks such as knowledge graph creation and completion + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Trajanoska et al. discusses using LLMs for knowledge graph creation, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Yao et al. discusses using LLMs for knowledge graph completion, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Ban et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Zhang et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Gao et al. discusses advanced RAG where the index is a knowledge graph + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + KAPING is a method where the index is a knowledge graph, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + G-Retriever is a method where subsets of the graph structure are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Graph-ToolFormer is a method where derived graph metrics are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + SURGE is a method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + FABULA is a method where retrieved event-plot subgraphs are serialized using narrative templates, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Wang et al. discusses a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Sensemaking questions are used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The evaluation of RAG systems focuses on corpora in the region of 1 million tokens + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Trade-offs are considerations involved in building a graph index for RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + A graph index is a data structure used in RAG systems to organize and retrieve information + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Performance of RAG systems varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Different data types are used in RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Dataset sizes affect the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Evaluation is the process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Corpora are collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Different question types are used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Target metrics are specific measures used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 4.0 + Graph RAG utilizes Large Language Models (LLMs) to construct a graph-based text index, enabling the generation of summaries and the answering of queries. In this approach, LLMs play a crucial role in analyzing and generating text based on the information retrieved through the graph structure. Additionally, LLMs leverage the Graph RAG framework to provide comprehensive overviews of public figures in the entertainment industry. This integration of LLMs within Graph RAG enhances the system's ability to process and synthesize complex text data, making it a powerful tool for information retrieval and natural language processing tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Text chunks are processed using LLM to extract elements of a graph index + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM is used to extract elements of a graph index from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + LLM (Large Language Model) and Few-Shot Examples are closely related in the context of Natural Language Processing and Information Retrieval. Few-shot examples are provided to the LLM for in-context learning, which helps tailor the extraction prompt. This technique is particularly useful for improving the performance of the LLM in specialized domains. By leveraging a small number of examples, the LLM can better understand and adapt to specific tasks, thereby enhancing its overall effectiveness in extracting and processing information within those specialized areas. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM extracts named entities from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Kuratov et al. (2024) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Liu et al. (2023) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM prompts are instructions given to the LLM for extracting elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Recall degradation occurs with longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + The extraction process involves using LLM to identify and extract elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Default prompt is the standard set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Secondary extraction prompt is an additional set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + The LLM uses covariate prompts to extract additional attributes associated with detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM uses multiple rounds of gleanings to ensure no entities are missed + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Logit bias is used to force a yes/no decision from the LLM during entity extraction + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM extracts element instances from source texts + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM detects and summarizes communities of entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + LLM generates intermediate answers and scores for each chunk + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + LLM generates a helpfulness score for each answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + LLM is used to generate questions for evaluating the Podcast Transcripts dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + LLM is used to generate questions for evaluating the News Articles dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + LLM uses Naive RAG to list public figures mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + LLM-generated responses are evaluated using assessment metrics + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + LLM-generated responses are evaluated using specific questions + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Ad-hoc LLM use involves the use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + LLMs are used for knowledge graph creation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph completion + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for the extraction of causal graphs + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph creation as per Trajanoska et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph completion as per Yao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for the extraction of causal graphs as per Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is proposed as a method to combine the strengths of RAG and QFS + e8d83e6e7a7c0f57b218cef24976b745 + + + 7.0 + Graph RAG is an approach that utilizes community summaries in various capacities to enhance its functionality. Specifically, community summaries are generated within the Graph RAG framework and are employed to generate partial responses. These summaries are also used to compare against source texts, serving as a form of self-memory for the system. By leveraging community summaries, Graph RAG aims to improve the comprehensiveness and diversity of answers. Additionally, Graph RAG incorporates summaries of root-level communities within an entity-based graph index, further enriching its analytical capabilities. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is designed to handle global sensemaking questions over large datasets + e8d83e6e7a7c0f57b218cef24976b745 + + + 2.0 + Graph RAG is implemented in Python. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The open-source implementation of Graph RAG will be available at this URL + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Graph RAG uses an entity knowledge graph to index text + e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG is an approach that is evaluated for its comprehensiveness, a target quality used to assess its effectiveness. The method aims to improve the comprehensiveness of generated answers, ensuring that the information provided is thorough and complete. This evaluation metric is crucial in determining the success of Graph RAG in producing detailed and accurate responses. + 21e52bc06a82796b1f4bcd73edda1f2a,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG is an approach in the domain of Natural Language Processing and Information Retrieval that focuses on improving the diversity of generated answers. Diversity, in this context, is a target quality used to evaluate the performance of the Graph RAG approach. By enhancing the diversity of responses, Graph RAG aims to provide a broader range of answers, thereby improving the overall effectiveness and robustness of the system. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG uses a knowledge graph for global summarization + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Community detection algorithms are used in the Graph RAG approach to partition graphs + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Podcast transcripts are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + News articles are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 2.0 + Graph RAG is evaluated using the target quality of Empowerment. Empowerment is specifically utilized to assess Graph RAG's capability in aiding users to achieve an informed understanding. This evaluation metric underscores the importance of user comprehension and the effectiveness of the Graph RAG approach in facilitating informed decision-making processes. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Graph RAG is compared to naive RAG in the evaluation. In this comparison, Graph RAG outperformed naive RAG on comprehensiveness and diversity. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Graph RAG is compared to global map-reduce summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Query-focused summarization is a method used in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Activity-centered sensemaking questions are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 2.0 + The "Graph RAG" approach is evaluated in terms of its performance by considering "Token Costs." Token costs are measured to assess the efficiency of the Graph RAG method, indicating that the computational expense associated with processing tokens is a critical factor in determining the overall effectiveness of this approach. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Data flow describes the high-level process of the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 3.0 + Design parameters are key settings in the Graph RAG approach and significantly influence the Graph RAG approach and pipeline. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Graph RAG uses global summarization to summarize information from a large dataset + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG aims to answer specific queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG uses a corpus for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Activity-centered sensemaking is used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Real-world datasets are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG is compared to source text summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Low-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Intermediate-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + High-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + The Graph RAG approach involves a specific pipeline for processing and summarizing text + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Techniques are specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Implementation details are specific configurations used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Graph RAG is a specific implementation of RAG systems + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + Graph RAG is a system that utilizes root-level community summaries, denoted as C0, to answer user queries. C0 represents these root-level community summaries within the Graph RAG analysis, serving as a foundational element in the system's ability to map out relationships and understand the structural dynamics within specialized communities. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses high-level community summaries (C1) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses intermediate-level community summaries (C2) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 2.0 + Graph RAG utilizes low-level community summaries, represented by C3, to answer user queries. C3 plays a crucial role in the Graph RAG analysis by providing detailed summaries of community structures, which are essential for effectively addressing user inquiries. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + 2.0 + Graph RAG is a key entity in the analysis, serving both as a condition being compared and as a tool for comparing multiple conditions. This dual role highlights its significance in the study, where it functions not only as a subject of comparison but also as a methodological framework for evaluating other conditions. The analysis likely involves a detailed examination of various conditions, with Graph RAG playing a central role in facilitating these comparisons. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses different levels of graph communities to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The Graph RAG mechanism uses an LLM evaluator for head-to-head comparison + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Graph RAG is a multi-stage mechanism + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Graph RAG mentions Taylor Swift as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Travis Kelce as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Britney Spears as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Justin Timberlake as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG is determined to be the winner based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Answer 1 was generated using the Graph RAG method, which provides a comprehensive and structured overview of public figures across various sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Graph RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Graph RAG is compared with source texts for answer comprehensiveness and diversity + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + TS represents source text summarization in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Root-level summaries are used in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Answer comprehensiveness is used to evaluate the performance of Graph RAG + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Win rate is used to measure the success rate of Graph RAG in providing comprehensive and diverse answers + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Element extraction prompts are used in Graph RAG to retain specific details in the index + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Graph RAG incorporates the concept of self-memory + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of iterative retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of federated retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts used in multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts used in multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG uses a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of a tree of clarifications + f35de4d9fb65f1d5a392064b20545c19 + + + 3.0 + Graph RAG utilizes a self-generated graph index. This self-generated graph index is a crucial component of Graph RAG, enabling it to efficiently manage and retrieve information within its graph-based framework. The use of a self-generated graph index suggests that Graph RAG has an inherent capability to construct and maintain its indexing structure, which likely enhances its performance and adaptability in handling complex data relationships. + e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Gao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Cheng et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Mao et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Shao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Wang et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Su et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Feng et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Trivedi et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Khattab et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Sarthi et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Kim et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG generates community answers in parallel + f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is compared to a graph-free approach for global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG is compared to map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses rich text annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses a hierarchical community structure + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can operate using embedding-based matching + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can be part of hybrid RAG schemes + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can employ map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can extend operations across the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Alonso contributed to the work on Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG includes local graph RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses an entity-based graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + 2.0 + NebulaGraph launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Community summaries are used to generate partial responses + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Community summaries are created from graph communities + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Community answers are created from community summaries. These answers are generated by synthesizing information from the summaries provided by the community, ensuring that the responses are comprehensive and reflective of the collective knowledge and insights within the community. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Domain-tailored summarization is used to create community summaries + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Community descriptions are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Partial answers are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Community summaries are created for each level in the hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Community summaries are useful for understanding the global structure and semantics of the dataset + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Community summaries are used to answer global queries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are generated from root communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are generated from sub-communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are added to the LLM context window until the token limit is reached + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Global answers are generated from community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The level of summary detail affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The scope of information affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are used for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Community summaries are divided into chunks of pre-specified token size + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Summary detail and scope affect the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are divided into chunks + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Community summaries are prepared to answer user queries + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Intermediate answers are generated from community summaries + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Community summaries are part of the graph community hierarchy + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Community summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Summaries of root-level communities are used in Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Global sensemaking questions are evaluated over datasets in the 1 million token range + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Global sensemaking questions are directed at an entire text corpus + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The Python-based implementation of Graph RAG approaches will be available at this URL + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Query-focused summarization is used to produce the global answer + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Map-reduce is used for query-focused summarization of an entire corpus + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Query-focused summarization is used for answering global queries + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + An entity knowledge graph is derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + + + 2.0 + In the domain of Natural Language Processing and Information Retrieval, "SOURCE DOCUMENTS" and "TEXT CHUNKS" are closely related entities. Text chunks are extracted from source documents, specifically for processing in the Graph RAG (Retrieval-Augmented Generation) approach. This method involves breaking down the source documents into smaller, manageable pieces, or text chunks, which can then be effectively utilized in various computational processes, enhancing the efficiency and accuracy of information retrieval and generation tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Intermediate-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Low-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Document corpus consists of source documents being processed + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Partial responses are summarized to generate a final response + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The LLM evaluator assesses answers based on the comprehensiveness metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Naive RAG is evaluated for comprehensiveness + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Comprehensiveness is a metric used to determine the decision + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) was universally better for comprehensiveness + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The final evaluation prioritized comprehensiveness in answers + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Global approaches achieved higher comprehensiveness win rates + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The LLM evaluator assesses answers based on the diversity metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on diversity + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The final evaluation prioritized diversity in answers + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Global approaches achieved higher diversity win rates + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Human endeavors rely on sensemaking to understand and reason about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Human endeavors rely on analyzing document collections for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + LLMs are used to automate sensemaking in complex domains + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Microsoft uses LLMs for automating sensemaking in scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Ranade uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Joshi uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + LLM prompts are used to tailor the responses of large language models + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Ranade and Joshi discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + GPT is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Llama is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Gemini is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Kuratov et al., 2024, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Liu et al., 2023, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Sensemaking is applied in the domain of scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Sensemaking is applied in the domain of intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Klein defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Klein et al. defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Element instances are extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Entity references are extracted from text chunks during processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Element instances are converted into element summaries by the LLM (Large Language Model). Element summaries are created from element instances, indicating a transformation process facilitated by the LLM. This process involves the LLM taking detailed element instances and generating concise element summaries, which encapsulate the essential information from the instances. + 2c6ed90897310eea2f28e33fff1c32b0,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Covariates are additional attributes associated with extracted element instances + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Domain-tailored summarization is used to create element summaries + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Element summaries include descriptions of entity nodes + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries include descriptions of relationship edges + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries include descriptions of claim covariates + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries are used to understand the structure and semantics of graph communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Element summaries include descriptions of nodes + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Element summaries include descriptions of edges + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Element summaries include descriptions of covariates + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Sub-community summaries are used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Community detection is a technique used to identify graph communities. Graph communities are groups of nodes within a graph that are more densely connected to each other than to the rest of the graph. This process of identifying such communities is crucial for understanding the structural dynamics and relationships within complex networks, particularly in the domain of Natural Language Processing and Information Retrieval. By leveraging community detection algorithms, researchers can uncover hidden patterns and insights within large datasets, facilitating more effective data analysis and interpretation. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Global answer is created from community answers + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Global answers are generated in response to user queries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Global answer is generated by sorting intermediate answers based on helpfulness scores + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Intermediate answers are combined to form the global answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + The final context window is used to generate the global answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Graph RAG pipeline operates at indexing time + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Graph RAG pipeline operates at query time + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Nodes are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Edges are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Covariates are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Leiden method is used in the graph RAG pipeline for community detection + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Graph RAG pipeline uses the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + The Podcast dataset graph consists of 8564 nodes + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The News dataset graph consists of 15754 nodes + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The Podcast dataset graph consists of 20691 edges + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The News dataset graph consists of 19520 edges + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Traag contributed to the development of the Leiden method + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Traag et al. are the authors of the Leiden algorithm and developed the Leiden method. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Leiden is a specific type of community detection algorithm used in various analytical pipelines. It is designed to identify and map out the structural dynamics within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. The algorithm is known for its efficiency and accuracy in detecting community structures, making it a valuable tool for researchers and practitioners in the field. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Leiden is known for its ability to recover hierarchical community structures efficiently + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The Leiden algorithm is used to detect graph communities in the MultiHop-RAG + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Figure 3 shows graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Lewis contributed to the development of the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Lewis et al. developed the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Lewis et al., 2020, are the authors who established the RAG approach + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Kevin Scott is the CTO of Microsoft + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + Microsoft conducted a study on the impact of large language models on scientific discovery using GPT-4 + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Preprint is available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Baumel, T. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Elhadad, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Es, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + James, J. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Espinosa-Anke, L. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Schockaert, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, Z. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, X. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Zhao, D. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Yang, M. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Qin, B. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + LangChain is an organization that has published on arXiv + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Wang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zuccon, G. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, R. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Derr, T. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Xu, Y. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lapata, M. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, J. published the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Gan, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yao, L. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, C. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Chiang, W.-L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Sheng, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wu, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lin, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Li, D. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Xing, E. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Preprint is classified under cs.CL on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Preprint was submitted on 24 Apr 2024 + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Preprint has the identifier 2404.16130v1 on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Community detection results in the partition of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The pipeline includes a step for community detection + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 2.0 + Dang, 2006, is the author who established the QFS approach + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Baumel et al., 2018, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Laskar et al., 2020, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Yao et al., 2017, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Goodwin et al., 2020, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Laskar et al., 2022, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Liu and Lapata, 2019, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Achiam et al., 2023, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Brown et al., 2020, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Touvron et al., 2023, are the authors who worked on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Anil et al., 2023, are the authors who worked on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Modularity is an inherent quality of knowledge graphs + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Brown et al. (2020) discuss in-context learning with few-shot examples + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Kuratov et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Liu et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Louvain is a type of community detection algorithm + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Community detection algorithms are used to partition the graph index into communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Fortunato has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Jin et al. have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + HotPotQA dataset is used to evaluate the entity extraction prompt with gpt-4-turbo + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Yang et al. (2018) introduced the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Yang et al. are the authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + GPT-4-Turbo was tested with varying context window sizes + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Tech journalist uses podcast transcripts to look for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + + + 3.0 + Kevin Scott is a participant in the podcast conversations compiled in the Podcast Transcripts dataset. His conversations are included as part of the podcast transcripts, contributing to the overall content and discussions captured within this dataset. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Technology leaders participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + RAG systems are used to evaluate the Podcast Transcripts dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + C0 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C1 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C2 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C3 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + TS is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + SS is a category used in the analysis of podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Units are used to measure the context in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Both are datasets used in the analysis + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Educator uses news articles to incorporate current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + RAG systems are used to evaluate the News Articles dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + C0 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Map-reduce is the method used in the text summarization condition + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The LLM evaluator assesses answers based on the empowerment metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on empowerment + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Empowerment has an average win rate of 51.3% + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Naive RAG mentions Taylor Swift as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Travis Kelce as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Britney Spears as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Justin Timberlake as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG is determined to be the loser based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Global approaches consistently outperformed the naive RAG + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Naive RAG produces the most direct responses + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + SS represents naive RAG in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Gao et al., 2023 discusses naive RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Community partitions enable divide-and-conquer global summarization + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Global summarization can be performed using a graph-free approach + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Source texts are used in global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Final global answer is generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Short descriptions are used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Low-level community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + The use of rich descriptive text for homogeneous nodes in a graph index aligns with the capabilities of LLMs + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Graph indexes differ from typical knowledge graphs in their use of rich descriptive text instead of concise knowledge triples + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The graph index supports condition C0 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C1 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C2 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C3 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index was created using generic prompts for entity and relationship extraction + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Few-shot examples tailored to the domain of the data were used in the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The decision to build a graph index depends on the expected number of lifetime queries per dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The decision to build a graph index depends on the value obtained from it + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The decision to build a graph index depends on the value obtained from other graph-related RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Recall measures the completeness of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Precision measures the accuracy of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to tailor the default prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to tailor the secondary extraction prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of science + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of medicine + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of law + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + A single extraction round is part of the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Domain refers to the specific area of knowledge of the document corpus + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Covariate prompts are used to extract claims linked to detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Source text span is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Start date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + End date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Description is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Subject is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Object is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Communities of entities help manage variations in a noisy graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Common entities are described using rich descriptive text for homogeneous nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + LLMs are used to generate metrics for evaluating natural language generation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Wang et al. (2023) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Zheng et al. (2024) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Relationship edges connect homogeneous nodes in a graph + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Edge weights represent the normalized counts of detected relationship instances on relationship edges + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Each level of the hierarchical community structure provides a community partition + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 2.0 + The hierarchical community structure is a framework used to organize and understand the relationships and dynamics within specialized communities. Root communities are an integral part of this structure, serving as the top-level communities. These root communities form the foundational layer in the hierarchical community structure, providing a basis for further subdivision and organization of more specific sub-communities. This hierarchical approach allows for a systematic analysis of complex networks, facilitating a deeper understanding of the interconnections and dependencies within the domain of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + The hierarchical community structure is a framework that organizes communities into different levels, with sub-communities representing the lower-level communities within this structure. Sub-communities are integral components of the hierarchical community structure, indicating that they are nested within larger communities and contribute to the overall organization and dynamics of the community. This hierarchical arrangement allows for a more detailed and nuanced understanding of the relationships and interactions within the community, facilitating more effective analysis and mapping of complex text data, particularly in specialized domains such as Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community levels are part of the hierarchical community structure + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The Leiden algorithm is used to detect communities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + OpenORD is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Force Atlas 2 is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Nodes represent entities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Edges represent connections between nodes in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Covariates are variables linked to nodes and edges in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Tang and Yang are the authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Questions are generated based on the target datasets + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + N represents the number of test questions per dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Root communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Level 0 represents the root-level communities in the hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Reports provide detailed information about specific subtopics within sub-communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Sub-communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Level 1 represents sub-communities within the root-level communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Partitions can be organized into a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Level 0 is the root level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Level 1 is a sub-level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The token limit defines the maximum number of tokens in the LLM context window + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Prominence is used to prioritize community edges + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Combined source and target node degree is used to measure prominence + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Chunks are divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Helpfulness scores are assigned to intermediate answers + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in episodes dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in how guests perceive the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in discussions about the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in discussions about collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in how news articles address the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in highlighting the importance of health literacy through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + The size of the context window and the prompts used for answer generation are the same across all conditions + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The task is an activity or goal that the user aims to achieve + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Questions are generated based on the user's task + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Datasets were used in combination with questions for the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Questions were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + + + 2.0 + Zheng et al. are the authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Zheng, L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Xing, E. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + MT-Bench and Chatbot Arena are both tools used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 2.0 + Koesten et al. authored a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + Xu and Lapata authored a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Text summarization method applies a map-reduce approach directly to source texts (TS) + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Text summarization is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Semantic search RAG is a na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached (SS) + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Semantic search RAG is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C0 uses root-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C0 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C0 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C0 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 uses high-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C1 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C1 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C1 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 uses intermediate-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C2 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C2 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C2 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 uses low-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C3 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C3 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C3 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + TS is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + SS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The graph indexing process used 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + A graph was created for the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Intermediate-level summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + The graph indexing process used 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + A graph was created for the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Datasets were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Wang et al., 2023a discusses the state-of-the-art results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Zheng et al., 2024 discusses the competitive results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Zheng et al., 2024 discusses the LLM-as-a-judge method + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Embedding-based matching is used to match user queries + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Query-time LLM use was evaluated with different context window sizes + 4c855404ee3d3c94aa2136f1513c666f + + + 2.0 + The **CONTEXT WINDOW SIZE** and **FINAL EVALUATION** are closely related in the given data. A fixed context window size of 8k tokens was used for the final evaluation. This indicates that during the final evaluation phase, the system or model was configured to process and analyze text data within a predefined window of 8,000 tokens, ensuring consistency and standardization in the evaluation process. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Natural Language Generation achieves state-of-the-art results + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation achieves competitive results + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation is compared against human judgements + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation can generate reference-based metrics + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation can measure qualities in a reference-free style + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Es et al., 2023 discusses the RAGAS method + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates context relevance + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates faithfulness + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates answer relevance + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator assesses answers based on the directness metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Table 2 shows an example of LLM-generated assessment + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses a head-to-head comparison approach + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator assesses answers based on target metrics + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses a control metric for validity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator accounts for stochasticity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses mean scores from multiple comparisons + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Directness is used to evaluate the straightforwardness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The question asks about public figures mentioned in entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Public figures are repeatedly mentioned across various entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Answer 1 covers a wide range of public figures from different sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Controversies involve public figures and impact public discourse. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Entertainment articles cover topics related to the entertainment industry + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Taylor Swift is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Travis Kelce is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Britney Spears is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Justin Timberlake is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Taylor Swift is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Travis Kelce is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Britney Spears is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Justin Timberlake is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Actors and Directors are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Musicians and Executives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Athletes and Coaches are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Influencers and Entrepreneurs are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Public Figures in Controversy are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Film is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Television is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Music is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Sports is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Digital Media is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Cultural Narratives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Trends are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Social Discussions are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Public Discourse is a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Reference-based metrics require gold standard answers + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Gold standard answers are lacking for sensemaking questions + 322e02986c8724eedbcf3ebfa20b989c + + + 3.0 + End users play a crucial role in the validation process of sensemaking questions and target metrics. Sensemaking questions are specifically validated with end users to ensure their relevance and accuracy. This collaborative approach ensures that the questions and metrics are aligned with the needs and expectations of the end users, thereby enhancing the overall effectiveness and applicability of the sensemaking process. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Target metrics are validated with end users + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The control metric is used as an indicator of validity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Taylor Swift is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Taylor Swift is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Travis Kelce is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Travis Kelce is a public figure in the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Britney Spears is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Britney Spears is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Justin Timberlake is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Justin Timberlake is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the film sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the television sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on public figures primarily from the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on public figures primarily from the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the digital media sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 cites specific data sources from the News article dataset for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 provides insights into controversies involving public figures and their impact on public discourse. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the gaming sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 cites specific data sources for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 was generated using the Naïve RAG method, which directly lists specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d + + + 2.0 + ANSWER 2 is a generated answer for a question in the NEWS ARTICLE DATASET. It relies heavily on a single source from the NEWS ARTICLE DATASET for data. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Answer 2 relies heavily on a single data source. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Naïve RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The global approach to Graph RAG shows improvements over naïve RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + LLM-generated assessments are used to evaluate the answers produced for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Example question is part of the News article dataset used for analysis + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Head-to-head win rate percentages were used to compare different conditions + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Win rate percentages were used to measure the performance of different conditions + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The overall winner per dataset and metric was determined for each condition + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Self-win rates were shown as the expected 50% for each condition + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The indexing process resulted in the creation of graphs + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Map-reduce summarization requires the highest number of context tokens + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Root-level community summaries require dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + + + 2.0 + Queries are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Self-memory is related to generation-augmented retrieval + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + CAiRE-COVID is a system for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + ITRG is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + IR-CoT is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + DSP is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + RAPTOR is a method for generating a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + The paper by Baek et al. discusses the KAPING method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by He et al. discusses the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Zhang discusses the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Kang et al. discusses the SURGE method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Ranade and Joshi discusses the FABULA method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Both LangChain and LlamaIndex support a variety of graph databases + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LangChain supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LangChain supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 1.0 + LangChain developed Langchain graphs + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + LlamaIndex supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LlamaIndex supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + NaLLM is a method that can create and reason over knowledge graphs in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Neo4J developed Project NaLLM + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + GraphRAG is a method that can create and reason over knowledge graphs in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Manakul et al. discusses the SelfCheckGPT method + 92e93fc6449756c0a60200636b297f65 + + + 1.0 + SelfCheckGPT is an approach mentioned in the work by Manakul et al., 2023 + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + SelfCheckGPT is used to compare fabrication rates + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Embedding-based matching is used to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Hybrid RAG schemes combine embedding-based matching against community reports + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The roll-up operation can be extended using map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The drill down mechanism follows the information scent in the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The global approach to Graph RAG achieves competitive performance at a fraction of the token cost + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The open-source implementation of Graph RAG approaches is Python-based + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The drill down mechanism follows the information scent + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Alonso Guevara Fernández and Amber Hoak both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Sarah Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Shane Solomon both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Adler co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Altenschmidt and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Altenschmidt and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Altman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and S. Borgeaud co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Schalkwyk and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Schalkwyk and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + A. M. Dai and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Baek and A. F. Aji co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Baek and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + A. F. Aji and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + T. Ban and L. Chen co-authored the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + 1.0 + Baumel, T. and Eyal, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Baumel, T. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Baumel, T. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Elhadad, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Guillaume, J.-L. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Guillaume, J.-L. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Guillaume, J.-L. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Lambiotte, R. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Mann, B. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Zhao, D. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Es, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + James, J. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Espinosa-Anke, L. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Schockaert, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, Z. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, X. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Yang, M. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Qin, B. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Gao, Y. and Xiong, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Dai, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Dai, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Sun, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Goodwin, T. R. and Savery, M. E. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Goodwin, T. R. and Demner-Fushman, D. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + 2.0 + Khattab, O. and Santhanam, K. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and highlights their collaborative work in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Khattab, O. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This collaboration is mentioned in the text, highlighting their joint contribution to the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Santhanam, K. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Li, X. L. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hall, D. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Liang, P. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + LIANG, P. and ZAHARIA, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Liang, P. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Potts, C. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Potts, C. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Zaharia, M. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Kim, S. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Park, J. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Park, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kang, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. and Moon, B. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Moon, B. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Moon, B. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoffman, R. R. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Gregory, K. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Groth, P. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Groth, P. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Simperl, E. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kuratov, Y. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Bulatov, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Anokhin, P. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Sorokin, D. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Sorokin, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Burtsev, M. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Laskar, M. T. R. and Hoque, E. co-authored two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning with transformer models to improve the effectiveness of query-focused abstractive summarization. Both works contribute to advancing the application of transformer models in specialized summarization tasks. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Laskar, M. T. R. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Hoque, E. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoque, E. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoque, E. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Hoque, E. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Huang, J. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + The 33rd Canadian Conference on Artificial Intelligence is also known as Canadian AI 2020 + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Springer published the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Huang, J. X. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Perez, E. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Karpukhin, V. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Xu, Y. and Lapata, M. co-authored the paper "Text summarization with latent queries" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Huang, M. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Duan, N. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 3.0 + Martin, S. and Brown, W. M. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with large graph structures, making it a valuable resource for researchers and practitioners in the domain of graph theory and network analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Martin, S. and Klavans, R. co-authored the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Martin, S. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the capabilities and applications of the Openord toolbox, emphasizing its utility in handling extensive graph data efficiently. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Brown, W. M. and Klavans, R. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Brown, W. M. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + KLAVANS, R. and BOYACK, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Newman, M. E. published the paper "Modularity and community structure in networks" in the Proceedings of the National Academy of Sciences + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Levine, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Shashua, A. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Shashua, A. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Leyton-Brown, K. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ranade, P. and Joshi, A. co-authored the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Abdullah, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Yang, Z. and Manning, C. D. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Huang, M. and Duan, N. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Xu, Y. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Siddique, F. B. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Siddique, F. B. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Barezi, E. J. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Tang, Y. and Yang, Y. co-authored the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Martin, L. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bhargava, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bhosale, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Liang, Y. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 2.0 + Li, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Li, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Li, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 2.0 + Xu, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Xu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Qu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Khramtsova, E. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wang, Y. and Lipka, N. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, R. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Qi, P. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Zhang, S. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Bengio, Y. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Cohen, W. W. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Salakhutdinov, R. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Chiang, W.-L. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Xing, E. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + \ No newline at end of file diff --git a/graphfleet/output/graphindex/artifacts/join_text_units_to_covariate_ids.parquet b/graphfleet/output/graphindex/artifacts/join_text_units_to_covariate_ids.parquet new file mode 100644 index 000000000..928ce7c53 Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/join_text_units_to_covariate_ids.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/join_text_units_to_entity_ids.parquet b/graphfleet/output/graphindex/artifacts/join_text_units_to_entity_ids.parquet new file mode 100644 index 000000000..ff65c72a5 Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/join_text_units_to_entity_ids.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/join_text_units_to_relationship_ids.parquet b/graphfleet/output/graphindex/artifacts/join_text_units_to_relationship_ids.parquet new file mode 100644 index 000000000..e8bcc6315 Binary files /dev/null and b/graphfleet/output/graphindex/artifacts/join_text_units_to_relationship_ids.parquet differ diff --git a/graphfleet/output/graphindex/artifacts/merged_graph.graphml b/graphfleet/output/graphindex/artifacts/merged_graph.graphml new file mode 100644 index 000000000..27878c09d --- /dev/null +++ b/graphfleet/output/graphindex/artifacts/merged_graph.graphml @@ -0,0 +1,10064 @@ + + + + + + + + + + + PERSON + Darren Edge is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Ha Trinh is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Newman Cheng is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Joshua Bradley is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Alex Chao is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Apurva Mody is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Steven Truitt is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Jonathan Larson is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Research is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Strategic Missions and Technologies is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Office of the CTO is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + METHOD + RAG (Retrieval-Augmented Generation) is a method used to retrieve relevant information from an external knowledge source to enable large language models to answer questions +RAG (Retrieval-Augmented Generation) is a method used for generating responses in text generation tasks +RAG (Retrieval-Augmented Generation) is a method that produces direct responses in text generation tasks + +Retrieval-Augmented Generation, a method that incorporates retrieval of relevant data to augment text generation +RAG (Retrieval-Augmented Generation) is a developing research area with multiple established directions, including knowledge graph creation, completion, and extraction of causal graphs + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY + LLM (Large Language Model) is a type of model used to automate human-like sensemaking and reasoning over large collections of documents +LLM (Large Language Model) is a type of AI model used for generating text and answering queries +LLM (Large Language Model) is used to process text chunks and extract elements of a graph index +LLM (Large Language Model) is a type of artificial intelligence used for tasks such as entity extraction, summarization, and understanding relationships in text +LLM (Large Language Model) is used to generate intermediate answers and scores for each chunk +Large Language Model used to automate the generation of questions for dataset evaluation +LLM (Large Language Model) is a type of artificial intelligence used for generating and assessing text +Large Language Models (LLMs) are used to analyze and generate text based on retrieved information and queries +Large Language Model, a type of AI model with a context window that can be exceeded by external datasets + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,2c6ed90897310eea2f28e33fff1c32b0,6f33a085ff3304e5994f7fbb86c881a4,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + METHOD + Graph RAG is an approach that combines retrieval-augmented generation with graph-based text indexing to answer questions over private text corpora +A new approach based on global summarization of an LLM-derived knowledge graph, targeting global summarization tasks +Graph RAG is a specific approach to RAG that focuses on global summarization using a knowledge graph +Graph RAG (Retrieval-Augmented Generation) is an approach that involves a high-level data flow and pipeline for processing and summarizing text +A specific implementation of RAG using four levels of graph communities +Graph RAG is a method using graph communities at different levels to answer user queries +A multi-stage mechanism for Retrieval-Augmented Generation (RAG) that involves comparing multiple conditions +Graph RAG is a method that provides a comprehensive overview of public figures in the entertainment industry +Graph RAG is a method used to generate responses that provide a comprehensive and structured overview of public figures across various sectors of the entertainment industry. +Graph RAG is a method used to generate answers for questions in the News article dataset +Graph RAG is a method that outperformed naive RAG on comprehensiveness and diversity in text generation tasks +A method used to compare community summaries to source texts, generally providing a small but consistent improvement in answer comprehensiveness and diversity +Graph RAG is a method that uses community summaries to improve answer comprehensiveness and diversity while requiring fewer tokens compared to source text summarization +An implementation of RAG that incorporates multiple concepts from other systems, including self-memory and parallel generation of community answers +A method that uses the natural modularity of graphs to partition data for global summarization +Graph RAG is a method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) to support human sensemaking over text corpora +Graph RAG (Retrieval-Augmented Generation) is a method that combines global and local approaches for efficient token usage in text generation tasks +Graph RAG is a retrieval-augmented generation technology based on knowledge graphs launched by NebulaGraph + 086021a89900a39bcb62036981737bfa,21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,718017a4871c909420f84b85b8ba969d,833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19,fb3c48579608fa28be585ceb6cd2f0fe + + + METHOD + QFS (Query-Focused Summarization) is a method used to generate summaries based on specific user queries + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + Community summaries are pre-generated summaries for groups of closely-related entities used in the Graph RAG approach +Community summaries are summaries of graph communities, tailored to the domain +Summaries generated from modular communities in the knowledge graph +Report-like summaries of each community in a hierarchical structure, useful for understanding the dataset +Community summaries are generated summaries of data clusters or communities, used to answer queries +Community summaries are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks +Summaries of different levels of each graph community hierarchy +Community summaries are summaries derived from community-generated content, used in the analysis to compare with source texts +Summaries that act as a kind of self-memory for generation-augmented retrieval +Summaries of root-level communities in an entity-based graph index + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + CONCEPT + Global sensemaking questions are questions that require understanding and summarizing large datasets, often beyond the explicit content of the source texts + e8d83e6e7a7c0f57b218cef24976b745 + + + METRIC + 1 million token range refers to the scale of datasets used in the evaluation of the Graph RAG approach + e8d83e6e7a7c0f57b218cef24976b745 + + + TECHNOLOGY + Python is a programming language used to implement the open-source version of the Graph RAG approach +Python is a programming language used for implementing both global and local Graph RAG approaches + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + + + URL + The URL where the open-source implementation of the Graph RAG approach will be available +A URL where the open-source, Python-based implementation of Graph RAG approaches will be available + e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745 + + + METHOD + Query-Focused Summarization (QFS) is a method used to generate summaries based on specific user queries +Query-focused summarization is a method used to generate summaries that are relevant to a specific query +A summarization method that focuses on answering specific queries using the entire corpus +A method for summarizing information based on specific queries + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + + + CONCEPT + An external knowledge source is a repository of information that can be accessed to retrieve relevant data for answering questions + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + A text corpus is a large collection of written texts used for analysis and research + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + Source documents are the original texts from which information is retrieved or summarized +Source documents are the original texts from which information is extracted and analyzed +Source documents are the original texts from which input texts are extracted for processing in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + + + CONCEPT + A partial response is an intermediate answer generated from community summaries before being combined into a final response + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + A final response is the comprehensive answer generated after combining all partial responses + e8d83e6e7a7c0f57b218cef24976b745 + + + METRIC + Comprehensiveness is a metric used to evaluate the extent to which generated answers cover the relevant information +A target quality used to evaluate the summarization approach, focusing on the completeness of the summary +A metric that measures how much detail an answer provides to cover all aspects and details of a question +A metric used to evaluate the comprehensiveness of the generated responses +A metric used to evaluate the thoroughness of the generated answers +A metric used to evaluate the quality of answers in terms of their completeness +A metric used to evaluate the thoroughness of responses, with win rates between 72-83% for Podcast transcripts and 72-80% for News articles + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + + + METRIC + Diversity is a metric used to evaluate the variety of information included in the generated answers +A target quality used to evaluate the summarization approach, focusing on the variety of information in the summary +A metric that measures how varied and rich an answer is in providing different perspectives and insights on a question +A metric used to evaluate the variety in the generated answers +A metric used to evaluate the variety of answers generated +A metric used to evaluate the variety of responses, with win rates ranging from 75-82% for Podcast transcripts and 62-71% for News articles +Diversity is a measure used to evaluate the variety of answers provided by different methods + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + + + ACTIVITY + Human endeavors refer to activities and efforts across various domains that rely on reading and reasoning about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + + + TECHNOLOGY + Large language models are advanced AI models designed to understand and generate human-like text, used in automating sensemaking in complex domains +Modern language models, including GPT, Llama, and Gemini, that can use in-context learning to summarize content + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + + + DOMAIN + Scientific discovery is a complex domain where sensemaking is applied to understand and generate new knowledge from scientific texts + f0306814bf64f5c9e79603fc6a52f4ea + + + DOMAIN + Intelligence analysis is a complex domain where sensemaking is applied to understand and generate insights from intelligence data + f0306814bf64f5c9e79603fc6a52f4ea + + + PROCESS + Sensemaking is the process of understanding connections among people, places, and events to anticipate their trajectories and act effectively +Sensemaking is the process of understanding and making sense of complex information + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Text chunks are segments of source documents that are extracted for further analysis +Text chunks are segments of input texts extracted from source documents, used for processing in the Graph RAG approach +Segments of text that are embedded into a vector space for analysis + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + DATA + Element instances are specific pieces of information extracted from text chunks, tailored to the domain +Element instances are identified and extracted instances of graph nodes and edges from text chunks +Element instances are individual occurrences of entities, relationships, and claims extracted from source texts + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Element summaries are concise representations of element instances, tailored to the domain +Element summaries are descriptive texts created by the LLM to summarize entities, relationships, and claims extracted from source texts +Summaries of elements within a graph, used to understand the structure and semantics of the dataset +Element summaries are detailed descriptions of nodes, edges, and covariates within a community + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Graph communities are groups of elements (nodes, edges, covariates) detected in a graph index, used for summarization +Groups of nodes within a graph that have stronger connections to each other than to other nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Community answers are query-focused summaries of community summaries +Community answers are responses generated from community summaries to answer user queries +Answers generated in parallel from community summaries + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + DATA + Global answer is the final query-focused summary produced from all relevant community summaries +A global answer is a comprehensive response generated from multiple community summaries to answer a user query +The final answer generated by combining intermediate community answers based on their helpfulness scores + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + TIME + Indexing time refers to the time when the graph index is created and elements are summarized + f0306814bf64f5c9e79603fc6a52f4ea + + + TIME + Query time refers to the time when a query is made and the relevant summaries are generated + f0306814bf64f5c9e79603fc6a52f4ea + + + PROCESS + Graph RAG pipeline is a process using an LLM-derived graph index to detect, extract, and summarize nodes, edges, and covariates in source documents + f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Nodes are entities detected in the graph index of source documents +The individual elements or points in a graph, with 8564 nodes for the Podcast dataset and 15754 nodes for the News dataset + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Edges are relationships detected in the graph index of source documents +The connections or links between nodes in a graph, with 20691 edges for the Podcast dataset and 19520 edges for the News dataset + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Covariates are claims or additional information detected in the graph index of source documents +Covariates are additional attributes associated with extracted node instances in the graph index + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Leiden is a community detection method used to partition the graph index into groups of elements +Leiden is a community detection algorithm used to partition graphs into modular communities +A community detection algorithm known for its ability to recover hierarchical community structure efficiently + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Retrieval-augmented generation is an approach to answering user questions over entire datasets by retrieving and generating relevant information +RAG is an established approach to answering user questions over entire datasets by retrieving relevant text regions to provide grounding for the generation task + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + + + ORGANIZATION + Microsoft is an organization involved in automating sensemaking in scientific discovery using LLMs +Microsoft is a technology company whose CTO, Kevin Scott, participates in the podcast conversations +Microsoft is an organization that conducted a study on the impact of large language models on scientific discovery using GPT-4 + 1d07b4248c2655081c7af0e373bd70c9,833e7d67dcd30790b26b71c9b5306f6b,f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Ranade is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Joshi is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Klein is an author who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Lewis is an author who contributed to the development of the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Traag is an author who contributed to the development of the Leiden community detection method + f0306814bf64f5c9e79603fc6a52f4ea + + + PUBLICATION + arXiv is a repository where the preprint of the discussed research paper is available +arXiv is a repository where the mentioned papers are published as preprints +The platform where the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" was published +arXiv is the platform where the paper "Retrieval-augmented generation for large language models: A survey" was publishedarXiv is the platform where the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" was published +arXiv is a repository of electronic preprints (known as e-prints) approved for publication after moderation, but not full peer review +The preprint server where the paper "Lost in the middle: How language models use long contexts" was published +arXiv is a preprint repository where several papers mentioned in the text were published +arXiv is a repository where the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" was publishedarXiv is a repository where the paper "Llama 2: Open foundation and fine-tuned chat models" was publishedarXiv is a repository where the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" was publishedarXiv is a repository where the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" was publishedarXiv is a repository where the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" was publishedarXiv is a repository where the paper "Enhancing knowledge graph construction using large language models" was publishedarXiv is a repository where the paper "Is chatgpt a good nlg evaluator? a preliminary study" was published +arXiv is the platform where the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" was publishedarXiv is the platform where the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" was publishedarXiv is the platform where the paper "Is chatgpt a good nlg evaluator? a preliminary study" was publishedarXiv is the platform where the paper "Causal graph discovery with retrieval-augmented generation based large language models" was publishedarXiv is the platform where the paper "Knowledge graph prompting for multi-document question answering" was publishedarXiv is the platform where the paper "Text summarization with latent queries" was published + 00e8e4e881bd0862022f4dfc913b900b,086021a89900a39bcb62036981737bfa,58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035,f0306814bf64f5c9e79603fc6a52f4ea,fc4b27d64f055b7fc30176ba110dd02e + + + PUBLICATION + Preprint refers to the version of the research paper that is under review and available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + CATEGORY + cs.CL is the category under which the research paper is classified on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + DATE + 24 Apr 2024 is the date when the research paper was submitted to arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + IDENTIFIER + 2404.16130v1 is the identifier for the research paper on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Document collections refer to large sets of documents that are analyzed for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + TECHNOLOGY + LLM prompts are specific instructions given to large language models to tailor their responses to the domain of the dataset +LLM prompts are specific instructions given to the LLM to extract elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Community detection is a method used to identify groups of related elements within a graph +The process of identifying communities within a graph + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Domain-tailored summarization is a method used to create summaries that are specific to the domain of the dataset + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Klein et al. are authors who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Ranade and Joshi are authors who discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Lewis et al. are authors who developed the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Traag et al. are authors who developed the Leiden community detection method +Authors of the Leiden algorithm, known for its efficiency in recovering hierarchical community structures + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + QFS is a task framing that focuses on generating summaries based on specific queries, rather than just concatenating excerpts + fb3c48579608fa28be585ceb6cd2f0fe + + + METHOD + A type of summarization that generates natural language summaries based on specific queries, rather than just extracting text + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A neural network architecture that has shown substantial improvements in various summarization tasks + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization +Gemini is a family of highly capable multimodal models described in an arXiv preprint + 086021a89900a39bcb62036981737bfa,fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A structured representation of knowledge used in the Graph RAG approach for global summarization +A knowledge graph is a structured representation of information, used in the Graph RAG approach for summarization + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + + + REFERENCE + Authors of a paper on Retrieval-augmented generation (RAG) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Author of a paper on query-focused summarization (QFS) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the GPT series of large language models +A reference to a publication by Brown et al. in 2020, discussing in-context learning with few-shot examples + bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the limitations of LLM context windows +A reference to a publication by Kuratov et al. in 2024, discussing the recall degradation of longer LLM context windows +A reference to a study by Kuratov et al. in 2024, discussing the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the limitations of LLM context windows +A reference to a publication by Liu et al. in 2023, discussing the recall degradation of longer LLM context windows +A reference to a study by Liu et al. in 2023, discussing the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + TECHNOLOGY + Algorithms used to partition graphs into modular communities of closely-related nodes +Algorithms used to partition a graph into communities of nodes with stronger connections to one another + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + ALGORITHM + Louvain is a community detection algorithm used to partition graphs into modular communities + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + HotPotQA is a dataset used to evaluate the entity extraction prompt with gpt-4-turbo +HotPotQA is a dataset used to observe the behavior of text chunk extraction in the Graph RAG approach +A benchmark dataset for open-domain question answering, targeting explicit fact retrieval + 21e52bc06a82796b1f4bcd73edda1f2a,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNOLOGY + GPT-4-Turbo is a version of the GPT-4 model used for entity extraction in the evaluation +GPT-4-Turbo is a model with a large context size of 128k tokens used in the analysis + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + + + DATASET + A dataset consisting of transcripts from podcasts used for analysis +A dataset consisting of compiled transcripts of podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders +Compiled transcripts of podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders. Size: 1669 × 600-token text chunks, with 100-token overlaps between chunks, approximately 1 million tokens +A dataset consisting of transcripts from podcasts used for analysis +A dataset consisting of transcripts from podcasts used for analysis + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620,ebf5249c888e07fedce6572a4c03f88c + + + DATASET + A dataset consisting of news articles used for analysis +A dataset consisting of news articles used for analysis +Benchmark dataset comprising news articles published from September 2013 to December 2023 in a range of categories, including entertainment, business, sports, technology, health, and science. Size: 3197 × 600-token text chunks, with 100-token overlaps between chunks, approximately 1.7 million tokens +A dataset consisting of news articles used for analysis + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620 + + + METHOD + Map-reduce is a method used for query-focused summarization of an entire corpus +A method used for text summarization by applying a map-reduce approach directly to source texts + 21e52bc06a82796b1f4bcd73edda1f2a,973164fa90bf2b4ee267f4fd795916bf + + + METRIC + A target quality used to evaluate the summarization approach, focusing on the ability to develop understanding of broad issues and themes +A metric that measures how well an answer helps the reader understand and make informed judgements about a topic +A metric used to evaluate how empowering the generated answers are +A metric used to evaluate the effectiveness of answers in empowering users +A concept or metric used in the evaluation, with an average win rate of 51.3% +Empowerment is a measure used to evaluate the ability of different methods to help users reach an informed understanding + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + Naive RAG is a basic approach to RAG used as a baseline for comparison +Naive RAG is a method that lists public figures mentioned in entertainment articles, focusing on their professional achievements and personal lives +Naive RAG is a baseline method used for comparison in text generation tasks +A method that produces the most direct responses but is outperformed by global approaches in comprehensiveness and diversity +Naive RAG is a basic retrieval-augmented generation method that converts documents to text, splits them into chunks, and embeds these chunks into a vector space for query matching + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c + + + METHOD + A method for summarizing source texts using a map-reduce approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Questions generated to evaluate the summarization approach, focusing on understanding activities + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + METRIC + The computational cost measured in tokens used in the summarization process +Token costs refer to the number of tokens required for processing text in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS + The high-level process of the Graph RAG approach and pipeline + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + Key parameters that influence the design of the Graph RAG approach and pipeline +Design parameters are key settings and configurations in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + PARAMETER + + + METHOD + Global summarization is a method that aims to summarize information from a large dataset or corpus +A method for summarizing the overall structure and semantics of a dataset +A method for summarizing information on a global scale + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e4d9b12cf2b4c691c74019eefff4fb39 + + + ATTRIBUTE + Modularity is an inherent quality of graphs that allows them to be partitioned into communities of closely-related nodes + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Descriptions generated from modular communities in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + + + INPUT + A specific question or request for information that the summarization methods aim to answer + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + A large collection of texts or documents used for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Intermediate answers generated from community summaries before being combined into a final global answer + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + The comprehensive answer generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + + + METHOD + A method that focuses on generating questions to understand activities from datasets + 21e52bc06a82796b1f4bcd73edda1f2a + + + INPUT + Brief descriptions of datasets used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + Datasets that represent real-world information, such as podcast transcripts and news articles + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + METHOD + A method that summarizes the original source texts directly + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Summaries generated from lower hierarchical levels of the community in the knowledge graph +Low-level community summaries are summaries that provide a detailed overview of the source text +Low-level community summaries are a type of community summary used in the News dataset for analysis + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + OUTPUT + Summaries generated from intermediate hierarchical levels of the community in the knowledge graph +Intermediate-level community summaries are summaries that provide a mid-level overview of the source text + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + OUTPUT + Summaries generated from higher hierarchical levels of the community in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + + + PROCESS, SYSTEM + The pipeline refers to the sequence of steps and processes involved in the Graph RAG approach +A series of processes or steps used to analyze and summarize a dataset + 7fb7d9ce2da9c940a32afdd87d1d9e56,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA STRUCTURE, OUTPUT + Graph index is a data structure that includes various elements extracted from text chunks using LLM prompts +An index created from a graph structure, used for query-focused summarization and other tasks +The graph index supporting conditions C0-C3, created using generic prompts for entity and relationship extraction +A self-generated index that enables Graph RAG +A data structure used in RAG systems to organize and retrieve information +An index built using a graph structure to organize and retrieve information + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + + + DATA, UNIT + Entity references are mentions of entities within text chunks, extracted during the processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC + Recall is a metric used to measure the completeness of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC + Precision is a metric used to measure the accuracy of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + Few-shot examples are sample inputs provided to the LLM for in-context learning to tailor the extraction prompt to the document corpus domain +Few-shot examples are specialized instances provided to the LLM to improve its performance in domains with specialized knowledge such as science, medicine, and law +Examples tailored to the domain of the data used in the graph indexing process + 2c6ed90897310eea2f28e33fff1c32b0,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA, UNIT + Named entities are specific types of entities such as people, places, and organizations, extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + REFERENCE, PUBLICATION + A reference to a publication by Yang et al. in 2018, introducing the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METHOD, APPROACH + Techniques refer to the specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Implementation details are specific configurations and settings used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS, METHOD + A single extraction round refers to one complete cycle of extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC, ISSUE + Recall degradation refers to the decrease in recall performance when using longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS, METHOD + The extraction process involves identifying and extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Domain refers to the specific area of knowledge or field to which the document corpus belongs + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA, INPUT + Document corpus refers to the collection of documents being processed in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + Default prompt is the standard set of instructions given to the LLM for extracting named entities + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + Secondary extraction prompt is an additional set of instructions given to the LLM for extracting covariates + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METHOD + A covariate prompt is used to extract additional attributes associated with detected entities, including claims linked to the entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + Claims are statements or assertions linked to detected entities, including details such as subject, object, type, description, source text span, and start and end dates + 2c6ed90897310eea2f28e33fff1c32b0 + + + METHOD + Gleanings refer to multiple rounds of entity extraction to ensure that no entities are missed in the process + 2c6ed90897310eea2f28e33fff1c32b0 + + + TECHNIQUE + Logit bias is a technique used to force a yes/no decision from the LLM during the entity extraction process + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + An entity node is a representation of an entity in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A relationship edge is a representation of a relationship between entities in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A claim covariate is an additional attribute or variable associated with a claim in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + Communities of entities are groups of closely-related entities detected and summarized by the LLM + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A noisy graph structure is a graph that may contain duplicate or inconsistent entity elements due to variations in text format +A graph structure that may contain inconsistencies or errors, making it challenging to analyze + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + DOMAIN + Science is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + DOMAIN + Medicine is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + DOMAIN + Law is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Source text span is an attribute of a claim that indicates the specific portion of text from which the claim was extracted + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Start date is an attribute of a claim that indicates when the event or fact described in the claim began + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + End date is an attribute of a claim that indicates when the event or fact described in the claim ended + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Description is an attribute of a claim that provides a detailed explanation of the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Subject is an attribute of a claim that indicates the main entity involved in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Object is an attribute of a claim that indicates the entity that is affected by the subject in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A concept referring to an entity that has multiple name variations but is resilient to such variations due to sufficient connectivity to closely-related entities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models capable of understanding and generating human-like text +Large Language Models used as evaluators of natural language generation +Large Language Models (LLMs) are used for various tasks such as knowledge graph creation and completion + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf + + + TECHNOLOGY + Structured representations of knowledge in the form of triples (subject, predicate, object) used for reasoning tasks + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + Nodes in a graph that are of the same type and are described using rich descriptive text + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + Edges in a graph that represent relationships between entity nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + METRIC + Weights assigned to edges in a graph, representing the normalized counts of detected relationship instances + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + A structure in which communities are organized in a hierarchy, with each level providing a partition of the graph nodes +Hierarchical community structure is a multi-level clustering of communities used to generate community summaries +A structure that organizes data into a hierarchy of communities + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39 + + + CONCEPT + A division of graph nodes into mutually-exclusive, collectively-exhaustive communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + TECHNOLOGY + A specific implementation or variant of Retrieval-Augmented Generation (RAG) used in the context of graph communities +MultiHop-RAG is a dataset used for community detection and analysis +A benchmark dataset comprising news articles published from September 2013 to December 2023 in a range of categories, including entertainment, business, sports, technology, health, and scienceA benchmark dataset for open-domain question answering, targeting explicit fact retrieval + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author who has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + PERSON + Authors who have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + A collection of data used for analysis and summarization +A collection of data used for analysis, such as podcast transcripts or news articles +A collection of data used for evaluation, including the Podcast and News datasets + 1d07b4248c2655081c7af0e373bd70c9,7fb7d9ce2da9c940a32afdd87d1d9e56,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + Queries that aim to retrieve information from a global perspective, covering the entire dataset +Global queries refer to questions or inquiries that require comprehensive answers derived from multiple sources or datasets + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + The top-level communities in a hierarchical community structure +Root communities are the top-level clusters in a hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Lower-level communities in a hierarchical community structure, providing more detailed information +Sub-communities are lower-level clusters within root communities in a hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + TECHNOLOGY + Detailed documents that provide information about specific subtopics within a community + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The division of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + A system in which elements are ranked or organized in levels + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The root level in a hierarchical community structure +Level 0 represents the root-level communities in the hierarchical clustering with maximum modularity + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + A sub-level in a hierarchical community structure, providing more detailed information +Level 1 represents sub-communities within the root-level communities, revealing internal structure + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + A visual representation of graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + METHOD + The Leiden algorithm is a method used for detecting communities in large networks + 843fc5421e086120ffa1c75856ecf6cd + + + TOOL + OpenORD is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + + + TOOL + Force Atlas 2 is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Nodes represent entities in a graph, with size proportional to their degree + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Edges represent connections between nodes in a graph + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Covariates are variables that are linked to nodes and edges in a graph + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + The LLM context window is the token limit within which summaries are added for processing by a language model + 843fc5421e086120ffa1c75856ecf6cd + + + METHOD + Hierarchical clustering is a method of clustering data into a tree-like structure with multiple levels + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + The token limit is the maximum number of tokens that can be processed in a single context window by a language model + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Summary detail refers to the level of detail provided in a summary + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Scope refers to the range or extent of information covered in a summary + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + A user query is a question or inquiry posed by a user seeking information +A query from the user that the system aims to answer + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd + CONCEPT + + + ELEMENT + Chunks are segments of community summaries divided into pre-specified token sizes + 843fc5421e086120ffa1c75856ecf6cd + ELEMENT + + + METRIC + Prominence is a metric used to prioritize community edges based on the combined degree of source and target nodes + 843fc5421e086120ffa1c75856ecf6cd + + + METRIC + Combined source and target node degree is a metric used to measure the overall prominence of community edges + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Community edges are connections between nodes within a community, prioritized based on prominence + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Sub-community summaries are shorter summaries of sub-communities used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + + + CATEGORY + Community level refers to the different levels in the hierarchical community structure used to generate summaries + 843fc5421e086120ffa1c75856ecf6cd + + + DATA + Chunks are segments of community summaries divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + + + METRIC + A score between 0-100 generated by the LLM to indicate how helpful an answer is in addressing the target question + 1d07b4248c2655081c7af0e373bd70c9 + + + USER + A user looking for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + + + USER + A user incorporating current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic addressing the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + The importance of health literacy highlighted through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + + + OUTPUT + Answers generated for each chunk of community summaries + 1d07b4248c2655081c7af0e373bd70c9 + + + METRIC + The pre-specified size of tokens used to divide community summaries into chunks + 1d07b4248c2655081c7af0e373bd70c9 + + + TECHNOLOGY + A window of text used to generate answers, limited by token size +The size of the context window used for answer generation, which is the same across all conditions + 1d07b4248c2655081c7af0e373bd70c9,973164fa90bf2b4ee267f4fd795916bf + + + PERSON + Kevin Scott is the CTO of Microsoft and a participant in the podcast conversations +Microsoft CTO who participates in podcast conversations compiled in the dataset + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + Individuals who are leaders in the technology industry and participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + + + INPUT + A specific activity or goal that the user aims to achieve using the datasets + 1d07b4248c2655081c7af0e373bd70c9 + + + INPUT + Specific questions generated by the LLM based on the user's task and the target datasets +The questions used in the analysis to evaluate the performance of different methods + 1d07b4248c2655081c7af0e373bd70c9,4c855404ee3d3c94aa2136f1513c666f + + + + + 1d07b4248c2655081c7af0e373bd70c9 + + + DATASET + A benchmark dataset for open-domain question answering, targeting explicit fact retrieval +MT-Bench is a benchmarking tool used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + 922778ce1cb2fdd6dbab1746c8795620,b1bbda43309e8e0e2175ea034aa88e13 + DATASET + + + PROCESS + The process through which people inspect, engage with, and contextualize data within the broader scope of real-world activities + 922778ce1cb2fdd6dbab1746c8795620 + PROCESS + + + TECHNOLOGY + Retrieval-Augmented Generation systems used for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + TECHNOLOGY + + + AUTHORS + Authors of a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors of a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + AUTHORS + Authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + + + 922778ce1cb2fdd6dbab1746c8795620 + + + PODCAST + A podcast series featuring conversations between Kevin Scott and other technology leaders +Behind the Tech is a media platform associated with Scott, K. + 833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + Kevin Scott, Microsoft CTO, who participates in the podcast conversations compiled in the dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + DATASET + A benchmark dataset for open-domain question answering, targeting explicit fact retrieval + 922778ce1cb2fdd6dbab1746c8795620 + + + METRIC + N represents the number of test questions per dataset used in the evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + METHOD + A method applying a map-reduce approach directly to source texts for summarization + 973164fa90bf2b4ee267f4fd795916bf + + + METHOD + A na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached + 973164fa90bf2b4ee267f4fd795916bf + + + CATEGORY + Root-level community summaries used to answer user queries, representing the fewest number of summaries +A category or cluster used in the analysis, representing a specific subset of the data +A category or condition used in the analysis, representing a specific subset of the data +A category or cluster used in the analysis, representing a specific subset of the data +A category representing root-level community summaries in the analysis + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + High-level community summaries used to answer user queries, representing sub-communities of C0 +A category or cluster used in the analysis, representing a specific subset of the data +A category or condition used in the analysis, representing a specific subset of the data +A category or cluster used in the analysis, representing a specific subset of the data + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + Intermediate-level community summaries used to answer user queries, representing sub-communities of C1 +A category or cluster used in the analysis, representing a specific subset of the data +A category or condition used in the analysis, representing a specific subset of the data +A category or cluster used in the analysis, representing a specific subset of the data + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + Low-level community summaries used to answer user queries, representing sub-communities of C2 +A category or cluster used in the analysis, representing a specific subset of the data +A category or condition used in the analysis, representing a specific subset of the data +A category or cluster used in the analysis, representing a specific subset of the data +A category representing low-level community summaries in the analysis + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + A text summarization method applying a map-reduce approach directly to source texts +A category or cluster used in the analysis, representing a specific subset of the data +A category or condition used in the analysis, representing a specific subset of the data +A category or cluster used in the analysis, representing a specific subset of the data +A category representing source text summarization in the analysis + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + A na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached +A category or cluster used in the analysis, representing a specific subset of the data +A baseline condition used in the analysis, representing a specific subset of the data +A category representing na¨ıve RAG in the analysis + 4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CONCEPT + The prompts used for answer generation, which are the same across all conditions with minor modifications + 973164fa90bf2b4ee267f4fd795916bf + + + DATASET + A dataset consisting of podcast transcripts used in the evaluation +A dataset consisting of podcast transcripts, used in the analysis +A dataset consisting of podcast transcripts used for analysis + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + DATASET + A dataset consisting of news articles used in the evaluation +A dataset consisting of news articles, used in the analysis +A dataset consisting of news articles used for analysis + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + Metrics used to evaluate natural language generation, including reference-based metrics and qualities of generated texts +The metrics used in the analysis to evaluate the performance of different methods + 4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + A reference to a study by Wang et al. in 2023, indicating the effectiveness of LLMs in evaluation +A reference to a study or paper authored by Wang and others in 2023 + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + A reference to a study by Zheng et al. in 2024, indicating the effectiveness of LLMs in evaluation +A reference to a study or paper authored by Zheng and others in 2024 + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + Different conditions compared in the analysis, including Graph RAG, text summarization, and semantic search RAG +Different scenarios or variables that are compared in an experiment + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + Queries made by users that are answered using different methods and conditions +Queries made by users to retrieve information + 973164fa90bf2b4ee267f4fd795916bf,e4d9b12cf2b4c691c74019eefff4fb39 + + + CONCEPT + Types of entities extracted during the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + + + METRIC + The size of the context window used in the graph indexing process, set to 600 tokens +The size of the context window used in the analysis, tested at 8k, 16k, 32k, and 64k tokens +The fixed size of the context window used for the final evaluation, set to 8k tokens + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + The process of extracting information, with 1 gleaning for the Podcast dataset and 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + TECHNOLOGY + Natural Language Generation (NLG) is a subfield of artificial intelligence that focuses on generating human-like text from data + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method where a Large Language Model (LLM) is used to compare and evaluate competing outputs + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method for evaluating the performance of Retrieval-Augmented Generation (RAG) systems, focusing on context relevance, faithfulness, and answer relevance + 322e02986c8724eedbcf3ebfa20b989c + + + PUBLICATION + A reference to a study or paper authored by Es and others in 2023 + 322e02986c8724eedbcf3ebfa20b989c + + + TOOL + A Large Language Model (LLM) used to evaluate and compare generated texts based on specific metrics + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how specifically and clearly an answer addresses a question +A metric used to evaluate the straightforwardness of the generated answers +A validity test metric used to measure the directness of responses, with naive RAG producing the most direct responses + 322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + DATA + An example of LLM-generated assessment shown in a table format + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + A specific query used in the evaluation process +A metric used to evaluate the generated responses by asking specific questions + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + ENTITY + Individuals who are well-known in the entertainment industry and are mentioned across various articles +Public figures are individuals who have gained fame or notoriety in various sectors such as entertainment, sports, and digital media. + 322e02986c8724eedbcf3ebfa20b989c,718017a4871c909420f84b85b8ba969d + + + DATASET + A collection of articles focused on the entertainment industry +A dataset consisting of articles related to the entertainment industry + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + DOMAIN + A sector that encompasses various forms of entertainment, including movies, music, and television +The entertainment industry encompasses film, television, music, sports, and digital media + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric indicating the highest level of development or achievement in a particular field + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric indicating results that are comparable to or better than those of others in the same field + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric based on evaluations made by humans + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + Metrics that require a gold standard or reference answers for evaluation + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + An evaluation method that does not require reference answers + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how relevant the generated text is to the given context + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how accurately the generated text reflects the source information + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how relevant the generated answer is to the question + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method involving multiple stages or steps + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The correct or ideal answers used as a benchmark in evaluations + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + Questions designed to help understand and make sense of complex information +A class of questions used to evaluate the performance of RAG systems +Questions designed to validate the understanding and interpretation of data + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method where two items are directly compared against each other + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + Specific metrics that are the focus of an evaluation +Specific measures used to evaluate the performance of RAG systems +Specific metrics aimed to be achieved or measured in the analysis + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A metric used as a baseline or standard for comparison + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures the accuracy and reliability of a method or result + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures the randomness or variability in a process + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The average scores obtained from multiple evaluations + 322e02986c8724eedbcf3ebfa20b989c + + + PERSON + Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to music and her high-profile personal life +Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to sports and his high-profile personal life +Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Britney Spears is a public figure frequently mentioned in entertainment articles, known for her contributions to music and her high-profile personal life +Britney Spears is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his contributions to music and his high-profile personal life +Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his contributions to the music industry. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in film and television + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in music + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in sports + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in digital media and business + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry who are involved in controversies + e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric used to determine the winner in the comparison of generated responses + e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric used to evaluate the quality of LLM-generated responses + e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + A sector within the entertainment industry that includes movies and cinema +The film sector includes public figures involved in the movie industry, including actors, directors, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + A sector within the entertainment industry that includes TV shows and series +The television sector includes public figures involved in TV shows, including actors, hosts, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + A sector within the entertainment industry that includes musical performances and recordings +The music sector includes public figures involved in the music industry, including singers, musicians, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + A sector within the entertainment industry that includes athletic events and competitions +The sports sector includes public figures involved in sports, including athletes, coaches, and sports commentators. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + A sector within the entertainment industry that includes online content and social media +The digital media sector includes public figures involved in online platforms, including influencers, content creators, and digital marketers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes stories and themes that shape culture + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes popular movements and styles + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes public conversations and debates + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes formal discussions and communications + e8c8f911135faf3ff35f24107eb3f99c + + + RESPONSE + Answer 1 provides a varied and rich response by covering a wide range of public figures from different sectors of the entertainment industry, including film, television, music, sports, gaming, and digital media. It offers insights into the contributions and influence of these figures, as well as controversies and their impact on public discourse. The answer also cites specific data sources for each mentioned figure, indicating a diverse range of evidence to support the claims. + 718017a4871c909420f84b85b8ba969d + + + RESPONSE + Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports, and relies heavily on a single source for data. It provides concise explanations for the frequent mentions of specific public figures such as Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake. +Answer 2 is a generated answer for the example question in the News article dataset + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + Naïve RAG is a method used to generate responses that directly list specific public figures who are repeatedly mentioned across various entertainment articles. +Naïve RAG is a method used to generate answers for questions in the News article dataset +A basic form of RAG that has certain drawbacks which advanced RAG systems aim to overcome +A baseline method for retrieval-augmented generation (RAG) that does not use advanced techniques + 718017a4871c909420f84b85b8ba969d,e4d9b12cf2b4c691c74019eefff4fb39,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19 + + + DATASET + A dataset consisting of news articles used for generating responses to questions about public figures in the entertainment industry. +A dataset consisting of news articles used for analysis + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + TOPIC + Controversies are events or issues involving public figures that generate public debate and impact public discourse. + 718017a4871c909420f84b85b8ba969d + + + SECTOR + The gaming sector includes public figures involved in the gaming industry, including gamers, developers, and streamers. + 718017a4871c909420f84b85b8ba969d + + + RESOURCE + Data sources are references or reports used to support claims about public figures and their influence. + 718017a4871c909420f84b85b8ba969d + + + METHOD + Assessments generated by large language models (LLMs) to evaluate the answers produced by different methods + ebf5249c888e07fedce6572a4c03f88c + + + DATASET + An example question used in the News article dataset for analysis + ebf5249c888e07fedce6572a4c03f88c + + + DATA + The datasets used in the analysis, consisting of various text sources + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + A metric used to compare the performance of different conditions in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + CATEGORY + A specific setup or scenario used in the analysis, such as C0, C1, C2, C3, TS, and SS + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + The percentage of times a condition outperformed another in the analysis +The percentage of times a particular approach or method achieves a win in a given context +Win rate is a measure used to evaluate the success rate of different methods in providing comprehensive and diverse answers + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4 + + + METRIC + The condition that performed the best across all comparisons in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + The expected win rate of a condition when compared to itself, shown as 50% for reference + 4c855404ee3d3c94aa2136f1513c666f + + + METHOD + The use of large language models (LLMs) at the time of querying, evaluated in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + METHOD + The last stage of the analysis where the best performing context window size was used + + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + + + PROCESS + The process that resulted in the creation of graphs for the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + + + STRUCTURE + A data structure consisting of nodes and edges, used to represent the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + + + METHOD + Approaches that consistently outperformed the naive RAG in comprehensiveness and diversity metrics + 36db32c37e1987e2c5863898ad882190 + + + METRIC + The number of context units, such as community summaries or text chunks, used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + + + METRIC + The number of tokens, or individual words, used in the analysis +The number of individual words used in the analysis, with the evaluation focusing on corpora in the region of 1 million tokens + 36db32c37e1987e2c5863898ad882190,92e93fc6449756c0a60200636b297f65 + METRIC + + + METRIC + The percentage of the maximum token count used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + + + METHOD + A summarization approach that is the most resource-intensive, requiring the highest number of context tokens +A method for summarizing source texts using a map-reduce approach + 36db32c37e1987e2c5863898ad882190,e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Summaries at the root level of the community hierarchy, requiring dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + + + DATASET + Source texts are the original texts used for comparison with community summaries in the analysis +Original texts from which summaries or analyses are derived + 6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39 + + + REFERENCE + A reference to a paper by Ram et al. in 2023 discussing RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + REFERENCE + A reference to a paper by Gao et al. in 2023 discussing naive RAG approaches +A reference to a publication by Gao et al. in 2023 +A paper by Gao et al. published in 2023, focusing on advanced RAG where the index is a knowledge graph + 6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + + + CATEGORY + Intermediate-level summaries are a type of community summary used in the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + CATEGORY + Root-level summaries are a type of community summary used in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + METRIC + Answer comprehensiveness is a measure used to evaluate the completeness of answers provided by different methods + 6f33a085ff3304e5994f7fbb86c881a4 + + + TECHNOLOGY + Ad-hoc LLM use refers to the spontaneous use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + + + TECHNOLOGY + Element extraction prompts are used to extract specific details in the Graph RAG index + 6f33a085ff3304e5994f7fbb86c881a4 + + + CONCEPT, TECHNOLOGY + A mathematical space in which text chunks and queries are embedded to represent similar semantics + f35de4d9fb65f1d5a392064b20545c19 + + + CONCEPT, DATA + Search inputs that are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A type of RAG system that includes patterns for iterative and dynamic cycles of interleaved retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, CONCEPT + A concept related to generation-augmented retrieval that facilitates future generation cycles + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method that facilitates future generation cycles by using self-memory + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A strategy for iterative retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A federated strategy for retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method that combines multiple concepts for summarizing multiple documents + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method for answering questions that require multiple steps or "hops" to gather information + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + An approach that involves generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A method for generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A method for answering multiple interpretations of ambiguous questions by generating a hierarchical structure + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A process that involves using LLMs to create knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A process that involves using LLMs to complete existing knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + Graphs that represent causal relationships, which can be extracted using LLMs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + REFERENCE, PUBLICATION + A reference to a publication by Cheng et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Mao et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Shao et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Wang et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Su et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Feng et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Trivedi et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Khattab et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Sarthi et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Kim et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Trajanoska et al. in 2023 +A paper by Trajanoska et al. published in 2023, focusing on using LLMs for knowledge graph creation + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Yao et al. in 2023 +A paper by Yao et al. published in 2023, focusing on using LLMs for knowledge graph completion + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + TECHNOLOGY, METHOD + A system that combines multiple concepts for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + Strategies used before the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Strategies used during the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Strategies used after the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A pattern in Modular RAG systems for iterative and dynamic cycles of retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Cycles of generation that are facilitated by self-memory in Graph RAG + f35de4d9fb65f1d5a392064b20545c19 + + + PUBLICATION + A paper by Ban et al. published in 2023, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + PUBLICATION + A paper by Zhang et al. published in 2024, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where the index is a knowledge graph, developed by Baek et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Baek et al. published in 2023, focusing on the KAPING method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where subsets of the graph structure are the objects of enquiry, developed by He et al. in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by He et al. published in 2024, focusing on the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where derived graph metrics are the objects of enquiry, developed by Zhang in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Zhang published in 2023, focusing on the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, developed by Kang et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Kang et al. published in 2023, focusing on the SURGE method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where retrieved event-plot subgraphs are serialized using narrative templates, developed by Ranade and Joshi in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Ranade and Joshi published in 2023, focusing on the FABULA method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + PUBLICATION + A paper by Wang et al. published in 2023, focusing on a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + ORGANIZATION + LangChain is an organization that supports a variety of graph databases +LangChain is an organization that developed Langchain graphs + 71f6daf11e64e5273a3847d46bf228e1,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + + + ORGANIZATION + LlamaIndex is an organization that supports a variety of graph databases +LlamaIndex is an organization that developed the LlamaIndex Knowledge Graph Index + 6cd82819982879bd164547d2773ba5c7,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + + + TECHNOLOGY + Neo4J is a graph database format supported by various RAG applications +Neo4J is an organization that developed Project NaLLM + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + + + METHOD + A method that can create and reason over knowledge graphs in Neo4J format, developed by Neo4J in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + TECHNOLOGY + NebulaGraph is a graph database format supported by various RAG applications +NebulaGraph is an organization that launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + + + METHOD + A method that can create and reason over knowledge graphs in NebulaGraph format, developed by NebulaGraph in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + METHOD + A method for comparing fabrication rates, developed by Manakul et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Manakul et al. published in 2023, focusing on the SelfCheckGPT method +A reference to the work by Manakul and colleagues published in 2023, related to SelfCheckGPT + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + PUBLICATION + + + STAKEHOLDER + Individuals who validate sensemaking questions and target metrics +Individuals who are the final users of the system or analysis + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + STAKEHOLDER + + + CONCEPT + Considerations and compromises involved in building a graph index + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METRIC + The effectiveness of RAG systems, which varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + METRIC + + + CONCEPT + Various forms of data used in RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METRIC + The scale of datasets used in RAG systems, which affects performance + 92e93fc6449756c0a60200636b297f65 + METRIC + + + PROCESS + The process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + PROCESS + + + DATASET + Collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + DATASET + + + CONCEPT + Different categories of questions used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METHOD + SelfCheckGPT is an approach used to compare fabrication rates in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method for global summarization of source texts that does not use a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + RESOURCE + The amount of computational resources allocated for a task + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The expected number of queries over the lifetime of a dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Annotations that provide detailed information about the text + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method that uses embeddings to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + RAG schemes that combine embedding-based matching with other approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Mechanisms used in map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A hierarchical organization of communities + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) for global text summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The cost associated with the number of tokens used in a text generation task + e4d9b12cf2b4c691c74019eefff4fb39 + + + TECHNOLOGY + An implementation of Graph RAG approaches using the Python programming language + e4d9b12cf2b4c691c74019eefff4fb39 + + + PERSON + A person who contributed to the work mentioned in the acknowledgements + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The rates at which fabrications occur in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The expected number of queries over the lifetime of a specific dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The benefits or value obtained from using a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Different methods related to retrieval-augmented generation that utilize graph structures + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Graph RAG approaches that operate in a more localized manner + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Annotations made on the graph to provide additional information + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Reports generated from community summaries + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + An operation that aggregates information across multiple levels of a hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A mechanism that allows for exploring detailed information by following higher-level summaries + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + The trail of information that guides users to more detailed data + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + The top-level communities in a hierarchical structure + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A graph index organized around entities + e4d9b12cf2b4c691c74019eefff4fb39 + + + TECHNOLOGY + A publicly available implementation of a technology + e4d9b12cf2b4c691c74019eefff4fb39 + + + PERSON + Alonso Guevara Fernández is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Amber Hoak is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Andrés Morales Esquivel is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Ben Cutler is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Billie Rinaldi is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Chris Sanchez is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Chris Trevino is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Christine Caggiano is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + David Tittsworth is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Dayenne de Souza is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Douglas Orbaker is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Ed Clark is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Gabriel Nieves-Ponce is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Gaudy Blanco Meneses is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Kate Lytvynets is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Katy Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Mónica Carvajal is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Nathan Evans is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Richard Ortega is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Rodrigo Racanicci is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Sarah Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Shane Solomon is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + A technical report on GPT-4 published as an arXiv preprint + 086021a89900a39bcb62036981737bfa + + + METHOD + A method for zero-shot knowledge graph question answering described in an arXiv preprint + 086021a89900a39bcb62036981737bfa + + + METHOD + A method for harnessing large language models for advanced causal discovery from data + 086021a89900a39bcb62036981737bfa + + + METHOD + A method incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Achiam is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Adler is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Agarwal is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + L. Ahmad is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + I. Akkaya is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + F. L. Aleman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + D. Almeida is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Altenschmidt is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Altman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Anadkat is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + R. Anil is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Borgeaud is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + Y. Wu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J.-B. Alayrac is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Yu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + R. Soricut is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Schalkwyk is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + A. M. Dai is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + A. Hauth is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Baek is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + A. F. Aji is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + A. Saffari is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + T. Ban is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + L. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + X. Wang is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + H. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + T. Baumel is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + M. Eyal is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + M. Elhadad is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + Baumel, T. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Eyal, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Elhadad, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Blondel, V. D. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Guillaume, J.-L. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Lambiotte, R. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Lefebvre, E. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The journal where the paper "Fast unfolding of communities in large networks" was published + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Brown, T. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Mann, B. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Ryder, N. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Subbiah, M. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Kaplan, J. D. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dhariwal, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Neelakantan, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Shyam, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Sastry, G. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Askell, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The conference where the paper "Language models are few-shot learners" was presented +The conference where the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" was presented +The journal where the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" was published + 58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,b1bbda43309e8e0e2175ea034aa88e13 + + + PERSON + Cheng, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Luo, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Chen, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Liu, L. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Zhao, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory"Zhao, D. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + PERSON + + + PERSON + Yan, R. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dang, H. T. is an author of the paper "Duc 2005: Evaluation of question-focused summarization systems" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The conference where the paper "Duc 2005: Evaluation of question-focused summarization systems" was presented + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Es, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + James, J. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Espinosa-Anke, L. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Schockaert, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Feng, Z. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Feng, X. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Yang, M. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Qin, B. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Fortunato, S. is an author of the paper "Community detection in graphs" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The journal where the paper "Community detection in graphs" was published + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Gao, Y. is an author of the paper "Retrieval-augmented generation" +Gao, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Xiong, Y. is an author of the paper "Retrieval-augmented generation" +Xiong, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Gao, X. is an author of the paper "Retrieval-augmented generation" +Gao, X. is an author of the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Jia, K. is an author of the paper "Retrieval-augmented generation" +Jia, K. is an author of the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Pan, J. is an author of the paper "Retrieval-augmented generation" +Pan, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Bi, Y. is an author of the paper "Retrieval-augmented generation" +Bi, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dai, Y. is an author of the paper "Retrieval-augmented generation" +Dai, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Sun, J. is an author of the paper "Retrieval-augmented generation" +Sun, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Wang, H. is an author of the paper "Retrieval-augmented generation" +Wang, H. is an author of the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Goodwin, T. R. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Savery, M. E. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Demner-Fushman, D. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + CONFERENCE + COLING (International Conference on Computational Linguistics) is the conference where the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" was presented + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + He, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Tian, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Sun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Chawla, N. V. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Laurent, T. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + LeCun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Bresson, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Hooi, B. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jacomy, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Venturini, T. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Heymann, S. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Bastian, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PUBLICATION + PLOS ONE is the journal where the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" was published + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jin, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Yu, Z. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jiao, P. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Pan, S. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + He, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Wu, J. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Philip, S. Y. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Zhang, W. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PUBLICATION + IEEE Transactions on Knowledge and Data Engineering is the journal where the paper "A survey of community detection approaches: From statistical modeling to deep learning" was published + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Kang, M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Kwak, J. M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Baek, J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Hwang, S. J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Khattab, O. is an author of the paper mentioned in the text +Khattab, O. is an author of the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Santhanam, K. is an author of the paper mentioned in the text +Santhanam, K. is an author of the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Li, X. L. is an author of the paper mentioned in the text +Li, X. L. is an author of the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hall, D. is an author of the paper mentioned in the text +Hall, D. is an author of the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Liang, P. is an author of the paper mentioned in the text +Liang, P. is an author of the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" +Liang, P. is an author of the paper "Lost in the middle: How language models use long contexts" + 00e8e4e881bd0862022f4dfc913b900b,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Potts, C. is an author of the paper mentioned in the text +Potts, C. is an author of the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Zaharia, M. is an author of the paper mentioned in the text +Zaharia, M. is an author of the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kim, G. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kim, S. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Jeon, B. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Park, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kang, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Klein, G. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Moon, B. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hoffman, R. R. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The journal where the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" were published + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Koesten, L. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Gregory, K. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Groth, P. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Simperl, E. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The journal where the paper "Talking datasets–understanding data sensemaking behaviours" was published + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kuratov, Y. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Bulatov, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Anokhin, P. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Sorokin, D. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Sorokin, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Burtsev, M. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + TECHNOLOGY + Langchain graphs is a technology developed by LangChain + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Laskar, M. T. R. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" +Laskar, M. T. R. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hoque, E. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" +Hoque, E. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Huang, J. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The conference where the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" was presented + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + arXiv preprint refers to a preprint of a paper that is available on the arXiv repository + 71f6daf11e64e5273a3847d46bf228e1 + + + EVENT + The 33rd Canadian Conference on Artificial Intelligence, held in Ottawa, ON, Canada, from May 13–15, 2020 + 6cd82819982879bd164547d2773ba5c7 + + + EVENT + The 2020 edition of the Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + PUBLISHER + Springer is the publisher of the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Huang, J. X. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + PUBLICATION + The journal where the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" was published + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lewis, P. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Perez, E. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Piktus, A. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Petroni, F. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks"Petroni, F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + PERSON + + + PERSON + Karpukhin, V. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Goyal, N. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Küttler, H. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lewis, M. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Yih, W.-T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Rocktäschel, T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, N. F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lin, K. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Hewitt, J. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Paranjape, A. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Bevilacqua, M. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, Y. is an author of the paper "Hierarchical transformers for multi-document summarization" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lapata, M. is an author of the paper "Hierarchical transformers for multi-document summarization" +Lapata, M. is an author of the paper "Text summarization with latent queries" + 6cd82819982879bd164547d2773ba5c7,fc4b27d64f055b7fc30176ba110dd02e + + + TECHNOLOGY + LlamaIndex Knowledge Graph Index is a technology developed by LlamaIndex + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Manakul, P. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liusie, A. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Gales, M. J. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Mao, Y. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + He, P. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, X. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Shen, Y. is an author of the paper "Generation-augmented retrieval for open-domain question answering" +Shen, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Gao, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Han, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Chen, W. is an author of the paper "Generation-augmented retrieval for open-domain question answering" +Chen, W. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" +Chen, W. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Martin, S. is an author of the paper "Openord: An open-source toolbox for large graph" +Martin, S. is an author of the paper "Openord: An open-source toolbox for large graph layout" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Brown, W. M. is an author of the paper "Openord: An open-source toolbox for large graph" +Brown, W. M. is an author of the paper "Openord: An open-source toolbox for large graph layout" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Klavans, R. is an author of the paper "Openord: An open-source toolbox for large graph" +Klavans, R. is an author of the paper "Openord: An open-source toolbox for large graph layout" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Boyack, K. is an author of the paper "Openord: An open-source toolbox for large graph" +Boyack, K. is an author of the paper "Openord: An open-source toolbox for large graph layout" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + EVENT + The conference where the paper "Openord: An open-source toolbox for large graph layout" was presented + 833e7d67dcd30790b26b71c9b5306f6b + EVENT + + + TECHNOLOGY + GPT-4 is a large language model used in Microsoft's study on scientific discovery + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + + + TECHNOLOGY + Project NaLLM is a project developed by Neo4J + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + + + PERSON + Newman, M. E. is the author of the paper "Modularity and community structure in networks" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PUBLICATION + The journal where the paper "Modularity and community structure in networks" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + + + PERSON + Ram, O. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Levine, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Dalmedigos, I. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Muhlgay, D. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shashua, A. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Leyton-Brown, K. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shoham, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PUBLICATION + The journal where the paper "In-context retrieval-augmented language models" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + + + PERSON + Ranade, P. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Joshi, A. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Sarthi, P. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Abdullah, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Tuli, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Khanna, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Goldie, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Manning, C. D. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" +Manning, C. D. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + 833e7d67dcd30790b26b71c9b5306f6b,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Scott, K. is associated with "Behind the Tech" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shao, Z. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Gong, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Huang, M. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" +Huang, M. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + + + PERSON + Duan, N. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" +Duan, N. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + + + PERSON + Su, D. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Xu, Y. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" +Xu, Y. is an author of the paper "Text summarization with latent queries" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yu, T. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Siddique, F. B. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Barezi, E. J. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Fung, P. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Tang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Yang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Touvron, H. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Martin, L. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Stone, K. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Albert, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Almahairi, A. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Babaei, Y. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bashlykov, N. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Batra, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bhargava, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bhosale, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Traag, V. A. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Waltman, L. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Van Eck, N. J. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PUBLICATION + Scientific Reports is the journal where the paper "From Louvain to Leiden: guaranteeing well-connected communities" was published + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trajanoska, M. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Stojanov, R. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trajanov, D. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trivedi, H. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Balasubramanian, N. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Khot, T. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Sabharwal, A. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Wang, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Liang, Y. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Meng, F. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Sun, Z. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Shi, H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Li, Z. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Li, Z. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Li, Z. is an author of the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + 8d87efac8c50cf20cdf26bf61e5e2035,b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Xu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Xu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Qu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Qu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhou, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Zhou, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wang, S. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Wang, S. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Khramtsova is an author mentioned in the text + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Khramtsova, E. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhuang, S. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" +Zhuang, S. is an author of the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zuccon, G. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wang, Y. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Lipka, N. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Rossi, R. A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Siu, A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhang, R. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Derr, T. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yang, Z. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Qi, P. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhang, S. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Bengio, Y. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Cohen, W. W. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Salakhutdinov, R. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + CONFERENCE + The conference where the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" was presented + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yao, J.-g. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wan, X. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Xiao, J. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PUBLICATION + The journal where the paper "Recent advances in document summarization" was published + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yao, L. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models"Yao, L. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Peng, J. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Mao, C. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Luo, Y. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhang, J. is an author of the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhang, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Gan, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Wang, C. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zheng, L. is an author of the paper "Exploring large language models for knowledge graph completion" +Zheng, L. is an author of the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Chiang, W.-L. is an author of the paper "Exploring large language models for knowledge graph completion" +Chiang, W.-L. is an author of the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Sheng, Y. is an author of the paper "Exploring large language models for knowledge graph completion" +Sheng, Y. is an author of the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Wu, Z. is an author of the paper "Exploring large language models for knowledge graph completion" +Wu, Z. is an author of the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhuang, Y. is an author of the paper "Exploring large language models for knowledge graph completion" +Zhuang, Y. is an author of the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Lin, Z. is an author of the paper "Exploring large language models for knowledge graph completion" +Lin, Z. is an author of the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Li, D. is an author of the paper "Exploring large language models for knowledge graph completion" +Li, D. is an author of the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Xing, E. is an author of the paper "Exploring large language models for knowledge graph completion" +Xing, E. is an author of the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + TECHNOLOGY + Chatbot Arena is a platform or tool used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Darren Edge and Ha Trinh co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Steven Truitt and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Steven Truitt is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Jonathan Larson is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 4.0 + RAG uses LLMs to retrieve relevant information from external knowledge sources +LLM uses RAG to generate and assess text +RAG is used to augment the capabilities of LLMs + e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 7.0 + Graph RAG combines the strengths of RAG with graph-based text indexing +Graph RAG is a specific approach to RAG +Graph RAG is a specific implementation of RAG +Graph RAG is an implementation of RAG +Graph RAG is a method that uses the natural modularity of graphs to partition data for global summarization + 21e52bc06a82796b1f4bcd73edda1f2a,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Query-Focused Summarization is a task that RAG fails to address effectively + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + RAG retrieves relevant information from an external knowledge source + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Naive RAG is a specific implementation of RAG + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Ram et al., 2023 discusses RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Naïve RAG is a basic form of RAG + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Modular RAG is an advanced form of RAG + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used in various RAG tasks such as knowledge graph creation and completion + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Trajanoska et al. discusses using LLMs for knowledge graph creation, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Yao et al. discusses using LLMs for knowledge graph completion, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Ban et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Zhang et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Gao et al. discusses advanced RAG where the index is a knowledge graph + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + KAPING is a method where the index is a knowledge graph, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + G-Retriever is a method where subsets of the graph structure are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Graph-ToolFormer is a method where derived graph metrics are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + SURGE is a method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + FABULA is a method where retrieved event-plot subgraphs are serialized using narrative templates, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Wang et al. discusses a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Sensemaking questions are used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The evaluation of RAG systems focuses on corpora in the region of 1 million tokens + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Trade-offs are considerations involved in building a graph index for RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + A graph index is a data structure used in RAG systems to organize and retrieve information + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Performance of RAG systems varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Different data types are used in RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Dataset sizes affect the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Evaluation is the process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Corpora are collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Different question types are used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Target metrics are specific measures used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 4.0 + Graph RAG uses LLMs to build a graph-based text index +LLM is used in the Graph RAG approach to generate summaries and answer queries +LLM uses Graph RAG to provide a comprehensive overview of public figures in the entertainment industry +LLMs are used in Graph RAG to analyze and generate text based on retrieved information and queries + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Text chunks are processed using LLM to extract elements of a graph index + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM is used to extract elements of a graph index from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Few-shot examples are provided to the LLM for in-context learning to tailor the extraction prompt +Few-shot examples are used to improve the performance of the LLM in specialized domains + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM extracts named entities from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Kuratov et al. (2024) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Liu et al. (2023) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM prompts are instructions given to the LLM for extracting elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Recall degradation occurs with longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + The extraction process involves using LLM to identify and extract elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Default prompt is the standard set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Secondary extraction prompt is an additional set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + The LLM uses covariate prompts to extract additional attributes associated with detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM uses multiple rounds of gleanings to ensure no entities are missed + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Logit bias is used to force a yes/no decision from the LLM during entity extraction + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM extracts element instances from source texts + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM detects and summarizes communities of entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + LLM generates intermediate answers and scores for each chunk + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + LLM generates a helpfulness score for each answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + LLM is used to generate questions for evaluating the Podcast Transcripts dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + LLM is used to generate questions for evaluating the News Articles dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + LLM uses Naive RAG to list public figures mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + LLM-generated responses are evaluated using assessment metrics + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + LLM-generated responses are evaluated using specific questions + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Ad-hoc LLM use involves the use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + LLMs are used for knowledge graph creation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph completion + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for the extraction of causal graphs + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph creation as per Trajanoska et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph completion as per Yao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for the extraction of causal graphs as per Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is proposed as a method to combine the strengths of RAG and QFS + e8d83e6e7a7c0f57b218cef24976b745 + + + 7.0 + Community summaries are used in the Graph RAG approach to generate partial responses +Community summaries are generated in the Graph RAG approach +Graph RAG is used to compare community summaries to source texts +Graph RAG uses community summaries to improve answer comprehensiveness and diversity +Graph RAG uses community summaries as a kind of self-memory +Graph RAG uses summaries of root-level communities in an entity-based graph index + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is designed to handle global sensemaking questions over large datasets + e8d83e6e7a7c0f57b218cef24976b745 + + + 2.0 + Graph RAG is implemented in Python +Graph RAG is implemented using Python + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The open-source implementation of Graph RAG will be available at this URL + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Graph RAG uses an entity knowledge graph to index text + e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG improves the comprehensiveness of generated answers +Comprehensiveness is a target quality used to evaluate the Graph RAG approach +Graph RAG is evaluated for comprehensiveness + 21e52bc06a82796b1f4bcd73edda1f2a,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG improves the diversity of generated answers +Diversity is a target quality used to evaluate the Graph RAG approach +Diversity is used to evaluate the performance of Graph RAG + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG uses a knowledge graph for global summarization +Graph RAG uses a knowledge graph for global summarization + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Community detection algorithms are used in the Graph RAG approach to partition graphs + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Podcast transcripts are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + News articles are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 2.0 + Empowerment is a target quality used to evaluate the Graph RAG approach +Empowerment is used to evaluate Graph RAG's ability to help users reach an informed understanding + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Graph RAG is compared to naive RAG in the evaluation +Graph RAG outperformed naive RAG on comprehensiveness and diversity + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Graph RAG is compared to global map-reduce summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Query-focused summarization is a method used in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Activity-centered sensemaking questions are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 2.0 + Token costs are measured to evaluate the efficiency of the Graph RAG approach +Token costs are a consideration in the performance of the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Data flow describes the high-level process of the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 3.0 + Design parameters influence the Graph RAG approach and pipeline +Design parameters are key settings in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Graph RAG uses global summarization to summarize information from a large dataset + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG aims to answer specific queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG uses a corpus for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Activity-centered sensemaking is used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Real-world datasets are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG is compared to source text summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Low-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Intermediate-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + High-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + The Graph RAG approach involves a specific pipeline for processing and summarizing text + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Techniques are specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Implementation details are specific configurations used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Graph RAG is a specific implementation of RAG systems + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + Graph RAG uses root-level community summaries (C0) to answer user queries +C0 represents root-level community summaries in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses high-level community summaries (C1) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses intermediate-level community summaries (C2) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 2.0 + Graph RAG uses low-level community summaries (C3) to answer user queries +C3 represents low-level community summaries in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + 2.0 + Graph RAG is one of the conditions compared in the analysis +Graph RAG compares multiple conditions + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses different levels of graph communities to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The Graph RAG mechanism uses an LLM evaluator for head-to-head comparison + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Graph RAG is a multi-stage mechanism + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Graph RAG mentions Taylor Swift as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Travis Kelce as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Britney Spears as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Justin Timberlake as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG is determined to be the winner based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Answer 1 was generated using the Graph RAG method, which provides a comprehensive and structured overview of public figures across various sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Graph RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Graph RAG is compared with source texts for answer comprehensiveness and diversity + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + TS represents source text summarization in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Root-level summaries are used in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Answer comprehensiveness is used to evaluate the performance of Graph RAG + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Win rate is used to measure the success rate of Graph RAG in providing comprehensive and diverse answers + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Element extraction prompts are used in Graph RAG to retain specific details in the index + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Graph RAG incorporates the concept of self-memory + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of iterative retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of federated retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts used in multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts used in multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG uses a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of a tree of clarifications + f35de4d9fb65f1d5a392064b20545c19 + + + 3.0 + Graph RAG uses a self-generated graph index +Graph RAG uses a graph index + e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Gao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Cheng et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Mao et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Shao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Wang et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Su et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Feng et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Trivedi et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Khattab et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Sarthi et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Kim et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG generates community answers in parallel + f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is compared to a graph-free approach for global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG is compared to map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses rich text annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses a hierarchical community structure + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can operate using embedding-based matching + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can be part of hybrid RAG schemes + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can employ map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can extend operations across the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Alonso contributed to the work on Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG includes local graph RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses an entity-based graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + 2.0 + NebulaGraph launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Community summaries are used to generate partial responses + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Community summaries are created from graph communities + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Community answers are created from community summaries +Community answers are generated from community summaries + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Domain-tailored summarization is used to create community summaries + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Community descriptions are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Partial answers are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Community summaries are created for each level in the hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Community summaries are useful for understanding the global structure and semantics of the dataset + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Community summaries are used to answer global queries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are generated from root communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are generated from sub-communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are added to the LLM context window until the token limit is reached + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Global answers are generated from community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The level of summary detail affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The scope of information affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are used for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Community summaries are divided into chunks of pre-specified token size + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Summary detail and scope affect the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are divided into chunks + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Community summaries are prepared to answer user queries + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Intermediate answers are generated from community summaries + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Community summaries are part of the graph community hierarchy + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Community summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Summaries of root-level communities are used in Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Global sensemaking questions are evaluated over datasets in the 1 million token range + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Global sensemaking questions are directed at an entire text corpus + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The Python-based implementation of Graph RAG approaches will be available at this URL + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Query-focused summarization is used to produce the global answer + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Map-reduce is used for query-focused summarization of an entire corpus + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Query-focused summarization is used for answering global queries + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + An entity knowledge graph is derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + + + 2.0 + Text chunks are extracted from source documents +Text chunks are extracted from source documents for processing in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Intermediate-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Low-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Document corpus consists of source documents being processed + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Partial responses are summarized to generate a final response + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The LLM evaluator assesses answers based on the comprehensiveness metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Naive RAG is evaluated for comprehensiveness + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Comprehensiveness is a metric used to determine the decision + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) was universally better for comprehensiveness + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The final evaluation prioritized comprehensiveness in answers + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Global approaches achieved higher comprehensiveness win rates + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The LLM evaluator assesses answers based on the diversity metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on diversity + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The final evaluation prioritized diversity in answers + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Global approaches achieved higher diversity win rates + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Human endeavors rely on sensemaking to understand and reason about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Human endeavors rely on analyzing document collections for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + LLMs are used to automate sensemaking in complex domains + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Microsoft uses LLMs for automating sensemaking in scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Ranade uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Joshi uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + LLM prompts are used to tailor the responses of large language models + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Ranade and Joshi discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + GPT is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Llama is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Gemini is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Kuratov et al., 2024, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Liu et al., 2023, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Sensemaking is applied in the domain of scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Sensemaking is applied in the domain of intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Klein defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Klein et al. defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Element instances are extracted from text chunks +Element instances are extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Entity references are extracted from text chunks during processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Element summaries are created from element instances +Element instances are converted into element summaries by the LLM + 2c6ed90897310eea2f28e33fff1c32b0,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Covariates are additional attributes associated with extracted element instances + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Domain-tailored summarization is used to create element summaries + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Element summaries include descriptions of entity nodes + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries include descriptions of relationship edges + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries include descriptions of claim covariates + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries are used to understand the structure and semantics of graph communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Element summaries include descriptions of nodes + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Element summaries include descriptions of edges + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Element summaries include descriptions of covariates + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Sub-community summaries are used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Community detection is used to identify graph communities +Graph communities are identified through community detection + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Global answer is created from community answers + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Global answers are generated in response to user queries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Global answer is generated by sorting intermediate answers based on helpfulness scores + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Intermediate answers are combined to form the global answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + The final context window is used to generate the global answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Graph RAG pipeline operates at indexing time + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Graph RAG pipeline operates at query time + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Nodes are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Edges are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Covariates are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Leiden method is used in the graph RAG pipeline for community detection + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Graph RAG pipeline uses the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + The Podcast dataset graph consists of 8564 nodes + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The News dataset graph consists of 15754 nodes + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The Podcast dataset graph consists of 20691 edges + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The News dataset graph consists of 19520 edges + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Traag contributed to the development of the Leiden method + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Traag et al. developed the Leiden method +Traag et al. are the authors of the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Leiden is a type of community detection algorithm +Leiden is a specific community detection algorithm used in the pipeline + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Leiden is known for its ability to recover hierarchical community structures efficiently + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The Leiden algorithm is used to detect graph communities in the MultiHop-RAG + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Figure 3 shows graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Lewis contributed to the development of the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Lewis et al. developed the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Lewis et al., 2020, are the authors who established the RAG approach + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Kevin Scott is the CTO of Microsoft + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + Microsoft conducted a study on the impact of large language models on scientific discovery using GPT-4 + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Preprint is available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Baumel, T. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Elhadad, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Es, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + James, J. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Espinosa-Anke, L. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Schockaert, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, Z. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, X. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Zhao, D. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Yang, M. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Qin, B. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + LangChain is an organization that has published on arXiv + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Wang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zuccon, G. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, R. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Derr, T. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Xu, Y. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lapata, M. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, J. published the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Gan, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yao, L. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, C. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Chiang, W.-L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Sheng, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wu, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lin, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Li, D. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Xing, E. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Preprint is classified under cs.CL on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Preprint was submitted on 24 Apr 2024 + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Preprint has the identifier 2404.16130v1 on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Community detection results in the partition of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The pipeline includes a step for community detection + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 2.0 + Dang, 2006, is the author who established the QFS approach + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Baumel et al., 2018, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Laskar et al., 2020, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Yao et al., 2017, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Goodwin et al., 2020, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Laskar et al., 2022, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Liu and Lapata, 2019, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Achiam et al., 2023, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Brown et al., 2020, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Touvron et al., 2023, are the authors who worked on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Anil et al., 2023, are the authors who worked on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Modularity is an inherent quality of knowledge graphs + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Brown et al. (2020) discuss in-context learning with few-shot examples + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Kuratov et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Liu et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Louvain is a type of community detection algorithm + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Community detection algorithms are used to partition the graph index into communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Fortunato has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Jin et al. have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + HotPotQA dataset is used to evaluate the entity extraction prompt with gpt-4-turbo + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Yang et al. (2018) introduced the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Yang et al. are the authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + GPT-4-Turbo was tested with varying context window sizes + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Tech journalist uses podcast transcripts to look for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + + + 3.0 + Kevin Scott's conversations are part of the podcast transcripts +Kevin Scott is a participant in the podcast conversations compiled in the Podcast Transcripts dataset + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Technology leaders participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + RAG systems are used to evaluate the Podcast Transcripts dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + C0 is a category used in the analysis of podcast transcripts +C0 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C1 is a category used in the analysis of podcast transcripts +C1 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C2 is a category used in the analysis of podcast transcripts +C2 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C3 is a category used in the analysis of podcast transcripts +C3 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + TS is a category used in the analysis of podcast transcripts +TS is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + SS is a category used in the analysis of podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Units are used to measure the context in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Both are datasets used in the analysis + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Educator uses news articles to incorporate current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + RAG systems are used to evaluate the News Articles dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + C0 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Map-reduce is the method used in the text summarization condition + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The LLM evaluator assesses answers based on the empowerment metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on empowerment + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Empowerment has an average win rate of 51.3% + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Naive RAG mentions Taylor Swift as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Travis Kelce as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Britney Spears as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Justin Timberlake as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG is determined to be the loser based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Global approaches consistently outperformed the naive RAG + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Naive RAG produces the most direct responses + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + SS represents naive RAG in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Gao et al., 2023 discusses naive RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Community partitions enable divide-and-conquer global summarization + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Global summarization can be performed using a graph-free approach + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Source texts are used in global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Final global answer is generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Short descriptions are used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Low-level community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + The use of rich descriptive text for homogeneous nodes in a graph index aligns with the capabilities of LLMs + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Graph indexes differ from typical knowledge graphs in their use of rich descriptive text instead of concise knowledge triples + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The graph index supports condition C0 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C1 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C2 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C3 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index was created using generic prompts for entity and relationship extraction + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Few-shot examples tailored to the domain of the data were used in the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The decision to build a graph index depends on the expected number of lifetime queries per dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The decision to build a graph index depends on the value obtained from it + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The decision to build a graph index depends on the value obtained from other graph-related RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Recall measures the completeness of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Precision measures the accuracy of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to tailor the default prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to tailor the secondary extraction prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of science + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of medicine + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of law + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + A single extraction round is part of the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Domain refers to the specific area of knowledge of the document corpus + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Covariate prompts are used to extract claims linked to detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Source text span is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Start date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + End date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Description is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Subject is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Object is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Communities of entities help manage variations in a noisy graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Common entities are described using rich descriptive text for homogeneous nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + LLMs are used to generate metrics for evaluating natural language generation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Wang et al. (2023) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Zheng et al. (2024) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Relationship edges connect homogeneous nodes in a graph + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Edge weights represent the normalized counts of detected relationship instances on relationship edges + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Each level of the hierarchical community structure provides a community partition + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 2.0 + Root communities are the top-level communities in a hierarchical community structure +Root communities are part of the hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Sub-communities are lower-level communities in a hierarchical community structure +Sub-communities are part of the hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community levels are part of the hierarchical community structure + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The Leiden algorithm is used to detect communities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + OpenORD is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Force Atlas 2 is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Nodes represent entities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Edges represent connections between nodes in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Covariates are variables linked to nodes and edges in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Tang and Yang are the authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Questions are generated based on the target datasets + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + N represents the number of test questions per dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Root communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Level 0 represents the root-level communities in the hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Reports provide detailed information about specific subtopics within sub-communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Sub-communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Level 1 represents sub-communities within the root-level communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Partitions can be organized into a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Level 0 is the root level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Level 1 is a sub-level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The token limit defines the maximum number of tokens in the LLM context window + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Prominence is used to prioritize community edges + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Combined source and target node degree is used to measure prominence + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Chunks are divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Helpfulness scores are assigned to intermediate answers + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in episodes dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in how guests perceive the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in discussions about the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in discussions about collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in how news articles address the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in highlighting the importance of health literacy through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + The size of the context window and the prompts used for answer generation are the same across all conditions + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The task is an activity or goal that the user aims to achieve + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Questions are generated based on the user's task + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Datasets were used in combination with questions for the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Questions were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + + + 2.0 + Zheng et al. are the authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Zheng, L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Xing, E. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + MT-Bench and Chatbot Arena are both tools used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 2.0 + Koesten et al. authored a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + Xu and Lapata authored a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Text summarization method applies a map-reduce approach directly to source texts (TS) + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Text summarization is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Semantic search RAG is a na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached (SS) + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Semantic search RAG is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C0 uses root-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C0 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C0 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C0 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 uses high-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C1 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C1 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C1 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 uses intermediate-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C2 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C2 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C2 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 uses low-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C3 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C3 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C3 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + TS is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + SS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The graph indexing process used 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + A graph was created for the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Intermediate-level summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + The graph indexing process used 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + A graph was created for the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Datasets were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Wang et al., 2023a discusses the state-of-the-art results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Zheng et al., 2024 discusses the competitive results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Zheng et al., 2024 discusses the LLM-as-a-judge method + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Embedding-based matching is used to match user queries + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Query-time LLM use was evaluated with different context window sizes + 4c855404ee3d3c94aa2136f1513c666f + + + 2.0 + The final evaluation used a fixed context window size of 8k tokens +A fixed context window size of 8k tokens was used for the final evaluation + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Natural Language Generation achieves state-of-the-art results + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation achieves competitive results + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation is compared against human judgements + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation can generate reference-based metrics + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation can measure qualities in a reference-free style + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Es et al., 2023 discusses the RAGAS method + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates context relevance + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates faithfulness + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates answer relevance + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator assesses answers based on the directness metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Table 2 shows an example of LLM-generated assessment + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses a head-to-head comparison approach + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator assesses answers based on target metrics + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses a control metric for validity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator accounts for stochasticity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses mean scores from multiple comparisons + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Directness is used to evaluate the straightforwardness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The question asks about public figures mentioned in entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Public figures are repeatedly mentioned across various entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Answer 1 covers a wide range of public figures from different sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Controversies involve public figures and impact public discourse. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Entertainment articles cover topics related to the entertainment industry + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Taylor Swift is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Travis Kelce is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Britney Spears is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Justin Timberlake is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Taylor Swift is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Travis Kelce is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Britney Spears is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Justin Timberlake is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Actors and Directors are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Musicians and Executives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Athletes and Coaches are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Influencers and Entrepreneurs are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Public Figures in Controversy are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Film is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Television is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Music is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Sports is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Digital Media is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Cultural Narratives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Trends are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Social Discussions are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Public Discourse is a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Reference-based metrics require gold standard answers + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Gold standard answers are lacking for sensemaking questions + 322e02986c8724eedbcf3ebfa20b989c + + + 3.0 + End users validate sensemaking questions and target metrics +Sensemaking questions are validated with end users + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Target metrics are validated with end users + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The control metric is used as an indicator of validity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Taylor Swift is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Taylor Swift is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Travis Kelce is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Travis Kelce is a public figure in the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Britney Spears is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Britney Spears is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Justin Timberlake is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Justin Timberlake is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the film sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the television sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on public figures primarily from the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on public figures primarily from the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the digital media sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 cites specific data sources from the News article dataset for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 provides insights into controversies involving public figures and their impact on public discourse. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the gaming sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 cites specific data sources for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 was generated using the Naïve RAG method, which directly lists specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d + + + 2.0 + Answer 2 relies heavily on a single source from the News article dataset for data. +Answer 2 is a generated answer for a question in the News article dataset + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Answer 2 relies heavily on a single data source. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Naïve RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The global approach to Graph RAG shows improvements over naïve RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + LLM-generated assessments are used to evaluate the answers produced for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Example question is part of the News article dataset used for analysis + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Head-to-head win rate percentages were used to compare different conditions + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Win rate percentages were used to measure the performance of different conditions + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The overall winner per dataset and metric was determined for each condition + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Self-win rates were shown as the expected 50% for each condition + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The indexing process resulted in the creation of graphs + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Map-reduce summarization requires the highest number of context tokens + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Root-level community summaries require dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + + + 2.0 + Queries are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Self-memory is related to generation-augmented retrieval + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + CAiRE-COVID is a system for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + ITRG is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + IR-CoT is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + DSP is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + RAPTOR is a method for generating a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + The paper by Baek et al. discusses the KAPING method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by He et al. discusses the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Zhang discusses the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Kang et al. discusses the SURGE method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Ranade and Joshi discusses the FABULA method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Both LangChain and LlamaIndex support a variety of graph databases + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LangChain supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LangChain supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 1.0 + LangChain developed Langchain graphs + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + LlamaIndex supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LlamaIndex supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + NaLLM is a method that can create and reason over knowledge graphs in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Neo4J developed Project NaLLM + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + GraphRAG is a method that can create and reason over knowledge graphs in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Manakul et al. discusses the SelfCheckGPT method + 92e93fc6449756c0a60200636b297f65 + + + 1.0 + SelfCheckGPT is an approach mentioned in the work by Manakul et al., 2023 + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + SelfCheckGPT is used to compare fabrication rates + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Embedding-based matching is used to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Hybrid RAG schemes combine embedding-based matching against community reports + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The roll-up operation can be extended using map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The drill down mechanism follows the information scent in the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The global approach to Graph RAG achieves competitive performance at a fraction of the token cost + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The open-source implementation of Graph RAG approaches is Python-based + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The drill down mechanism follows the information scent + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Alonso Guevara Fernández and Amber Hoak both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Sarah Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Shane Solomon both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Adler co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Altenschmidt and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Altenschmidt and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Altman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and S. Borgeaud co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Schalkwyk and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Schalkwyk and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + A. M. Dai and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Baek and A. F. Aji co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Baek and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + A. F. Aji and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + T. Ban and L. Chen co-authored the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + 1.0 + Baumel, T. and Eyal, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Baumel, T. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Baumel, T. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Elhadad, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Guillaume, J.-L. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Guillaume, J.-L. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Guillaume, J.-L. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Lambiotte, R. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Mann, B. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Zhao, D. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Es, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + James, J. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Espinosa-Anke, L. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Schockaert, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, Z. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, X. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Yang, M. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Qin, B. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Gao, Y. and Xiong, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Dai, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Dai, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Sun, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Goodwin, T. R. and Savery, M. E. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Goodwin, T. R. and Demner-Fushman, D. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + 2.0 + Khattab, O. and Santhanam, K. co-authored a paper mentioned in the text +Khattab, O. and Santhanam, K. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Li, X. L. co-authored a paper mentioned in the text +Khattab, O. and Li, X. L. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Hall, D. co-authored a paper mentioned in the text +Khattab, O. and Hall, D. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Liang, P. co-authored a paper mentioned in the text +Khattab, O. and Liang, P. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Potts, C. co-authored a paper mentioned in the text +Khattab, O. and Potts, C. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Zaharia, M. co-authored a paper mentioned in the text +Khattab, O. and Zaharia, M. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Khattab, O. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Li, X. L. co-authored a paper mentioned in the text +Santhanam, K. and Li, X. L. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Hall, D. co-authored a paper mentioned in the text +Santhanam, K. and Hall, D. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Liang, P. co-authored a paper mentioned in the text +Santhanam, K. and Liang, P. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Potts, C. co-authored a paper mentioned in the text +Santhanam, K. and Potts, C. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Zaharia, M. co-authored a paper mentioned in the text +Santhanam, K. and Zaharia, M. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Santhanam, K. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Hall, D. co-authored a paper mentioned in the text +Li, X. L. and Hall, D. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Liang, P. co-authored a paper mentioned in the text +Li, X. L. and Liang, P. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Potts, C. co-authored a paper mentioned in the text +Li, X. L. and Potts, C. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Zaharia, M. co-authored a paper mentioned in the text +Li, X. L. and Zaharia, M. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Li, X. L. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Liang, P. co-authored a paper mentioned in the text +Hall, D. and Liang, P. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Potts, C. co-authored a paper mentioned in the text +Hall, D. and Potts, C. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Zaharia, M. co-authored a paper mentioned in the text +Hall, D. and Zaharia, M. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hall, D. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Liang, P. and Potts, C. co-authored a paper mentioned in the text +Liang, P. and Potts, C. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Liang, P. and Zaharia, M. co-authored a paper mentioned in the text +Liang, P. and Zaharia, M. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Liang, P. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Potts, C. and Zaharia, M. co-authored a paper mentioned in the text +Potts, C. and Zaharia, M. co-authored the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Potts, C. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Zaharia, M. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Kim, S. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Park, J. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Park, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kang, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. and Moon, B. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Moon, B. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Moon, B. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoffman, R. R. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Gregory, K. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Groth, P. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Groth, P. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Simperl, E. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kuratov, Y. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Bulatov, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Anokhin, P. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Sorokin, D. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Sorokin, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Burtsev, M. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Laskar, M. T. R. and Hoque, E. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" +Laskar, M. T. R. and Hoque, E. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Laskar, M. T. R. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Hoque, E. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoque, E. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoque, E. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Hoque, E. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Huang, J. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + The 33rd Canadian Conference on Artificial Intelligence is also known as Canadian AI 2020 + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Springer published the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Huang, J. X. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Perez, E. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Karpukhin, V. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Xu, Y. and Lapata, M. co-authored the paper "Text summarization with latent queries" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Huang, M. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Duan, N. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 3.0 + Martin, S. and Brown, W. M. co-authored the paper "Openord: An open-source toolbox for large graph" +Martin, S. and Brown, W. M. co-authored the paper "Openord: An open-source toolbox for large graph layout" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Martin, S. and Klavans, R. co-authored the paper "Openord: An open-source toolbox for large graph" +Martin, S. and Klavans, R. co-authored the paper "Openord: An open-source toolbox for large graph layout" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Martin, S. and Boyack, K. co-authored the paper "Openord: An open-source toolbox for large graph" +Martin, S. and Boyack, K. co-authored the paper "Openord: An open-source toolbox for large graph layout" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Brown, W. M. and Klavans, R. co-authored the paper "Openord: An open-source toolbox for large graph" +Brown, W. M. and Klavans, R. co-authored the paper "Openord: An open-source toolbox for large graph layout" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Brown, W. M. and Boyack, K. co-authored the paper "Openord: An open-source toolbox for large graph" +Brown, W. M. and Boyack, K. co-authored the paper "Openord: An open-source toolbox for large graph layout" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Klavans, R. and Boyack, K. co-authored the paper "Openord: An open-source toolbox for large graph" +Klavans, R. and Boyack, K. co-authored the paper "Openord: An open-source toolbox for large graph layout" + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Newman, M. E. published the paper "Modularity and community structure in networks" in the Proceedings of the National Academy of Sciences + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Levine, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Shashua, A. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Shashua, A. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Leyton-Brown, K. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ranade, P. and Joshi, A. co-authored the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Abdullah, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Yang, Z. and Manning, C. D. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Huang, M. and Duan, N. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Xu, Y. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Siddique, F. B. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Siddique, F. B. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Barezi, E. J. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Tang, Y. and Yang, Y. co-authored the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Martin, L. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bhargava, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bhosale, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Liang, Y. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 2.0 + Li, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Li, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Li, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Li, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Li, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Li, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 2.0 + Xu, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Xu, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Xu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Xu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Qu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" +Qu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Khramtsova, E. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wang, Y. and Lipka, N. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, R. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Qi, P. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Zhang, S. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Bengio, Y. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Cohen, W. W. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Salakhutdinov, R. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Chiang, W.-L. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Xing, E. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + \ No newline at end of file diff --git a/graphfleet/output/graphindex/artifacts/raw_extracted_entities.json b/graphfleet/output/graphindex/artifacts/raw_extracted_entities.json new file mode 100644 index 000000000..d936f2d8b --- /dev/null +++ b/graphfleet/output/graphindex/artifacts/raw_extracted_entities.json @@ -0,0 +1,30 @@ +{"id":"e8d83e6e7a7c0f57b218cef24976b745","chunk":"From Local to Global: A Graph RAG Approach to\nQuery-Focused Summarization\nDarren Edge1\u2020Ha Trinh1\u2020Newman Cheng2Joshua Bradley2Alex Chao3\nApurva Mody3Steven Truitt2\nJonathan Larson1\n1Microsoft Research\n2Microsoft Strategic Missions and Technologies\n3Microsoft Office of the CTO\n{daedge,trinhha,newmancheng,joshbradley,achao,moapurva,steventruitt,jolarso }\n@microsoft.com\n\u2020These authors contributed equally to this work\nAbstract\nThe use of retrieval-augmented generation (RAG) to retrieve relevant informa-\ntion from an external knowledge source enables large language models (LLMs)\nto answer questions over private and\/or previously unseen document collections.\nHowever, RAG fails on global questions directed at an entire text corpus, such\nas \u201cWhat are the main themes in the dataset?\u201d, since this is inherently a query-\nfocused summarization (QFS) task, rather than an explicit retrieval task. Prior\nQFS methods, meanwhile, fail to scale to the quantities of text indexed by typical\nRAG systems. To combine the strengths of these contrasting methods, we propose\na Graph RAG approach to question answering over private text corpora that scales\nwith both the generality of user questions and the quantity of source text to be in-\ndexed. Our approach uses an LLM to build a graph-based text index in two stages:\nfirst to derive an entity knowledge graph from the source documents, then to pre-\ngenerate community summaries for all groups of closely-related entities. Given a\nquestion, each community summary is used to generate a partial response, before\nall partial responses are again summarized in a final response to the user. For a\nclass of global sensemaking questions over datasets in the 1 million token range,\nwe show that Graph RAG leads to substantial improvements over a na \u00a8\u0131ve RAG\nbaseline for both the comprehensiveness and diversity of generated answers. An\nopen-source, Python-based implementation of both global and local Graph RAG\napproaches is forthcoming at https:\/\/aka .ms\/graphrag .\n1 Introduction\nHuman endeavors across a range of domains rely on our ability to read and reason about large\ncollections of documents, often reaching conclusions that go beyond anything stated in the source\ntexts themselves. With the emergence of large language models (LLMs), we are already witnessing\nattempts to automate human-like sense","chunk_id":"e8d83e6e7a7c0f57b218cef24976b745","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"DARREN EDGE","type":"PERSON","description":"Darren Edge is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"HA TRINH","type":"PERSON","description":"Ha Trinh is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"NEWMAN CHENG","type":"PERSON","description":"Newman Cheng is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"JOSHUA BRADLEY","type":"PERSON","description":"Joshua Bradley is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"ALEX CHAO","type":"PERSON","description":"Alex Chao is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"APURVA MODY","type":"PERSON","description":"Apurva Mody is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"STEVEN TRUITT","type":"PERSON","description":"Steven Truitt is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"JONATHAN LARSON","type":"PERSON","description":"Jonathan Larson is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"MICROSOFT RESEARCH","type":"ORGANIZATION","description":"Microsoft Research is an organization where some of the authors of the paper are affiliated","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"MICROSOFT STRATEGIC MISSIONS AND TECHNOLOGIES","type":"ORGANIZATION","description":"Microsoft Strategic Missions and Technologies is an organization where some of the authors of the paper are affiliated","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"MICROSOFT OFFICE OF THE CTO","type":"ORGANIZATION","description":"Microsoft Office of the CTO is an organization where some of the authors of the paper are affiliated","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"RAG","type":"METHOD","description":"RAG (Retrieval-Augmented Generation) is a method used to retrieve relevant information from an external knowledge source to enable large language models to answer questions","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"LLM","type":"TECHNOLOGY","description":"LLM (Large Language Model) is a type of model used to automate human-like sensemaking and reasoning over large collections of documents","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"GRAPH RAG","type":"METHOD","description":"Graph RAG is an approach that combines retrieval-augmented generation with graph-based text indexing to answer questions over private text corpora","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"QFS","type":"METHOD","description":"QFS (Query-Focused Summarization) is a method used to generate summaries based on specific user queries","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"COMMUNITY SUMMARIES","type":"CONCEPT","description":"Community summaries are pre-generated summaries for groups of closely-related entities used in the Graph RAG approach","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"GLOBAL SENSEMAKING QUESTIONS","type":"CONCEPT","description":"Global sensemaking questions are questions that require understanding and summarizing large datasets, often beyond the explicit content of the source texts","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"1 MILLION TOKEN RANGE","type":"METRIC","description":"1 million token range refers to the scale of datasets used in the evaluation of the Graph RAG approach","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"PYTHON","type":"TECHNOLOGY","description":"Python is a programming language used to implement the open-source version of the Graph RAG approach","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"HTTPS:\/\/AKA.MS\/GRAPHRAG","type":"URL","description":"The URL where the open-source implementation of the Graph RAG approach will be available","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"QUERY-FOCUSED SUMMARIZATION","type":"METHOD","description":"Query-Focused Summarization (QFS) is a method used to generate summaries based on specific user queries","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"EXTERNAL KNOWLEDGE SOURCE","type":"CONCEPT","description":"An external knowledge source is a repository of information that can be accessed to retrieve relevant data for answering questions","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"TEXT CORPUS","type":"CONCEPT","description":"A text corpus is a large collection of written texts used for analysis and research","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"ENTITY KNOWLEDGE GRAPH","type":"CONCEPT","description":"An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"SOURCE DOCUMENTS","type":"CONCEPT","description":"Source documents are the original texts from which information is retrieved or summarized","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"PARTIAL RESPONSE","type":"CONCEPT","description":"A partial response is an intermediate answer generated from community summaries before being combined into a final response","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"FINAL RESPONSE","type":"CONCEPT","description":"A final response is the comprehensive answer generated after combining all partial responses","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"COMPREHENSIVENESS","type":"METRIC","description":"Comprehensiveness is a metric used to evaluate the extent to which generated answers cover the relevant information","source_id":"e8d83e6e7a7c0f57b218cef24976b745"},{"name":"DIVERSITY","type":"METRIC","description":"Diversity is a metric used to evaluate the variety of information included in the generated answers","source_id":"e8d83e6e7a7c0f57b218cef24976b745"}],"entity_graph":" PERSON<\/data> Darren Edge is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> PERSON<\/data> Ha Trinh is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> PERSON<\/data> Newman Cheng is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> PERSON<\/data> Joshua Bradley is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> PERSON<\/data> Alex Chao is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> PERSON<\/data> Apurva Mody is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> PERSON<\/data> Steven Truitt is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> PERSON<\/data> Jonathan Larson is an author of the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> ORGANIZATION<\/data> Microsoft Research is an organization where some of the authors of the paper are affiliated<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> ORGANIZATION<\/data> Microsoft Strategic Missions and Technologies is an organization where some of the authors of the paper are affiliated<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> ORGANIZATION<\/data> Microsoft Office of the CTO is an organization where some of the authors of the paper are affiliated<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> METHOD<\/data> RAG (Retrieval-Augmented Generation) is a method used to retrieve relevant information from an external knowledge source to enable large language models to answer questions<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> TECHNOLOGY<\/data> LLM (Large Language Model) is a type of model used to automate human-like sensemaking and reasoning over large collections of documents<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> METHOD<\/data> Graph RAG is an approach that combines retrieval-augmented generation with graph-based text indexing to answer questions over private text corpora<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> METHOD<\/data> QFS (Query-Focused Summarization) is a method used to generate summaries based on specific user queries<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> CONCEPT<\/data> Community summaries are pre-generated summaries for groups of closely-related entities used in the Graph RAG approach<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> CONCEPT<\/data> Global sensemaking questions are questions that require understanding and summarizing large datasets, often beyond the explicit content of the source texts<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> METRIC<\/data> 1 million token range refers to the scale of datasets used in the evaluation of the Graph RAG approach<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> TECHNOLOGY<\/data> Python is a programming language used to implement the open-source version of the Graph RAG approach<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> URL<\/data> The URL where the open-source implementation of the Graph RAG approach will be available<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> METHOD<\/data> Query-Focused Summarization (QFS) is a method used to generate summaries based on specific user queries<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> CONCEPT<\/data> An external knowledge source is a repository of information that can be accessed to retrieve relevant data for answering questions<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> CONCEPT<\/data> A text corpus is a large collection of written texts used for analysis and research<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> CONCEPT<\/data> An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> CONCEPT<\/data> Source documents are the original texts from which information is retrieved or summarized<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> CONCEPT<\/data> A partial response is an intermediate answer generated from community summaries before being combined into a final response<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> CONCEPT<\/data> A final response is the comprehensive answer generated after combining all partial responses<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> METRIC<\/data> Comprehensiveness is a metric used to evaluate the extent to which generated answers cover the relevant information<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> METRIC<\/data> Diversity is a metric used to evaluate the variety of information included in the generated answers<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/node> 1.0<\/data> Darren Edge and Ha Trinh co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Darren Edge and Newman Cheng co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Darren Edge and Joshua Bradley co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Darren Edge and Alex Chao co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Darren Edge and Apurva Mody co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Darren Edge and Steven Truitt co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Darren Edge and Jonathan Larson co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Darren Edge is affiliated with Microsoft Research<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Ha Trinh and Newman Cheng co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Ha Trinh and Joshua Bradley co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Ha Trinh and Alex Chao co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Ha Trinh and Apurva Mody co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Ha Trinh and Steven Truitt co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Ha Trinh and Jonathan Larson co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Ha Trinh is affiliated with Microsoft Research<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Newman Cheng and Joshua Bradley co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Newman Cheng and Alex Chao co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Newman Cheng and Apurva Mody co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Newman Cheng and Steven Truitt co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Newman Cheng and Jonathan Larson co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Newman Cheng is affiliated with Microsoft Strategic Missions and Technologies<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Joshua Bradley and Alex Chao co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Joshua Bradley and Apurva Mody co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Joshua Bradley and Steven Truitt co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Joshua Bradley and Jonathan Larson co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Joshua Bradley is affiliated with Microsoft Strategic Missions and Technologies<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Alex Chao and Apurva Mody co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Alex Chao and Steven Truitt co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Alex Chao and Jonathan Larson co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Alex Chao is affiliated with Microsoft Office of the CTO<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Apurva Mody and Steven Truitt co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Apurva Mody and Jonathan Larson co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Apurva Mody is affiliated with Microsoft Office of the CTO<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Steven Truitt and Jonathan Larson co-authored the paper \"From Local to Global: A Graph RAG Approach to Query-Focused Summarization\"<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Steven Truitt is affiliated with Microsoft Strategic Missions and Technologies<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Jonathan Larson is affiliated with Microsoft Research<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> RAG uses LLMs to retrieve relevant information from external knowledge sources<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Graph RAG combines the strengths of RAG with graph-based text indexing<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Query-Focused Summarization is a task that RAG fails to address effectively<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> RAG retrieves relevant information from an external knowledge source<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Graph RAG uses LLMs to build a graph-based text index<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Graph RAG is proposed as a method to combine the strengths of RAG and QFS<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Community summaries are used in the Graph RAG approach to generate partial responses<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Graph RAG is designed to handle global sensemaking questions over large datasets<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Graph RAG is implemented in Python<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> The open-source implementation of Graph RAG will be available at this URL<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Graph RAG uses an entity knowledge graph to index text<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Graph RAG improves the comprehensiveness of generated answers<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Graph RAG improves the diversity of generated answers<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Community summaries are used to generate partial responses<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Global sensemaking questions are evaluated over datasets in the 1 million token range<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Global sensemaking questions are directed at an entire text corpus<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> An entity knowledge graph is derived from source documents<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> 1.0<\/data> Partial responses are summarized to generate a final response<\/data> e8d83e6e7a7c0f57b218cef24976b745<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"f0306814bf64f5c9e79603fc6a52f4ea","chunk":" .\n1 Introduction\nHuman endeavors across a range of domains rely on our ability to read and reason about large\ncollections of documents, often reaching conclusions that go beyond anything stated in the source\ntexts themselves. With the emergence of large language models (LLMs), we are already witnessing\nattempts to automate human-like sensemaking in complex domains like scientific discovery (Mi-\ncrosoft, 2023) and intelligence analysis (Ranade and Joshi, 2023), where sensemaking is defined as\nPreprint. Under review.arXiv:2404.16130v1 [cs.CL] 24 Apr 2024Source Documents\nText Chunkstext extraction\nand chunking\nElement Instancesdomain-tailored\nsummarization\nElement Summariesdomain-tailored\nsummarization\nGraph Communitiescommunity\ndetectionCommunity Summaries\ndomain-tailored\nsummarizationCommunity Answers\nquery-focused\nsummarizationGlobal Answer\nquery-focused\nsummarization\nIndexing Time Query Time Pipeline Stage\nFigure 1: Graph RAG pipeline using an LLM-derived graph index of source document text. This\nindex spans nodes (e.g., entities), edges (e.g., relationships), and covariates (e.g., claims) that have\nbeen detected, extracted, and summarized by LLM prompts tailored to the domain of the dataset.\nCommunity detection (e.g., Leiden, Traag et al., 2019) is used to partition the graph index into\ngroups of elements (nodes, edges, covariates) that the LLM can summarize in parallel at both index-\ning time and query time. The \u201cglobal answer\u201d to a given query is produced using a final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\n\u201ca motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively \u201d (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions","chunk_id":"f0306814bf64f5c9e79603fc6a52f4ea","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"HUMAN ENDEAVORS","type":"ACTIVITY","description":"Human endeavors refer to activities and efforts across various domains that rely on reading and reasoning about large collections of documents","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"LARGE LANGUAGE MODELS (LLMS)","type":"TECHNOLOGY","description":"Large language models are advanced AI models designed to understand and generate human-like text, used in automating sensemaking in complex domains","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"SCIENTIFIC DISCOVERY","type":"DOMAIN","description":"Scientific discovery is a complex domain where sensemaking is applied to understand and generate new knowledge from scientific texts","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"INTELLIGENCE ANALYSIS","type":"DOMAIN","description":"Intelligence analysis is a complex domain where sensemaking is applied to understand and generate insights from intelligence data","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"SENSEMAKING","type":"PROCESS","description":"Sensemaking is the process of understanding connections among people, places, and events to anticipate their trajectories and act effectively","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"SOURCE DOCUMENTS","type":"DATA","description":"Source documents are the original texts from which information is extracted and analyzed","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"TEXT CHUNKS","type":"DATA","description":"Text chunks are segments of source documents that are extracted for further analysis","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"ELEMENT INSTANCES","type":"DATA","description":"Element instances are specific pieces of information extracted from text chunks, tailored to the domain","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"ELEMENT SUMMARIES","type":"DATA","description":"Element summaries are concise representations of element instances, tailored to the domain","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"GRAPH COMMUNITIES","type":"DATA","description":"Graph communities are groups of elements (nodes, edges, covariates) detected in a graph index, used for summarization","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"COMMUNITY SUMMARIES","type":"DATA","description":"Community summaries are summaries of graph communities, tailored to the domain","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"COMMUNITY ANSWERS","type":"DATA","description":"Community answers are query-focused summaries of community summaries","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"GLOBAL ANSWER","type":"DATA","description":"Global answer is the final query-focused summary produced from all relevant community summaries","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"INDEXING TIME","type":"TIME","description":"Indexing time refers to the time when the graph index is created and elements are summarized","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"QUERY TIME","type":"TIME","description":"Query time refers to the time when a query is made and the relevant summaries are generated","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"GRAPH RAG PIPELINE","type":"PROCESS","description":"Graph RAG pipeline is a process using an LLM-derived graph index to detect, extract, and summarize nodes, edges, and covariates in source documents","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"NODES","type":"DATA","description":"Nodes are entities detected in the graph index of source documents","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"EDGES","type":"DATA","description":"Edges are relationships detected in the graph index of source documents","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"COVARIATES","type":"DATA","description":"Covariates are claims or additional information detected in the graph index of source documents","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"LEIDEN","type":"METHOD","description":"Leiden is a community detection method used to partition the graph index into groups of elements","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"RETRIEVAL-AUGMENTED GENERATION (RAG)","type":"METHOD","description":"Retrieval-augmented generation is an approach to answering user questions over entire datasets by retrieving and generating relevant information","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"MICROSOFT","type":"ORGANIZATION","description":"Microsoft is an organization involved in automating sensemaking in scientific discovery using LLMs","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"RANADE","type":"PERSON","description":"Ranade is an author involved in research on automating sensemaking in intelligence analysis using LLMs","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"JOSHI","type":"PERSON","description":"Joshi is an author involved in research on automating sensemaking in intelligence analysis using LLMs","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"KLEIN","type":"PERSON","description":"Klein is an author who defined sensemaking and discussed its importance in understanding connections among people, places, and events","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"LEWIS","type":"PERSON","description":"Lewis is an author who contributed to the development of the retrieval-augmented generation (RAG) approach","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"TRAAG","type":"PERSON","description":"Traag is an author who contributed to the development of the Leiden community detection method","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"ARXIV","type":"PUBLICATION","description":"arXiv is a repository where the preprint of the discussed research paper is available","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"PREPRINT","type":"PUBLICATION","description":"Preprint refers to the version of the research paper that is under review and available on arXiv","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"CS.CL","type":"CATEGORY","description":"cs.CL is the category under which the research paper is classified on arXiv","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"24 APR 2024","type":"DATE","description":"24 Apr 2024 is the date when the research paper was submitted to arXiv","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"2404.16130V1","type":"IDENTIFIER","description":"2404.16130v1 is the identifier for the research paper on arXiv","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"DOCUMENT COLLECTIONS","type":"DATA","description":"Document collections refer to large sets of documents that are analyzed for sensemaking","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"LLM PROMPTS","type":"TECHNOLOGY","description":"LLM prompts are specific instructions given to large language models to tailor their responses to the domain of the dataset","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"COMMUNITY DETECTION","type":"METHOD","description":"Community detection is a method used to identify groups of related elements within a graph","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"QUERY-FOCUSED SUMMARIZATION","type":"METHOD","description":"Query-focused summarization is a method used to generate summaries that are relevant to a specific query","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"DOMAIN-TAILORED SUMMARIZATION","type":"METHOD","description":"Domain-tailored summarization is a method used to create summaries that are specific to the domain of the dataset","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"KLEIN ET AL.","type":"PERSON","description":"Klein et al. are authors who defined sensemaking and discussed its importance in understanding connections among people, places, and events","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"RANADE AND JOSHI","type":"PERSON","description":"Ranade and Joshi are authors who discussed the use of LLMs in intelligence analysis","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"LEWIS ET AL.","type":"PERSON","description":"Lewis et al. are authors who developed the retrieval-augmented generation (RAG) approach","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"},{"name":"TRAAG ET AL.","type":"PERSON","description":"Traag et al. are authors who developed the Leiden community detection method","source_id":"f0306814bf64f5c9e79603fc6a52f4ea"}],"entity_graph":" ACTIVITY<\/data> Human endeavors refer to activities and efforts across various domains that rely on reading and reasoning about large collections of documents<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> TECHNOLOGY<\/data> Large language models are advanced AI models designed to understand and generate human-like text, used in automating sensemaking in complex domains<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DOMAIN<\/data> Scientific discovery is a complex domain where sensemaking is applied to understand and generate new knowledge from scientific texts<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DOMAIN<\/data> Intelligence analysis is a complex domain where sensemaking is applied to understand and generate insights from intelligence data<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PROCESS<\/data> Sensemaking is the process of understanding connections among people, places, and events to anticipate their trajectories and act effectively<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Source documents are the original texts from which information is extracted and analyzed<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Text chunks are segments of source documents that are extracted for further analysis<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Element instances are specific pieces of information extracted from text chunks, tailored to the domain<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Element summaries are concise representations of element instances, tailored to the domain<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Graph communities are groups of elements (nodes, edges, covariates) detected in a graph index, used for summarization<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Community summaries are summaries of graph communities, tailored to the domain<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Community answers are query-focused summaries of community summaries<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Global answer is the final query-focused summary produced from all relevant community summaries<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> TIME<\/data> Indexing time refers to the time when the graph index is created and elements are summarized<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> TIME<\/data> Query time refers to the time when a query is made and the relevant summaries are generated<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PROCESS<\/data> Graph RAG pipeline is a process using an LLM-derived graph index to detect, extract, and summarize nodes, edges, and covariates in source documents<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Nodes are entities detected in the graph index of source documents<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Edges are relationships detected in the graph index of source documents<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Covariates are claims or additional information detected in the graph index of source documents<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> METHOD<\/data> Leiden is a community detection method used to partition the graph index into groups of elements<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> METHOD<\/data> Retrieval-augmented generation is an approach to answering user questions over entire datasets by retrieving and generating relevant information<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> ORGANIZATION<\/data> Microsoft is an organization involved in automating sensemaking in scientific discovery using LLMs<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PERSON<\/data> Ranade is an author involved in research on automating sensemaking in intelligence analysis using LLMs<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PERSON<\/data> Joshi is an author involved in research on automating sensemaking in intelligence analysis using LLMs<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PERSON<\/data> Klein is an author who defined sensemaking and discussed its importance in understanding connections among people, places, and events<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PERSON<\/data> Lewis is an author who contributed to the development of the retrieval-augmented generation (RAG) approach<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PERSON<\/data> Traag is an author who contributed to the development of the Leiden community detection method<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PUBLICATION<\/data> arXiv is a repository where the preprint of the discussed research paper is available<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PUBLICATION<\/data> Preprint refers to the version of the research paper that is under review and available on arXiv<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> CATEGORY<\/data> cs.CL is the category under which the research paper is classified on arXiv<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATE<\/data> 24 Apr 2024 is the date when the research paper was submitted to arXiv<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> IDENTIFIER<\/data> 2404.16130v1 is the identifier for the research paper on arXiv<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> DATA<\/data> Document collections refer to large sets of documents that are analyzed for sensemaking<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> TECHNOLOGY<\/data> LLM prompts are specific instructions given to large language models to tailor their responses to the domain of the dataset<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> METHOD<\/data> Community detection is a method used to identify groups of related elements within a graph<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> METHOD<\/data> Query-focused summarization is a method used to generate summaries that are relevant to a specific query<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> METHOD<\/data> Domain-tailored summarization is a method used to create summaries that are specific to the domain of the dataset<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PERSON<\/data> Klein et al. are authors who defined sensemaking and discussed its importance in understanding connections among people, places, and events<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PERSON<\/data> Ranade and Joshi are authors who discussed the use of LLMs in intelligence analysis<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PERSON<\/data> Lewis et al. are authors who developed the retrieval-augmented generation (RAG) approach<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> PERSON<\/data> Traag et al. are authors who developed the Leiden community detection method<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/node> 1.0<\/data> Human endeavors rely on sensemaking to understand and reason about large collections of documents<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Human endeavors rely on analyzing document collections for sensemaking<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> LLMs are used to automate sensemaking in complex domains<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Microsoft uses LLMs for automating sensemaking in scientific discovery<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Ranade uses LLMs for automating sensemaking in intelligence analysis<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Joshi uses LLMs for automating sensemaking in intelligence analysis<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> LLM prompts are used to tailor the responses of large language models<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Ranade and Joshi discussed the use of LLMs in intelligence analysis<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Sensemaking is applied in the domain of scientific discovery<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Sensemaking is applied in the domain of intelligence analysis<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Klein defined and discussed the importance of sensemaking<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Klein et al. defined and discussed the importance of sensemaking<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Text chunks are extracted from source documents<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Element instances are extracted from text chunks<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Element summaries are created from element instances<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Domain-tailored summarization is used to create element summaries<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Community summaries are created from graph communities<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Community detection is used to identify graph communities<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Community answers are created from community summaries<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Domain-tailored summarization is used to create community summaries<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Global answer is created from community answers<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Query-focused summarization is used to produce the global answer<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Graph RAG pipeline operates at indexing time<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Graph RAG pipeline operates at query time<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Nodes are detected in the graph RAG pipeline<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Edges are detected in the graph RAG pipeline<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Covariates are detected in the graph RAG pipeline<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Leiden method is used in the graph RAG pipeline for community detection<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Graph RAG pipeline uses the RAG approach<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Traag contributed to the development of the Leiden method<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Traag et al. developed the Leiden method<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Lewis contributed to the development of the RAG approach<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Lewis et al. developed the RAG approach<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Preprint is available on arXiv<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Preprint is classified under cs.CL on arXiv<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Preprint was submitted on 24 Apr 2024<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> 1.0<\/data> Preprint has the identifier 2404.16130v1 on arXiv<\/data> f0306814bf64f5c9e79603fc6a52f4ea<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"fb3c48579608fa28be585ceb6cd2f0fe","chunk":" et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular, query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., 2023) series,\nall of which can use in-context learning to summarize any content provided in their context window.\nThe challenge remains, however, for query-focused abstractive summarization over an entire corpus.\nSuch volumes of text can greatly exceed the limits of LLM context windows, and the expansion of\nsuch windows may not be enough given that information can be \u201clost in the middle\u201d of longer\ncontexts (Kuratov et al., 2024; Liu et al., 2023). In addition, although the direct retrieval of text\nchunks in na \u00a8\u0131ve RAG is likely inadequate for QFS tasks, it is possible that an alternative form of\npre-indexing could support a new RAG approach specifically targeting global summarization.\nIn this paper, we present a Graph RAG approach based on global summarization of an LLM-derived\nknowledge graph (Figure 1). In contrast with related work that exploits the structured retrieval\nand traversal affordances of graph indexes (subsection 4.2","chunk_id":"fb3c48579608fa28be585ceb6cd2f0fe","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"RETRIEVAL-AUGMENTED GENERATION (RAG)","type":"METHOD","description":"RAG is an established approach to answering user questions over entire datasets by retrieving relevant text regions to provide grounding for the generation task","source_id":"fb3c48579608fa28be585ceb6cd2f0fe"},{"name":"QUERY-FOCUSED SUMMARIZATION (QFS)","type":"METHOD","description":"QFS is a task framing that focuses on generating summaries based on specific queries, rather than just concatenating excerpts","source_id":"fb3c48579608fa28be585ceb6cd2f0fe"},{"name":"QUERY-FOCUSED ABSTRACTIVE SUMMARIZATION","type":"METHOD","description":"A type of summarization that generates natural language summaries based on specific queries, rather than just extracting text","source_id":"fb3c48579608fa28be585ceb6cd2f0fe"},{"name":"TRANSFORMER ARCHITECTURE","type":"TECHNOLOGY","description":"A neural network architecture that has shown substantial improvements in various summarization tasks","source_id":"fb3c48579608fa28be585ceb6cd2f0fe"},{"name":"LARGE LANGUAGE MODELS (LLMS)","type":"TECHNOLOGY","description":"Modern language models, including GPT, Llama, and Gemini, that can use in-context learning to summarize content","source_id":"fb3c48579608fa28be585ceb6cd2f0fe"},{"name":"GPT","type":"TECHNOLOGY","description":"A series of large language models known for their ability to perform in-context learning and summarization","source_id":"fb3c48579608fa28be585ceb6cd2f0fe"},{"name":"LLAMA","type":"TECHNOLOGY","description":"A series of large language models known for their ability to perform in-context learning and summarization","source_id":"fb3c48579608fa28be585ceb6cd2f0fe"},{"name":"GEMINI","type":"TECHNOLOGY","description":"A series of large language models known for their ability to perform in-context learning and summarization","source_id":"fb3c48579608fa28be585ceb6cd2f0fe"},{"name":"GRAPH RAG","type":"METHOD","description":"A new approach based on global summarization of an LLM-derived knowledge graph, targeting global summarization tasks","source_id":"fb3c48579608fa28be585ceb6cd2f0fe"},{"name":"KNOWLEDGE GRAPH","type":"TECHNOLOGY","description":"A structured representation of knowledge used in the Graph RAG approach for global summarization","source_id":"fb3c48579608fa28be585ceb6cd2f0fe"},{"name":"LEWIS ET AL., 2020","type":"REFERENCE","description":"Authors of a paper on Retrieval-augmented generation (RAG)","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"DANG, 2006","type":"REFERENCE","description":"Author of a paper on query-focused summarization (QFS)","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"BAUMEL ET AL., 2018","type":"REFERENCE","description":"Authors of a paper on query-focused abstractive summarization","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"LASKAR ET AL., 2020","type":"REFERENCE","description":"Authors of a paper on query-focused abstractive summarization","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"YAO ET AL., 2017","type":"REFERENCE","description":"Authors of a paper on query-focused abstractive summarization","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"GOODWIN ET AL., 2020","type":"REFERENCE","description":"Authors of a paper on the early applications of the transformer architecture","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"LASKAR ET AL., 2022","type":"REFERENCE","description":"Authors of a paper on the early applications of the transformer architecture","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"LIU AND LAPATA, 2019","type":"REFERENCE","description":"Authors of a paper on the early applications of the transformer architecture","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"ACHIAM ET AL., 2023","type":"REFERENCE","description":"Authors of a paper on the GPT series of large language models","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"BROWN ET AL., 2020","type":"REFERENCE","description":"Authors of a paper on the GPT series of large language models","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"TOUVRON ET AL., 2023","type":"REFERENCE","description":"Authors of a paper on the Llama series of large language models","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"ANIL ET AL., 2023","type":"REFERENCE","description":"Authors of a paper on the Gemini series of large language models","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"KURATOV ET AL., 2024","type":"REFERENCE","description":"Authors of a paper on the limitations of LLM context windows","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"},{"name":"LIU ET AL., 2023","type":"REFERENCE","description":"Authors of a paper on the limitations of LLM context windows","source_id":"fb3c48579608fa28be585ceb6cd2f0fe","entity_type":"REFERENCE"}],"entity_graph":" METHOD<\/data> RAG is an established approach to answering user questions over entire datasets by retrieving relevant text regions to provide grounding for the generation task<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/node> METHOD<\/data> QFS is a task framing that focuses on generating summaries based on specific queries, rather than just concatenating excerpts<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/node> METHOD<\/data> A type of summarization that generates natural language summaries based on specific queries, rather than just extracting text<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/node> TECHNOLOGY<\/data> A neural network architecture that has shown substantial improvements in various summarization tasks<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/node> TECHNOLOGY<\/data> Modern language models, including GPT, Llama, and Gemini, that can use in-context learning to summarize content<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/node> TECHNOLOGY<\/data> A series of large language models known for their ability to perform in-context learning and summarization<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/node> TECHNOLOGY<\/data> A series of large language models known for their ability to perform in-context learning and summarization<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/node> TECHNOLOGY<\/data> A series of large language models known for their ability to perform in-context learning and summarization<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/node> METHOD<\/data> A new approach based on global summarization of an LLM-derived knowledge graph, targeting global summarization tasks<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/node> TECHNOLOGY<\/data> A structured representation of knowledge used in the Graph RAG approach for global summarization<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/node> REFERENCE<\/data> Authors of a paper on Retrieval-augmented generation (RAG)<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Author of a paper on query-focused summarization (QFS)<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on query-focused abstractive summarization<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on query-focused abstractive summarization<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on query-focused abstractive summarization<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on the early applications of the transformer architecture<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on the early applications of the transformer architecture<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on the early applications of the transformer architecture<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on the GPT series of large language models<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on the GPT series of large language models<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on the Llama series of large language models<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on the Gemini series of large language models<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on the limitations of LLM context windows<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> REFERENCE<\/data> Authors of a paper on the limitations of LLM context windows<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> REFERENCE<\/data> <\/node> 2.0<\/data> Lewis et al., 2020, are the authors who established the RAG approach<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Dang, 2006, is the author who established the QFS approach<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Baumel et al., 2018, are the authors who worked on query-focused abstractive summarization<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Laskar et al., 2020, are the authors who worked on query-focused abstractive summarization<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Yao et al., 2017, are the authors who worked on query-focused abstractive summarization<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Goodwin et al., 2020, are the authors who worked on the early applications of the transformer architecture<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Laskar et al., 2022, are the authors who worked on the early applications of the transformer architecture<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Liu and Lapata, 2019, are the authors who worked on the early applications of the transformer architecture<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> GPT is a type of large language model<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Llama is a type of large language model<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Gemini is a type of large language model<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Kuratov et al., 2024, discussed the limitations of LLM context windows<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Liu et al., 2023, discussed the limitations of LLM context windows<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Achiam et al., 2023, are the authors who worked on the GPT series of large language models<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Brown et al., 2020, are the authors who worked on the GPT series of large language models<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Touvron et al., 2023, are the authors who worked on the Llama series of large language models<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Anil et al., 2023, are the authors who worked on the Gemini series of large language models<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> 2.0<\/data> Graph RAG uses a knowledge graph for global summarization<\/data> fb3c48579608fa28be585ceb6cd2f0fe<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"21e52bc06a82796b1f4bcd73edda1f2a","chunk":" a new RAG approach specifically targeting global summarization.\nIn this paper, we present a Graph RAG approach based on global summarization of an LLM-derived\nknowledge graph (Figure 1). In contrast with related work that exploits the structured retrieval\nand traversal affordances of graph indexes (subsection 4.2), we focus on a previously unexplored\nquality of graphs in this context: their inherent modularity (Newman, 2006) and the ability of com-\nmunity detection algorithms to partition graphs into modular communities of closely-related nodes\n(e.g., Louvain, Blondel et al., 2008; Leiden, Traag et al., 2019). LLM-generated summaries of these\n20 1 2 30100002000030000\nNumber of gleanings performedEntity references detected600 chunk size\n1200 chunk size\n2400 chunk size\nFigure 2: How the entity references detected in the HotPotQA dataset (Yang et al., 2018)\nvaries with chunk size and gleanings for our generic entity extraction prompt with gpt-4-turbo .\ncommunity descriptions provide complete coverage of the underlying graph index and the input doc-\numents it represents. Query-focused summarization of an entire corpus is then made possible using\na map-reduce approach: first using each community summary to answer the query independently\nand in parallel, then summarizing all relevant partial answers into a final global answer.\nTo evaluate this approach, we used an LLM to generate a diverse set of activity-centered sense-\nmaking questions from short descriptions of two representative real-world datasets, containing pod-\ncast transcripts and news articles respectively. For the target qualities of comprehensiveness, diver-\nsity, and empowerment (defined in subsection 3.4) that develop understanding of broad issues and\nthemes, we both explore the impact of varying the the hierarchical level of community summaries\nused to answer queries, as well as compare to na \u00a8\u0131ve RAG and global map-reduce summarization\nof source texts. We show that all global approaches outperform na \u00a8\u0131ve RAG on comprehensiveness\nand diversity, and that Graph RAG with intermediate- and low-level community summaries shows\nfavorable performance over source text summarization on these same metrics, at lower token costs.\n2 Graph RAG Approach & Pipeline\nWe now unpack the high-level data flow of the Graph RAG approach (Figure 1) and pipeline, de-\nscribing key design parameters","chunk_id":"21e52bc06a82796b1f4bcd73edda1f2a","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"RAG","type":"METHOD","description":"RAG (Retrieval-Augmented Generation) is a method used for generating responses in text generation tasks","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"GRAPH RAG","type":"METHOD","description":"Graph RAG is a specific approach to RAG that focuses on global summarization using a knowledge graph","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"LLM","type":"TECHNOLOGY","description":"LLM (Large Language Model) is a type of AI model used for generating text and answering queries","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"KNOWLEDGE GRAPH","type":"TECHNOLOGY","description":"A knowledge graph is a structured representation of information, used in the Graph RAG approach for summarization","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"COMMUNITY DETECTION ALGORITHMS","type":"TECHNOLOGY","description":"Algorithms used to partition graphs into modular communities of closely-related nodes","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"LOUVAIN","type":"ALGORITHM","description":"Louvain is a community detection algorithm used to partition graphs into modular communities","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"LEIDEN","type":"ALGORITHM","description":"Leiden is a community detection algorithm used to partition graphs into modular communities","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"HOTPOTQA","type":"DATASET","description":"HotPotQA is a dataset used to evaluate the entity extraction prompt with gpt-4-turbo","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"GPT-4-TURBO","type":"TECHNOLOGY","description":"GPT-4-Turbo is a version of the GPT-4 model used for entity extraction in the evaluation","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"PODCAST TRANSCRIPTS","type":"DATASET","description":"A dataset consisting of transcripts from podcasts used for analysis","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"NEWS ARTICLES","type":"DATASET","description":"A dataset consisting of news articles used for analysis","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"MAP-REDUCE","type":"METHOD","description":"Map-reduce is a method used for query-focused summarization of an entire corpus","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"COMPREHENSIVENESS","type":"METRIC","description":"A target quality used to evaluate the summarization approach, focusing on the completeness of the summary","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"DIVERSITY","type":"METRIC","description":"A target quality used to evaluate the summarization approach, focusing on the variety of information in the summary","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"EMPOWERMENT","type":"METRIC","description":"A target quality used to evaluate the summarization approach, focusing on the ability to develop understanding of broad issues and themes","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"NAIVE RAG","type":"METHOD","description":"Naive RAG is a basic approach to RAG used as a baseline for comparison","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"GLOBAL MAP-REDUCE SUMMARIZATION","type":"METHOD","description":"A method for summarizing source texts using a map-reduce approach","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"COMMUNITY SUMMARIES","type":"OUTPUT","description":"Summaries generated from modular communities in the knowledge graph","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"QUERY-FOCUSED SUMMARIZATION","type":"METHOD","description":"A summarization method that focuses on answering specific queries using the entire corpus","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"ACTIVITY-CENTERED SENSEMAKING QUESTIONS","type":"OUTPUT","description":"Questions generated to evaluate the summarization approach, focusing on understanding activities","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"HIERARCHICAL LEVEL","type":"PARAMETER","description":"The level of detail in community summaries used to answer queries","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"TOKEN COSTS","type":"METRIC","description":"The computational cost measured in tokens used in the summarization process","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"DATA FLOW","type":"PROCESS","description":"The high-level process of the Graph RAG approach and pipeline","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"DESIGN PARAMETERS","type":"PARAMETER","description":"Key parameters that influence the design of the Graph RAG approach and pipeline","source_id":"21e52bc06a82796b1f4bcd73edda1f2a","entity_type":"PARAMETER"},{"name":"GLOBAL SUMMARIZATION","type":"METHOD","description":"Global summarization is a method that aims to summarize information from a large dataset or corpus","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"MODULARITY","type":"ATTRIBUTE","description":"Modularity is an inherent quality of graphs that allows them to be partitioned into communities of closely-related nodes","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"COMMUNITY DESCRIPTIONS","type":"OUTPUT","description":"Descriptions generated from modular communities in the knowledge graph","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"QUERY","type":"INPUT","description":"A specific question or request for information that the summarization methods aim to answer","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"CORPUS","type":"DATASET","description":"A large collection of texts or documents used for analysis and summarization","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"PARTIAL ANSWERS","type":"OUTPUT","description":"Intermediate answers generated from community summaries before being combined into a final global answer","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"FINAL GLOBAL ANSWER","type":"OUTPUT","description":"The comprehensive answer generated by combining all relevant partial answers","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"ACTIVITY-CENTERED SENSEMAKING","type":"METHOD","description":"A method that focuses on generating questions to understand activities from datasets","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"SHORT DESCRIPTIONS","type":"INPUT","description":"Brief descriptions of datasets used to generate sensemaking questions","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"REAL-WORLD DATASETS","type":"DATASET","description":"Datasets that represent real-world information, such as podcast transcripts and news articles","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"HIERARCHICAL LEVEL OF COMMUNITY SUMMARIES","type":"PARAMETER","description":"The level of detail in community summaries used to answer queries","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"SOURCE TEXT SUMMARIZATION","type":"METHOD","description":"A method that summarizes the original source texts directly","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"LOW-LEVEL COMMUNITY SUMMARIES","type":"OUTPUT","description":"Summaries generated from lower hierarchical levels of the community in the knowledge graph","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"INTERMEDIATE-LEVEL COMMUNITY SUMMARIES","type":"OUTPUT","description":"Summaries generated from intermediate hierarchical levels of the community in the knowledge graph","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"},{"name":"HIGH-LEVEL COMMUNITY SUMMARIES","type":"OUTPUT","description":"Summaries generated from higher hierarchical levels of the community in the knowledge graph","source_id":"21e52bc06a82796b1f4bcd73edda1f2a"}],"entity_graph":" METHOD<\/data> RAG (Retrieval-Augmented Generation) is a method used for generating responses in text generation tasks<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> METHOD<\/data> Graph RAG is a specific approach to RAG that focuses on global summarization using a knowledge graph<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> TECHNOLOGY<\/data> LLM (Large Language Model) is a type of AI model used for generating text and answering queries<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> TECHNOLOGY<\/data> A knowledge graph is a structured representation of information, used in the Graph RAG approach for summarization<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> TECHNOLOGY<\/data> Algorithms used to partition graphs into modular communities of closely-related nodes<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> ALGORITHM<\/data> Louvain is a community detection algorithm used to partition graphs into modular communities<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> ALGORITHM<\/data> Leiden is a community detection algorithm used to partition graphs into modular communities<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> DATASET<\/data> HotPotQA is a dataset used to evaluate the entity extraction prompt with gpt-4-turbo<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> TECHNOLOGY<\/data> GPT-4-Turbo is a version of the GPT-4 model used for entity extraction in the evaluation<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> DATASET<\/data> A dataset consisting of transcripts from podcasts used for analysis<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> DATASET<\/data> A dataset consisting of news articles used for analysis<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> METHOD<\/data> Map-reduce is a method used for query-focused summarization of an entire corpus<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> METRIC<\/data> A target quality used to evaluate the summarization approach, focusing on the completeness of the summary<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> METRIC<\/data> A target quality used to evaluate the summarization approach, focusing on the variety of information in the summary<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> METRIC<\/data> A target quality used to evaluate the summarization approach, focusing on the ability to develop understanding of broad issues and themes<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> METHOD<\/data> Naive RAG is a basic approach to RAG used as a baseline for comparison<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> METHOD<\/data> A method for summarizing source texts using a map-reduce approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> OUTPUT<\/data> Summaries generated from modular communities in the knowledge graph<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> METHOD<\/data> A summarization method that focuses on answering specific queries using the entire corpus<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> OUTPUT<\/data> Questions generated to evaluate the summarization approach, focusing on understanding activities<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> PARAMETER<\/data> The level of detail in community summaries used to answer queries<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> METRIC<\/data> The computational cost measured in tokens used in the summarization process<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> PROCESS<\/data> The high-level process of the Graph RAG approach and pipeline<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> PARAMETER<\/data> Key parameters that influence the design of the Graph RAG approach and pipeline<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> PARAMETER<\/data> <\/node> METHOD<\/data> Global summarization is a method that aims to summarize information from a large dataset or corpus<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> ATTRIBUTE<\/data> Modularity is an inherent quality of graphs that allows them to be partitioned into communities of closely-related nodes<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> OUTPUT<\/data> Descriptions generated from modular communities in the knowledge graph<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> INPUT<\/data> A specific question or request for information that the summarization methods aim to answer<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> DATASET<\/data> A large collection of texts or documents used for analysis and summarization<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> OUTPUT<\/data> Intermediate answers generated from community summaries before being combined into a final global answer<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> OUTPUT<\/data> The comprehensive answer generated by combining all relevant partial answers<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> METHOD<\/data> A method that focuses on generating questions to understand activities from datasets<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> INPUT<\/data> Brief descriptions of datasets used to generate sensemaking questions<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> DATASET<\/data> Datasets that represent real-world information, such as podcast transcripts and news articles<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> PARAMETER<\/data> The level of detail in community summaries used to answer queries<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> METHOD<\/data> A method that summarizes the original source texts directly<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> OUTPUT<\/data> Summaries generated from lower hierarchical levels of the community in the knowledge graph<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> OUTPUT<\/data> Summaries generated from intermediate hierarchical levels of the community in the knowledge graph<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> OUTPUT<\/data> Summaries generated from higher hierarchical levels of the community in the knowledge graph<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/node> 1.0<\/data> Graph RAG is a specific approach to RAG<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Graph RAG uses a knowledge graph for global summarization<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> LLM is used in the Graph RAG approach to generate summaries and answer queries<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Community detection algorithms are used in the Graph RAG approach to partition graphs<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Podcast transcripts are used as a dataset to evaluate the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> News articles are used as a dataset to evaluate the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Comprehensiveness is a target quality used to evaluate the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Diversity is a target quality used to evaluate the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Empowerment is a target quality used to evaluate the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Graph RAG is compared to naive RAG in the evaluation<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Graph RAG is compared to global map-reduce summarization in the evaluation<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Community summaries are generated in the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Query-focused summarization is a method used in the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Activity-centered sensemaking questions are used to evaluate the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Hierarchical level of community summaries is varied to evaluate the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Token costs are measured to evaluate the efficiency of the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Data flow describes the high-level process of the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 2.0<\/data> Design parameters influence the Graph RAG approach and pipeline<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Graph RAG uses global summarization to summarize information from a large dataset<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Graph RAG aims to answer specific queries<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Graph RAG uses a corpus for analysis and summarization<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Activity-centered sensemaking is used to evaluate the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Real-world datasets are used to evaluate the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Hierarchical level of community summaries is varied to evaluate the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Graph RAG is compared to source text summarization in the evaluation<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Low-level community summaries are generated in the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Intermediate-level community summaries are generated in the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> High-level community summaries are generated in the Graph RAG approach<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Modularity is an inherent quality of knowledge graphs<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Louvain is a type of community detection algorithm<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Leiden is a type of community detection algorithm<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> HotPotQA dataset is used to evaluate the entity extraction prompt with gpt-4-turbo<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Map-reduce is used for query-focused summarization of an entire corpus<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Community descriptions are generated from community summaries<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Partial answers are generated from community summaries<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Final global answer is generated by combining all relevant partial answers<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> 1.0<\/data> Short descriptions are used to generate sensemaking questions<\/data> 21e52bc06a82796b1f4bcd73edda1f2a<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"bc9e2c9e369c4108cf4f6dd5f60960f4","chunk":" intermediate- and low-level community summaries shows\nfavorable performance over source text summarization on these same metrics, at lower token costs.\n2 Graph RAG Approach & Pipeline\nWe now unpack the high-level data flow of the Graph RAG approach (Figure 1) and pipeline, de-\nscribing key design parameters, techniques, and implementation details for each step.\n2.1 Source Documents \u2192Text Chunks\nA fundamental design decision is the granularity with which input texts extracted from source doc-\numents should be split into text chunks for processing. In the following step, each of these chunks\nwill be passed to a set of LLM prompts designed to extract the various elements of a graph index.\nLonger text chunks require fewer LLM calls for such extraction, but suffer from the recall degrada-\ntion of longer LLM context windows (Kuratov et al., 2024; Liu et al., 2023). This behavior can be\nobserved in Figure 2 in the case of a single extraction round (i.e., with zero gleanings): on a sample\ndataset (HotPotQA, Yang et al., 2018), using a chunk size of 600 token extracted almost twice as\nmany entity references as when using a chunk size of 2400. While more references are generally\nbetter, any extraction process needs to balance recall and precision for the target activity.\n2.2 Text Chunks \u2192Element Instances\nThe baseline requirement for this step is to identify and extract instances of graph nodes and edges\nfrom each chunk of source text. We do this using a multipart LLM prompt that first identifies all\nentities in the text, including their name, type, and description, before identifying all relationships\nbetween clearly-related entities, including the source and target entities and a description of their\nrelationship. Both kinds of element instance are output in a single list of delimited tuples.\nThe primary opportunity to tailor this prompt to the domain of the document corpus lies in the\nchoice of few-shot examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of \u201cnamed entities\u201d like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt","chunk_id":"bc9e2c9e369c4108cf4f6dd5f60960f4","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"GRAPH RAG","type":"METHOD, TECHNOLOGY","description":"Graph RAG (Retrieval-Augmented Generation) is an approach that involves a high-level data flow and pipeline for processing and summarizing text","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"PIPELINE","type":"PROCESS, SYSTEM","description":"The pipeline refers to the sequence of steps and processes involved in the Graph RAG approach","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"SOURCE DOCUMENTS","type":"DATA, INPUT","description":"Source documents are the original texts from which input texts are extracted for processing in the Graph RAG approach","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"TEXT CHUNKS","type":"DATA, UNIT","description":"Text chunks are segments of input texts extracted from source documents, used for processing in the Graph RAG approach","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"LLM","type":"TECHNOLOGY, METHOD","description":"LLM (Large Language Model) is used to process text chunks and extract elements of a graph index","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"GRAPH INDEX","type":"DATA STRUCTURE, OUTPUT","description":"Graph index is a data structure that includes various elements extracted from text chunks using LLM prompts","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"HOTPOTQA","type":"DATASET","description":"HotPotQA is a dataset used to observe the behavior of text chunk extraction in the Graph RAG approach","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"ENTITY REFERENCES","type":"DATA, UNIT","description":"Entity references are mentions of entities within text chunks, extracted during the processing","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"RECALL","type":"METRIC","description":"Recall is a metric used to measure the completeness of entity extraction from text chunks","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"PRECISION","type":"METRIC","description":"Precision is a metric used to measure the accuracy of entity extraction from text chunks","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"ELEMENT INSTANCES","type":"DATA, UNIT","description":"Element instances are identified and extracted instances of graph nodes and edges from text chunks","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"FEW-SHOT EXAMPLES","type":"TECHNIQUE, METHOD","description":"Few-shot examples are sample inputs provided to the LLM for in-context learning to tailor the extraction prompt to the document corpus domain","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"NAMED ENTITIES","type":"DATA, UNIT","description":"Named entities are specific types of entities such as people, places, and organizations, extracted from text chunks","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"COVARIATES","type":"DATA, ATTRIBUTE","description":"Covariates are additional attributes associated with extracted node instances in the graph index","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"KURATOV ET AL., 2024","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Kuratov et al. in 2024, discussing the recall degradation of longer LLM context windows","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"LIU ET AL., 2023","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Liu et al. in 2023, discussing the recall degradation of longer LLM context windows","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"YANG ET AL., 2018","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Yang et al. in 2018, introducing the HotPotQA dataset","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"BROWN ET AL., 2020","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Brown et al. in 2020, discussing in-context learning with few-shot examples","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"INTERMEDIATE-LEVEL COMMUNITY SUMMARIES","type":"DATA, UNIT","description":"Intermediate-level community summaries are summaries that provide a mid-level overview of the source text","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"LOW-LEVEL COMMUNITY SUMMARIES","type":"DATA, UNIT","description":"Low-level community summaries are summaries that provide a detailed overview of the source text","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"TOKEN COSTS","type":"METRIC","description":"Token costs refer to the number of tokens required for processing text in the Graph RAG approach","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"DESIGN PARAMETERS","type":"ATTRIBUTE, CONFIGURATION","description":"Design parameters are key settings and configurations in the Graph RAG approach","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"TECHNIQUES","type":"METHOD, APPROACH","description":"Techniques refer to the specific methods used in the Graph RAG approach","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"IMPLEMENTATION DETAILS","type":"ATTRIBUTE, CONFIGURATION","description":"Implementation details are specific configurations and settings used in the Graph RAG approach","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"LLM PROMPTS","type":"TECHNIQUE, METHOD","description":"LLM prompts are specific instructions given to the LLM to extract elements from text chunks","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"SINGLE EXTRACTION ROUND","type":"PROCESS, METHOD","description":"A single extraction round refers to one complete cycle of extracting elements from text chunks using LLM prompts","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"CHUNK SIZE","type":"ATTRIBUTE, CONFIGURATION","description":"Chunk size refers to the length of text chunks used in the extraction process","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"RECALL DEGRADATION","type":"METRIC, ISSUE","description":"Recall degradation refers to the decrease in recall performance when using longer LLM context windows","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"EXTRACTION PROCESS","type":"PROCESS, METHOD","description":"The extraction process involves identifying and extracting elements from text chunks using LLM prompts","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"DOMAIN","type":"ATTRIBUTE, CONFIGURATION","description":"Domain refers to the specific area of knowledge or field to which the document corpus belongs","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"DOCUMENT CORPUS","type":"DATA, INPUT","description":"Document corpus refers to the collection of documents being processed in the Graph RAG approach","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"DEFAULT PROMPT","type":"TECHNIQUE, METHOD","description":"Default prompt is the standard set of instructions given to the LLM for extracting named entities","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"},{"name":"SECONDARY EXTRACTION PROMPT","type":"TECHNIQUE, METHOD","description":"Secondary extraction prompt is an additional set of instructions given to the LLM for extracting covariates","source_id":"bc9e2c9e369c4108cf4f6dd5f60960f4"}],"entity_graph":" METHOD, TECHNOLOGY<\/data> Graph RAG (Retrieval-Augmented Generation) is an approach that involves a high-level data flow and pipeline for processing and summarizing text<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> PROCESS, SYSTEM<\/data> The pipeline refers to the sequence of steps and processes involved in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> DATA, INPUT<\/data> Source documents are the original texts from which input texts are extracted for processing in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> DATA, UNIT<\/data> Text chunks are segments of input texts extracted from source documents, used for processing in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> TECHNOLOGY, METHOD<\/data> LLM (Large Language Model) is used to process text chunks and extract elements of a graph index<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> DATA STRUCTURE, OUTPUT<\/data> Graph index is a data structure that includes various elements extracted from text chunks using LLM prompts<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> DATASET<\/data> HotPotQA is a dataset used to observe the behavior of text chunk extraction in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> DATA, UNIT<\/data> Entity references are mentions of entities within text chunks, extracted during the processing<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> METRIC<\/data> Recall is a metric used to measure the completeness of entity extraction from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> METRIC<\/data> Precision is a metric used to measure the accuracy of entity extraction from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> DATA, UNIT<\/data> Element instances are identified and extracted instances of graph nodes and edges from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> TECHNIQUE, METHOD<\/data> Few-shot examples are sample inputs provided to the LLM for in-context learning to tailor the extraction prompt to the document corpus domain<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> DATA, UNIT<\/data> Named entities are specific types of entities such as people, places, and organizations, extracted from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> DATA, ATTRIBUTE<\/data> Covariates are additional attributes associated with extracted node instances in the graph index<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Kuratov et al. in 2024, discussing the recall degradation of longer LLM context windows<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Liu et al. in 2023, discussing the recall degradation of longer LLM context windows<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Yang et al. in 2018, introducing the HotPotQA dataset<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Brown et al. in 2020, discussing in-context learning with few-shot examples<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> DATA, UNIT<\/data> Intermediate-level community summaries are summaries that provide a mid-level overview of the source text<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> DATA, UNIT<\/data> Low-level community summaries are summaries that provide a detailed overview of the source text<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> METRIC<\/data> Token costs refer to the number of tokens required for processing text in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> ATTRIBUTE, CONFIGURATION<\/data> Design parameters are key settings and configurations in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> METHOD, APPROACH<\/data> Techniques refer to the specific methods used in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> ATTRIBUTE, CONFIGURATION<\/data> Implementation details are specific configurations and settings used in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> TECHNIQUE, METHOD<\/data> LLM prompts are specific instructions given to the LLM to extract elements from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> PROCESS, METHOD<\/data> A single extraction round refers to one complete cycle of extracting elements from text chunks using LLM prompts<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> ATTRIBUTE, CONFIGURATION<\/data> Chunk size refers to the length of text chunks used in the extraction process<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> METRIC, ISSUE<\/data> Recall degradation refers to the decrease in recall performance when using longer LLM context windows<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> PROCESS, METHOD<\/data> The extraction process involves identifying and extracting elements from text chunks using LLM prompts<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> ATTRIBUTE, CONFIGURATION<\/data> Domain refers to the specific area of knowledge or field to which the document corpus belongs<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> DATA, INPUT<\/data> Document corpus refers to the collection of documents being processed in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> TECHNIQUE, METHOD<\/data> Default prompt is the standard set of instructions given to the LLM for extracting named entities<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> TECHNIQUE, METHOD<\/data> Secondary extraction prompt is an additional set of instructions given to the LLM for extracting covariates<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/node> 1.0<\/data> The Graph RAG approach involves a specific pipeline for processing and summarizing text<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Token costs are a consideration in the performance of the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Design parameters are key settings in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Techniques are specific methods used in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Implementation details are specific configurations used in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Text chunks are extracted from source documents for processing in the Graph RAG approach<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Intermediate-level community summaries are derived from source documents<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Low-level community summaries are derived from source documents<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Document corpus consists of source documents being processed<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Text chunks are processed using LLM to extract elements of a graph index<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Entity references are extracted from text chunks during processing<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Element instances are extracted from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Chunk size refers to the length of text chunks used in the extraction process<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> LLM is used to extract elements of a graph index from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Few-shot examples are provided to the LLM for in-context learning to tailor the extraction prompt<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> LLM extracts named entities from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Kuratov et al. (2024) discuss the recall degradation of longer LLM context windows<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Liu et al. (2023) discuss the recall degradation of longer LLM context windows<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> LLM prompts are instructions given to the LLM for extracting elements from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Recall degradation occurs with longer LLM context windows<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> The extraction process involves using LLM to identify and extract elements from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Default prompt is the standard set of instructions given to the LLM<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Secondary extraction prompt is an additional set of instructions given to the LLM<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Yang et al. (2018) introduced the HotPotQA dataset<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Recall measures the completeness of entity references extracted from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Precision measures the accuracy of entity references extracted from text chunks<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Covariates are additional attributes associated with extracted element instances<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Brown et al. (2020) discuss in-context learning with few-shot examples<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Few-shot examples are used to tailor the default prompt to the domain<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Few-shot examples are used to tailor the secondary extraction prompt to the domain<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> A single extraction round is part of the extraction process<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> 1.0<\/data> Domain refers to the specific area of knowledge of the document corpus<\/data> bc9e2c9e369c4108cf4f6dd5f60960f4<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"2c6ed90897310eea2f28e33fff1c32b0","chunk":", and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of \u201cgleanings\u201d, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities were extracted, using a logit bias of 100 to force a yes\/no decision. If the LLM\nresponds that entities were missed, then a continuation indicating that \u201cMANY entities were missed\nin the last extraction\u201d encourages the LLM to glean these missing entities. This approach allows us\nto use larger chunk sizes without a drop in quality (Figure 2) or the forced introduction of noise.\n2.3 Element Instances \u2192Element Summaries\nThe use of an LLM to \u201cextract\u201d descriptions of entities, relationships, and claims represented in\nsource texts is already a form of abstractive summarization, relying on the LLM to create inde-\npendently meaningful summaries of concepts that may be implied but not stated by the text itself\n(e.g., the presence of implied relationships). To convert all such instance-level summaries into sin-\ngle blocks of descriptive text for each graph element (i.e., entity node, relationship edge, and claim\ncovariate) requires a further round of LLM summarization over matching groups of instances.\nA potential concern at this stage is that the LLM may not consistently extract references to the\nsame entity in the same text format, resulting in duplicate entity elements and thus duplicate nodes\nin the entity graph. However, since all closely-related \u201ccommunities\u201d of entities will be detected\nand summarized in the following step, and given that LLMs can understand the common entity\nbehind multiple name variations, our overall approach is resilient to such variations provided there\nis sufficient connectivity from all variations to a shared set of closely-related entities.\nOverall, our use of rich descriptive text for homogeneous nodes in a potentially noisy graph structure\nis aligned with both the capabilities of LLMs and","chunk_id":"2c6ed90897310eea2f28e33fff1c32b0","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"LLM","type":"TECHNOLOGY","description":"LLM (Large Language Model) is a type of artificial intelligence used for tasks such as entity extraction, summarization, and understanding relationships in text","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"FEW-SHOT EXAMPLES","type":"METHOD","description":"Few-shot examples are specialized instances provided to the LLM to improve its performance in domains with specialized knowledge such as science, medicine, and law","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"COVARIATE PROMPT","type":"METHOD","description":"A covariate prompt is used to extract additional attributes associated with detected entities, including claims linked to the entities","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"CLAIMS","type":"CONCEPT","description":"Claims are statements or assertions linked to detected entities, including details such as subject, object, type, description, source text span, and start and end dates","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"GLEANINGS","type":"METHOD","description":"Gleanings refer to multiple rounds of entity extraction to ensure that no entities are missed in the process","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"LOGIT BIAS","type":"TECHNIQUE","description":"Logit bias is a technique used to force a yes\/no decision from the LLM during the entity extraction process","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"ELEMENT INSTANCES","type":"CONCEPT","description":"Element instances are individual occurrences of entities, relationships, and claims extracted from source texts","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"ELEMENT SUMMARIES","type":"CONCEPT","description":"Element summaries are descriptive texts created by the LLM to summarize entities, relationships, and claims extracted from source texts","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"ENTITY NODE","type":"CONCEPT","description":"An entity node is a representation of an entity in a graph structure","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"RELATIONSHIP EDGE","type":"CONCEPT","description":"A relationship edge is a representation of a relationship between entities in a graph structure","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"CLAIM COVARIATE","type":"CONCEPT","description":"A claim covariate is an additional attribute or variable associated with a claim in a graph structure","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"COMMUNITIES OF ENTITIES","type":"CONCEPT","description":"Communities of entities are groups of closely-related entities detected and summarized by the LLM","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"NOISY GRAPH STRUCTURE","type":"CONCEPT","description":"A noisy graph structure is a graph that may contain duplicate or inconsistent entity elements due to variations in text format","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"SCIENCE","type":"DOMAIN","description":"Science is a specialized domain that benefits from few-shot examples to improve LLM performance","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"MEDICINE","type":"DOMAIN","description":"Medicine is a specialized domain that benefits from few-shot examples to improve LLM performance","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"LAW","type":"DOMAIN","description":"Law is a specialized domain that benefits from few-shot examples to improve LLM performance","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"SOURCE TEXT SPAN","type":"ATTRIBUTE","description":"Source text span is an attribute of a claim that indicates the specific portion of text from which the claim was extracted","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"START DATE","type":"ATTRIBUTE","description":"Start date is an attribute of a claim that indicates when the event or fact described in the claim began","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"END DATE","type":"ATTRIBUTE","description":"End date is an attribute of a claim that indicates when the event or fact described in the claim ended","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"DESCRIPTION","type":"ATTRIBUTE","description":"Description is an attribute of a claim that provides a detailed explanation of the claim","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"SUBJECT","type":"ATTRIBUTE","description":"Subject is an attribute of a claim that indicates the main entity involved in the claim","source_id":"2c6ed90897310eea2f28e33fff1c32b0"},{"name":"OBJECT","type":"ATTRIBUTE","description":"Object is an attribute of a claim that indicates the entity that is affected by the subject in the claim","source_id":"2c6ed90897310eea2f28e33fff1c32b0"}],"entity_graph":" TECHNOLOGY<\/data> LLM (Large Language Model) is a type of artificial intelligence used for tasks such as entity extraction, summarization, and understanding relationships in text<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> METHOD<\/data> Few-shot examples are specialized instances provided to the LLM to improve its performance in domains with specialized knowledge such as science, medicine, and law<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> METHOD<\/data> A covariate prompt is used to extract additional attributes associated with detected entities, including claims linked to the entities<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> CONCEPT<\/data> Claims are statements or assertions linked to detected entities, including details such as subject, object, type, description, source text span, and start and end dates<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> METHOD<\/data> Gleanings refer to multiple rounds of entity extraction to ensure that no entities are missed in the process<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> TECHNIQUE<\/data> Logit bias is a technique used to force a yes\/no decision from the LLM during the entity extraction process<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> CONCEPT<\/data> Element instances are individual occurrences of entities, relationships, and claims extracted from source texts<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> CONCEPT<\/data> Element summaries are descriptive texts created by the LLM to summarize entities, relationships, and claims extracted from source texts<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> CONCEPT<\/data> An entity node is a representation of an entity in a graph structure<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> CONCEPT<\/data> A relationship edge is a representation of a relationship between entities in a graph structure<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> CONCEPT<\/data> A claim covariate is an additional attribute or variable associated with a claim in a graph structure<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> CONCEPT<\/data> Communities of entities are groups of closely-related entities detected and summarized by the LLM<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> CONCEPT<\/data> A noisy graph structure is a graph that may contain duplicate or inconsistent entity elements due to variations in text format<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> DOMAIN<\/data> Science is a specialized domain that benefits from few-shot examples to improve LLM performance<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> DOMAIN<\/data> Medicine is a specialized domain that benefits from few-shot examples to improve LLM performance<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> DOMAIN<\/data> Law is a specialized domain that benefits from few-shot examples to improve LLM performance<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> ATTRIBUTE<\/data> Source text span is an attribute of a claim that indicates the specific portion of text from which the claim was extracted<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> ATTRIBUTE<\/data> Start date is an attribute of a claim that indicates when the event or fact described in the claim began<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> ATTRIBUTE<\/data> End date is an attribute of a claim that indicates when the event or fact described in the claim ended<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> ATTRIBUTE<\/data> Description is an attribute of a claim that provides a detailed explanation of the claim<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> ATTRIBUTE<\/data> Subject is an attribute of a claim that indicates the main entity involved in the claim<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> ATTRIBUTE<\/data> Object is an attribute of a claim that indicates the entity that is affected by the subject in the claim<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/node> 1.0<\/data> Few-shot examples are used to improve the performance of the LLM in specialized domains<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> The LLM uses covariate prompts to extract additional attributes associated with detected entities<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> The LLM uses multiple rounds of gleanings to ensure no entities are missed<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Logit bias is used to force a yes\/no decision from the LLM during entity extraction<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> The LLM extracts element instances from source texts<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> The LLM detects and summarizes communities of entities<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Few-shot examples are used to improve LLM performance in the domain of science<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Few-shot examples are used to improve LLM performance in the domain of medicine<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Few-shot examples are used to improve LLM performance in the domain of law<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Covariate prompts are used to extract claims linked to detected entities<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Source text span is an attribute of claims<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Start date is an attribute of claims<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> End date is an attribute of claims<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Description is an attribute of claims<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Subject is an attribute of claims<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Object is an attribute of claims<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Element instances are converted into element summaries by the LLM<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Element summaries include descriptions of entity nodes<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Element summaries include descriptions of relationship edges<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Element summaries include descriptions of claim covariates<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> 1.0<\/data> Communities of entities help manage variations in a noisy graph structure<\/data> 2c6ed90897310eea2f28e33fff1c32b0<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"7fb7d9ce2da9c940a32afdd87d1d9e56","chunk":" common entity\nbehind multiple name variations, our overall approach is resilient to such variations provided there\nis sufficient connectivity from all variations to a shared set of closely-related entities.\nOverall, our use of rich descriptive text for homogeneous nodes in a potentially noisy graph structure\nis aligned with both the capabilities of LLMs and the needs of global, query-focused summarization.\nThese qualities also differentiate our graph index from typical knowledge graphs, which rely on\nconcise and consistent knowledge triples (subject, predicate, object) for downstream reasoning tasks.\n2.4 Element Summaries \u2192Graph Communities\nThe index created in the previous step can be modelled as an homogeneous undirected weighted\ngraph in which entity nodes are connected by relationship edges, with edge weights representing the\nnormalized counts of detected relationship instances. Given such a graph, a variety of community\ndetection algorithms may be used to partition the graph into communities of nodes with stronger\nconnections to one another than to the other nodes in the graph (e.g., see the surveys by Fortu-\nnato, 2010 and Jin et al., 2021). In our pipeline, we use Leiden (Traag et al., 2019) on account of\nits ability to recover hierarchical community structure of large-scale graphs efficiently (Figure 3).\nEach level of this hierarchy provides a community partition that covers the nodes of the graph in a\nmutually-exclusive, collective-exhaustive way, enabling divide-and-conquer global summarization.\n2.5 Graph Communities \u2192Community Summaries\nThe next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T","chunk_id":"7fb7d9ce2da9c940a32afdd87d1d9e56","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"COMMON ENTITY","type":"CONCEPT","description":"A concept referring to an entity that has multiple name variations but is resilient to such variations due to sufficient connectivity to closely-related entities","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"LLMS","type":"TECHNOLOGY","description":"Large Language Models (LLMs) are advanced AI models capable of understanding and generating human-like text","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"KNOWLEDGE GRAPHS","type":"TECHNOLOGY","description":"Structured representations of knowledge in the form of triples (subject, predicate, object) used for reasoning tasks","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"HOMOGENEOUS NODES","type":"CONCEPT","description":"Nodes in a graph that are of the same type and are described using rich descriptive text","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"NOISY GRAPH STRUCTURE","type":"CONCEPT","description":"A graph structure that may contain inconsistencies or errors, making it challenging to analyze","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"GRAPH INDEX","type":"TECHNOLOGY","description":"An index created from a graph structure, used for query-focused summarization and other tasks","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"RELATIONSHIP EDGES","type":"CONCEPT","description":"Edges in a graph that represent relationships between entity nodes","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"EDGE WEIGHTS","type":"METRIC","description":"Weights assigned to edges in a graph, representing the normalized counts of detected relationship instances","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"COMMUNITY DETECTION ALGORITHMS","type":"TECHNOLOGY","description":"Algorithms used to partition a graph into communities of nodes with stronger connections to one another","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"LEIDEN","type":"TECHNOLOGY","description":"A community detection algorithm known for its ability to recover hierarchical community structure efficiently","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"HIERARCHICAL COMMUNITY STRUCTURE","type":"CONCEPT","description":"A structure in which communities are organized in a hierarchy, with each level providing a partition of the graph nodes","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"COMMUNITY PARTITION","type":"CONCEPT","description":"A division of graph nodes into mutually-exclusive, collectively-exhaustive communities","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"GLOBAL SUMMARIZATION","type":"TECHNOLOGY","description":"A method for summarizing the overall structure and semantics of a dataset","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"COMMUNITY SUMMARIES","type":"TECHNOLOGY","description":"Report-like summaries of each community in a hierarchical structure, useful for understanding the dataset","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"MULTIHOP-RAG","type":"TECHNOLOGY","description":"A specific implementation or variant of Retrieval-Augmented Generation (RAG) used in the context of graph communities","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"FORTUNATO","type":"PERSON","description":"An author who has conducted surveys on community detection algorithms","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"JIN ET AL.","type":"PERSON","description":"Authors who have conducted surveys on community detection algorithms","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"TRAAG ET AL.","type":"PERSON","description":"Authors of the Leiden algorithm, known for its efficiency in recovering hierarchical community structures","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"DATASET","type":"CONCEPT","description":"A collection of data used for analysis and summarization","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"QUERY-FOCUSED SUMMARIZATION","type":"TECHNOLOGY","description":"A method for summarizing information based on specific queries","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"GLOBAL QUERIES","type":"CONCEPT","description":"Queries that aim to retrieve information from a global perspective, covering the entire dataset","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"ROOT COMMUNITIES","type":"CONCEPT","description":"The top-level communities in a hierarchical community structure","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"SUB-COMMUNITIES","type":"CONCEPT","description":"Lower-level communities in a hierarchical community structure, providing more detailed information","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"REPORTS","type":"TECHNOLOGY","description":"Detailed documents that provide information about specific subtopics within a community","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"ELEMENT SUMMARIES","type":"TECHNOLOGY","description":"Summaries of elements within a graph, used to understand the structure and semantics of the dataset","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"GRAPH COMMUNITIES","type":"TECHNOLOGY","description":"Groups of nodes within a graph that have stronger connections to each other than to other nodes","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"COMMUNITY DETECTION","type":"TECHNOLOGY","description":"The process of identifying communities within a graph","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"PARTITION","type":"CONCEPT","description":"The division of a graph into distinct communities","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"HIERARCHY","type":"CONCEPT","description":"A system in which elements are ranked or organized in levels","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"PIPELINE","type":"TECHNOLOGY","description":"A series of processes or steps used to analyze and summarize a dataset","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"LEVEL 0","type":"CONCEPT","description":"The root level in a hierarchical community structure","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"LEVEL 1","type":"CONCEPT","description":"A sub-level in a hierarchical community structure, providing more detailed information","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"},{"name":"FIGURE 3","type":"CONCEPT","description":"A visual representation of graph communities detected using the Leiden algorithm","source_id":"7fb7d9ce2da9c940a32afdd87d1d9e56"}],"entity_graph":" CONCEPT<\/data> A concept referring to an entity that has multiple name variations but is resilient to such variations due to sufficient connectivity to closely-related entities<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> Large Language Models (LLMs) are advanced AI models capable of understanding and generating human-like text<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> Structured representations of knowledge in the form of triples (subject, predicate, object) used for reasoning tasks<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> Nodes in a graph that are of the same type and are described using rich descriptive text<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> A graph structure that may contain inconsistencies or errors, making it challenging to analyze<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> An index created from a graph structure, used for query-focused summarization and other tasks<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> Edges in a graph that represent relationships between entity nodes<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> METRIC<\/data> Weights assigned to edges in a graph, representing the normalized counts of detected relationship instances<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> Algorithms used to partition a graph into communities of nodes with stronger connections to one another<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> A community detection algorithm known for its ability to recover hierarchical community structure efficiently<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> A structure in which communities are organized in a hierarchy, with each level providing a partition of the graph nodes<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> A division of graph nodes into mutually-exclusive, collectively-exhaustive communities<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> A method for summarizing the overall structure and semantics of a dataset<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> Report-like summaries of each community in a hierarchical structure, useful for understanding the dataset<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> A specific implementation or variant of Retrieval-Augmented Generation (RAG) used in the context of graph communities<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> PERSON<\/data> An author who has conducted surveys on community detection algorithms<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> PERSON<\/data> Authors who have conducted surveys on community detection algorithms<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> PERSON<\/data> Authors of the Leiden algorithm, known for its efficiency in recovering hierarchical community structures<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> A collection of data used for analysis and summarization<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> A method for summarizing information based on specific queries<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> Queries that aim to retrieve information from a global perspective, covering the entire dataset<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> The top-level communities in a hierarchical community structure<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> Lower-level communities in a hierarchical community structure, providing more detailed information<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> Detailed documents that provide information about specific subtopics within a community<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> Summaries of elements within a graph, used to understand the structure and semantics of the dataset<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> Groups of nodes within a graph that have stronger connections to each other than to other nodes<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> The process of identifying communities within a graph<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> The division of a graph into distinct communities<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> A system in which elements are ranked or organized in levels<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> TECHNOLOGY<\/data> A series of processes or steps used to analyze and summarize a dataset<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> The root level in a hierarchical community structure<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> A sub-level in a hierarchical community structure, providing more detailed information<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> CONCEPT<\/data> A visual representation of graph communities detected using the Leiden algorithm<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/node> 1.0<\/data> Common entities are described using rich descriptive text for homogeneous nodes<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> The use of rich descriptive text for homogeneous nodes in a graph index aligns with the capabilities of LLMs<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Graph indexes differ from typical knowledge graphs in their use of rich descriptive text instead of concise knowledge triples<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Relationship edges connect homogeneous nodes in a graph<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Community detection algorithms are used to partition the graph index into communities<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Edge weights represent the normalized counts of detected relationship instances on relationship edges<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Leiden is a specific community detection algorithm used in the pipeline<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Fortunato has conducted surveys on community detection algorithms<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Jin et al. have conducted surveys on community detection algorithms<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Leiden is known for its ability to recover hierarchical community structures efficiently<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> The Leiden algorithm is used to detect graph communities in the MultiHop-RAG<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Traag et al. are the authors of the Leiden algorithm<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Figure 3 shows graph communities detected using the Leiden algorithm<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Each level of the hierarchical community structure provides a community partition<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Community summaries are created for each level in the hierarchical community structure<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Root communities are the top-level communities in a hierarchical community structure<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Sub-communities are lower-level communities in a hierarchical community structure<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Community partitions enable divide-and-conquer global summarization<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Community summaries are useful for understanding the global structure and semantics of the dataset<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Query-focused summarization is used for answering global queries<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Reports provide detailed information about specific subtopics within sub-communities<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Element summaries are used to understand the structure and semantics of graph communities<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Graph communities are identified through community detection<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Community detection results in the partition of a graph into distinct communities<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> The pipeline includes a step for community detection<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Partitions can be organized into a hierarchy<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Level 0 is the root level in a hierarchy<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> 1.0<\/data> Level 1 is a sub-level in a hierarchy<\/data> 7fb7d9ce2da9c940a32afdd87d1d9e56<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"843fc5421e086120ffa1c75856ecf6cd","chunk":" answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (Tang and Yang, 2024) dataset as indexed. Circles represent entity nodes with size\nproportional to their degree. Node layout was performed via OpenORD (Martin et al., 2011) and\nForce Atlas 2 (Jacomy et al., 2014). Node colors represent entity communities, shown at two levels\nof hierarchical clustering: (a) Level 0, corresponding to the hierarchical partition with maximum\nmodularity, and (b) Level 1, which reveals internal structure within these root-level communities.\n\u2022Leaf-level communities . The element summaries of a leaf-level community (nodes, edges,\ncovariates) are prioritized and then iteratively added to the LLM context window until\nthe token limit is reached. The prioritization is as follows: for each community edge in\ndecreasing order of combined source and target node degree (i.e., overall prominance), add\ndescriptions of the source node, target node, linked covariates, and the edge itself.\n\u2022Higher-level communities . If all element summaries fit within the token limit of the con-\ntext window, proceed as for leaf-level communities and summarize all element summaries\nwithin the community. Otherwise, rank sub-communities in decreasing order of element\nsummary tokens and iteratively substitute sub-community summaries (shorter) for their\nassociated element summaries (longer) until fit within the context window is achieved.\n2.6 Community Summaries \u2192Community Answers \u2192Global Answer\nGiven a user query, the community summaries generated in the previous step can be used to generate\na final answer in a multi-stage process. The hierarchical nature of the community structure also\nmeans that questions can be answered using the community summaries from different levels, raising\nthe question of whether a particular level in the hierarchical community structure offers the best\nbalance of summary detail and scope for general sensemaking questions (evaluated in section 3).\nFor a given community level, the global answer to any user query is generated as follows:\n\u2022Prepare community summaries . Community summaries are randomly shuffled and divided\ninto chunks of pre-specified token size. This ensures relevant information is distributed\nacross chunks, rather than concentrated (and potentially lost)","chunk_id":"843fc5421e086120ffa1c75856ecf6cd","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"GLOBAL QUERIES","type":"CONCEPT","description":"Global queries refer to questions or inquiries that require comprehensive answers derived from multiple sources or datasets","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"COMMUNITY SUMMARIES","type":"CONCEPT","description":"Community summaries are generated summaries of data clusters or communities, used to answer queries","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"ROOT COMMUNITIES","type":"CONCEPT","description":"Root communities are the top-level clusters in a hierarchical community structure","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"SUB-COMMUNITIES","type":"CONCEPT","description":"Sub-communities are lower-level clusters within root communities in a hierarchical community structure","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"LEIDEN ALGORITHM","type":"METHOD","description":"The Leiden algorithm is a method used for detecting communities in large networks","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"MULTIHOP-RAG","type":"DATASET","description":"MultiHop-RAG is a dataset used for community detection and analysis","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"OPENORD","type":"TOOL","description":"OpenORD is a tool used for node layout in graph visualizations","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"FORCE ATLAS 2","type":"TOOL","description":"Force Atlas 2 is a tool used for node layout in graph visualizations","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"NODE","type":"ELEMENT","description":"Nodes represent entities in a graph, with size proportional to their degree","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"EDGE","type":"ELEMENT","description":"Edges represent connections between nodes in a graph","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"COVARIATE","type":"ELEMENT","description":"Covariates are variables that are linked to nodes and edges in a graph","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"LLM CONTEXT WINDOW","type":"CONCEPT","description":"The LLM context window is the token limit within which summaries are added for processing by a language model","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"HIERARCHICAL CLUSTERING","type":"METHOD","description":"Hierarchical clustering is a method of clustering data into a tree-like structure with multiple levels","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"COMMUNITY ANSWERS","type":"CONCEPT","description":"Community answers are responses generated from community summaries to answer user queries","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"GLOBAL ANSWER","type":"CONCEPT","description":"A global answer is a comprehensive response generated from multiple community summaries to answer a user query","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"TOKEN LIMIT","type":"CONCEPT","description":"The token limit is the maximum number of tokens that can be processed in a single context window by a language model","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"SUMMARY DETAIL","type":"CONCEPT","description":"Summary detail refers to the level of detail provided in a summary","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"SCOPE","type":"CONCEPT","description":"Scope refers to the range or extent of information covered in a summary","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"SENSEMAKING","type":"CONCEPT","description":"Sensemaking is the process of understanding and making sense of complex information","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"USER QUERY","type":"CONCEPT","description":"A user query is a question or inquiry posed by a user seeking information","source_id":"843fc5421e086120ffa1c75856ecf6cd","entity_type":"CONCEPT"},{"name":"CHUNK","type":"ELEMENT","description":"Chunks are segments of community summaries divided into pre-specified token sizes","source_id":"843fc5421e086120ffa1c75856ecf6cd","entity_type":"ELEMENT"},{"name":"LEVEL 0","type":"CATEGORY","description":"Level 0 represents the root-level communities in the hierarchical clustering with maximum modularity","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"LEVEL 1","type":"CATEGORY","description":"Level 1 represents sub-communities within the root-level communities, revealing internal structure","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"ELEMENT SUMMARIES","type":"CONCEPT","description":"Element summaries are detailed descriptions of nodes, edges, and covariates within a community","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"PROMINENCE","type":"METRIC","description":"Prominence is a metric used to prioritize community edges based on the combined degree of source and target nodes","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"COMBINED SOURCE AND TARGET NODE DEGREE","type":"METRIC","description":"Combined source and target node degree is a metric used to measure the overall prominence of community edges","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"COMMUNITY EDGE","type":"ELEMENT","description":"Community edges are connections between nodes within a community, prioritized based on prominence","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"SUB-COMMUNITY SUMMARIES","type":"CONCEPT","description":"Sub-community summaries are shorter summaries of sub-communities used when element summaries exceed the token limit","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"HIERARCHICAL COMMUNITY STRUCTURE","type":"CONCEPT","description":"Hierarchical community structure is a multi-level clustering of communities used to generate community summaries","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"SUMMARY DETAIL AND SCOPE","type":"CONCEPT","description":"Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking","source_id":"843fc5421e086120ffa1c75856ecf6cd"},{"name":"COMMUNITY LEVEL","type":"CATEGORY","description":"Community level refers to the different levels in the hierarchical community structure used to generate summaries","source_id":"843fc5421e086120ffa1c75856ecf6cd"}],"entity_graph":" CONCEPT<\/data> Global queries refer to questions or inquiries that require comprehensive answers derived from multiple sources or datasets<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> Community summaries are generated summaries of data clusters or communities, used to answer queries<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> Root communities are the top-level clusters in a hierarchical community structure<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> Sub-communities are lower-level clusters within root communities in a hierarchical community structure<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> METHOD<\/data> The Leiden algorithm is a method used for detecting communities in large networks<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> DATASET<\/data> MultiHop-RAG is a dataset used for community detection and analysis<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> TOOL<\/data> OpenORD is a tool used for node layout in graph visualizations<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> TOOL<\/data> Force Atlas 2 is a tool used for node layout in graph visualizations<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> ELEMENT<\/data> Nodes represent entities in a graph, with size proportional to their degree<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> ELEMENT<\/data> Edges represent connections between nodes in a graph<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> ELEMENT<\/data> Covariates are variables that are linked to nodes and edges in a graph<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> The LLM context window is the token limit within which summaries are added for processing by a language model<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> METHOD<\/data> Hierarchical clustering is a method of clustering data into a tree-like structure with multiple levels<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> Community answers are responses generated from community summaries to answer user queries<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> A global answer is a comprehensive response generated from multiple community summaries to answer a user query<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> The token limit is the maximum number of tokens that can be processed in a single context window by a language model<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> Summary detail refers to the level of detail provided in a summary<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> Scope refers to the range or extent of information covered in a summary<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> Sensemaking is the process of understanding and making sense of complex information<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> A user query is a question or inquiry posed by a user seeking information<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> CONCEPT<\/data> <\/node> ELEMENT<\/data> Chunks are segments of community summaries divided into pre-specified token sizes<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> ELEMENT<\/data> <\/node> CATEGORY<\/data> Level 0 represents the root-level communities in the hierarchical clustering with maximum modularity<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CATEGORY<\/data> Level 1 represents sub-communities within the root-level communities, revealing internal structure<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> Element summaries are detailed descriptions of nodes, edges, and covariates within a community<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> METRIC<\/data> Prominence is a metric used to prioritize community edges based on the combined degree of source and target nodes<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> METRIC<\/data> Combined source and target node degree is a metric used to measure the overall prominence of community edges<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> ELEMENT<\/data> Community edges are connections between nodes within a community, prioritized based on prominence<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> Sub-community summaries are shorter summaries of sub-communities used when element summaries exceed the token limit<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> Hierarchical community structure is a multi-level clustering of communities used to generate community summaries<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CONCEPT<\/data> Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> CATEGORY<\/data> Community level refers to the different levels in the hierarchical community structure used to generate summaries<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/node> 1.0<\/data> Community summaries are used to answer global queries<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Community summaries are generated from root communities<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Community summaries are generated from sub-communities<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Community summaries are added to the LLM context window until the token limit is reached<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Community answers are generated from community summaries<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Global answers are generated from community summaries<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> The level of summary detail affects the content of community summaries<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> The scope of information affects the content of community summaries<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Community summaries are used for sensemaking<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 2.0<\/data> Community summaries are divided into chunks of pre-specified token size<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Summary detail and scope affect the content of community summaries<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Root communities are identified through hierarchical clustering<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Level 0 represents the root-level communities in the hierarchical clustering<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Root communities are part of the hierarchical community structure<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Sub-communities are identified through hierarchical clustering<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Level 1 represents sub-communities within the root-level communities<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Sub-communities are part of the hierarchical community structure<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> The Leiden algorithm is used to detect communities in the MultiHop-RAG dataset<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> OpenORD is used for node layout in the MultiHop-RAG dataset<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Force Atlas 2 is used for node layout in the MultiHop-RAG dataset<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Nodes represent entities in the MultiHop-RAG dataset<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Edges represent connections between nodes in the MultiHop-RAG dataset<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Covariates are variables linked to nodes and edges in the MultiHop-RAG dataset<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Element summaries include descriptions of nodes<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Element summaries include descriptions of edges<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Element summaries include descriptions of covariates<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> The token limit defines the maximum number of tokens in the LLM context window<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 2.0<\/data> Global answers are generated in response to user queries<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Sub-community summaries are used when element summaries exceed the token limit<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Prominence is used to prioritize community edges<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Combined source and target node degree is used to measure prominence<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> 1.0<\/data> Community levels are part of the hierarchical community structure<\/data> 843fc5421e086120ffa1c75856ecf6cd<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"1d07b4248c2655081c7af0e373bd70c9","chunk":" in section 3).\nFor a given community level, the global answer to any user query is generated as follows:\n\u2022Prepare community summaries . Community summaries are randomly shuffled and divided\ninto chunks of pre-specified token size. This ensures relevant information is distributed\nacross chunks, rather than concentrated (and potentially lost) in a single context window.\n\u2022Map community answers . Generate intermediate answers in parallel, one for each chunk.\nThe LLM is also asked to generate a score between 0-100 indicating how helpful the gen-\nerated answer is in answering the target question. Answers with score 0 are filtered out.\n\u2022Reduce to global answer . Intermediate community answers are sorted in descending order\nof helpfulness score and iteratively added into a new context window until the token limit\nis reached. This final context is used to generate the global answer returned to the user.\n5Dataset Example activity framing and generation of global sensemaking questions\nPodcast\ntranscriptsUser : A tech journalist looking for insights and trends in the tech industry\nTask: Understanding how tech leaders view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\n\u2022Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (","chunk_id":"1d07b4248c2655081c7af0e373bd70c9","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"COMMUNITY SUMMARIES","type":"DATA","description":"Community summaries are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"CHUNKS","type":"DATA","description":"Chunks are segments of community summaries divided based on a pre-specified token size","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"LLM","type":"TECHNOLOGY","description":"LLM (Large Language Model) is used to generate intermediate answers and scores for each chunk","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"HELPFULNESS SCORE","type":"METRIC","description":"A score between 0-100 generated by the LLM to indicate how helpful an answer is in addressing the target question","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"GLOBAL ANSWER","type":"OUTPUT","description":"The final answer generated by combining intermediate community answers based on their helpfulness scores","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"PODCAST TRANSCRIPTS","type":"DATASET","description":"A dataset consisting of compiled transcripts of podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"NEWS ARTICLES","type":"DATASET","description":"A dataset consisting of news articles used for analysis","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"TECH JOURNALIST","type":"USER","description":"A user looking for insights and trends in the tech industry","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"EDUCATOR","type":"USER","description":"A user incorporating current affairs into curricula","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"TECH POLICY","type":"TOPIC","description":"A topic dealing with tech policy and government regulation","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"PRIVACY LAWS","type":"TOPIC","description":"A topic discussing the impact of privacy laws on technology development","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"INNOVATION AND ETHICS","type":"TOPIC","description":"A topic discussing the balance between innovation and ethical considerations","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"POLICY CHANGES","type":"TOPIC","description":"A topic discussing suggested changes to current policies","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"COLLABORATIONS","type":"TOPIC","description":"A topic discussing collaborations between tech companies and governments","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"HEALTH TOPICS","type":"TOPIC","description":"Current topics in health that can be integrated into health education curricula","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"PREVENTIVE MEDICINE","type":"TOPIC","description":"A topic addressing the concepts of preventive medicine and wellness","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"CONTRADICTORY ARTICLES","type":"TOPIC","description":"Examples of health articles that contradict each other","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"PUBLIC HEALTH PRIORITIES","type":"TOPIC","description":"Insights about public health priorities based on news coverage","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"HEALTH LITERACY","type":"TOPIC","description":"The importance of health literacy highlighted through the dataset","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"USER QUERY","type":"INPUT","description":"A query from the user that the system aims to answer","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"INTERMEDIATE ANSWERS","type":"OUTPUT","description":"Answers generated for each chunk of community summaries","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"TOKEN SIZE","type":"METRIC","description":"The pre-specified size of tokens used to divide community summaries into chunks","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"CONTEXT WINDOW","type":"TECHNOLOGY","description":"A window of text used to generate answers, limited by token size","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"KEVIN SCOTT","type":"PERSON","description":"Kevin Scott is the CTO of Microsoft and a participant in the podcast conversations","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"MICROSOFT","type":"ORGANIZATION","description":"Microsoft is a technology company whose CTO, Kevin Scott, participates in the podcast conversations","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"TECHNOLOGY LEADERS","type":"PERSON","description":"Individuals who are leaders in the technology industry and participate in the podcast conversations","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"TASK","type":"INPUT","description":"A specific activity or goal that the user aims to achieve using the datasets","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"QUESTIONS","type":"INPUT","description":"Specific questions generated by the LLM based on the user's task and the target datasets","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"DATASET","type":"DATA","description":"A collection of data used for analysis, such as podcast transcripts or news articles","source_id":"1d07b4248c2655081c7af0e373bd70c9"},{"name":"USER","type":"","description":"","source_id":"1d07b4248c2655081c7af0e373bd70c9"}],"entity_graph":" DATA<\/data> Community summaries are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> DATA<\/data> Chunks are segments of community summaries divided based on a pre-specified token size<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TECHNOLOGY<\/data> LLM (Large Language Model) is used to generate intermediate answers and scores for each chunk<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> METRIC<\/data> A score between 0-100 generated by the LLM to indicate how helpful an answer is in addressing the target question<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> OUTPUT<\/data> The final answer generated by combining intermediate community answers based on their helpfulness scores<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> DATASET<\/data> A dataset consisting of compiled transcripts of podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> DATASET<\/data> A dataset consisting of news articles used for analysis<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> USER<\/data> A user looking for insights and trends in the tech industry<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> USER<\/data> A user incorporating current affairs into curricula<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TOPIC<\/data> A topic dealing with tech policy and government regulation<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TOPIC<\/data> A topic discussing the impact of privacy laws on technology development<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TOPIC<\/data> A topic discussing the balance between innovation and ethical considerations<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TOPIC<\/data> A topic discussing suggested changes to current policies<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TOPIC<\/data> A topic discussing collaborations between tech companies and governments<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TOPIC<\/data> Current topics in health that can be integrated into health education curricula<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TOPIC<\/data> A topic addressing the concepts of preventive medicine and wellness<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TOPIC<\/data> Examples of health articles that contradict each other<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TOPIC<\/data> Insights about public health priorities based on news coverage<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TOPIC<\/data> The importance of health literacy highlighted through the dataset<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> INPUT<\/data> A query from the user that the system aims to answer<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> OUTPUT<\/data> Answers generated for each chunk of community summaries<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> METRIC<\/data> The pre-specified size of tokens used to divide community summaries into chunks<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> TECHNOLOGY<\/data> A window of text used to generate answers, limited by token size<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> PERSON<\/data> Kevin Scott is the CTO of Microsoft and a participant in the podcast conversations<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> ORGANIZATION<\/data> Microsoft is a technology company whose CTO, Kevin Scott, participates in the podcast conversations<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> PERSON<\/data> Individuals who are leaders in the technology industry and participate in the podcast conversations<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> INPUT<\/data> A specific activity or goal that the user aims to achieve using the datasets<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> INPUT<\/data> Specific questions generated by the LLM based on the user's task and the target datasets<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> DATA<\/data> A collection of data used for analysis, such as podcast transcripts or news articles<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/node> 1.0<\/data> Community summaries are divided into chunks<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Community summaries are prepared to answer user queries<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Intermediate answers are generated from community summaries<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> LLM generates intermediate answers and scores for each chunk<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Chunks are divided based on a pre-specified token size<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> LLM generates a helpfulness score for each answer<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Global answer is generated by sorting intermediate answers based on helpfulness scores<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Helpfulness scores are assigned to intermediate answers<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Intermediate answers are combined to form the global answer<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> The final context window is used to generate the global answer<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Tech journalist uses podcast transcripts to look for insights and trends in the tech industry<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Kevin Scott's conversations are part of the podcast transcripts<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Technology leaders participate in the podcast conversations<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Educator uses news articles to incorporate current affairs into curricula<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Tech journalist is interested in episodes dealing with tech policy and government regulation<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Tech journalist is interested in how guests perceive the impact of privacy laws on technology development<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Tech journalist is interested in discussions about the balance between innovation and ethical considerations<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Tech journalist is interested in suggested changes to current policies<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Tech journalist is interested in discussions about collaborations between tech companies and governments<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Educator is interested in current topics in health that can be integrated into health education curricula<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Educator is interested in how news articles address the concepts of preventive medicine and wellness<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Educator is interested in examples of health articles that contradict each other<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Educator is interested in insights about public health priorities based on news coverage<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Educator is interested in highlighting the importance of health literacy through the dataset<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Kevin Scott is the CTO of Microsoft<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> The task is an activity or goal that the user aims to achieve<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Questions are generated based on the user's task<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> 1.0<\/data> Questions are generated based on the target datasets<\/data> 1d07b4248c2655081c7af0e373bd70c9<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"922778ce1cb2fdd6dbab1746c8795620","chunk":"atasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\n\u2022Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\n\u00d7600-token text chunks, with 100-token overlaps between chunks ( \u223c1 million tokens).\n\u2022News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 \u00d7\n600-token text chunks, with 100-token overlaps between chunks ( \u223c1.7 million tokens).\n3.2 Queries\nMany benchmark datasets for open-domain question answering exist, including HotPotQA (Yang\net al., 2018), MultiHop-RAG (Tang and Yang, 2024), and MT-Bench (Zheng et al., 2024). However,\nthe associated question sets target explicit fact retrieval rather than summarization for the purpose\nof data sensemaking, i.e., the process though which people inspect, engage with, and contextualize\ndata within the broader scope of real-world activities (Koesten et al., 2021). Similarly, methods for\nextracting latent summarization queries from source texts also exist (Xu and Lapata, 2021), but such\nextracted questions can target details that betray prior knowledge of the texts.\nTo evaluate the effectiveness of RAG systems for more global sensemaking tasks, we need questions\nthat convey only a high-level understanding of dataset contents, and not the details of specific texts.\nWe used an activity-centered approach to automate the generation of such questions: given a short\ndescription of a dataset, we asked the LLM to identify Npotential users and Ntasks per user,\nthen for each (user, task) combination, we asked the LLM to generate Nquestions that require\nunderstanding of the entire corpus. For our evaluation, a value of N= 5 resulted in 125 test questions\nper dataset. Table 1 shows example questions for each of the two evaluation datasets.\n63.3 Conditions\nWe compare six different conditions in our analysis, including Graph RAG using four levels of graph\ncommunities ( C","chunk_id":"922778ce1cb2fdd6dbab1746c8795620","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"PODCAST TRANSCRIPTS","type":"DATASET","description":"Compiled transcripts of podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders. Size: 1669 \u00d7 600-token text chunks, with 100-token overlaps between chunks, approximately 1 million tokens","source_id":"922778ce1cb2fdd6dbab1746c8795620"},{"name":"NEWS ARTICLES","type":"DATASET","description":"Benchmark dataset comprising news articles published from September 2013 to December 2023 in a range of categories, including entertainment, business, sports, technology, health, and science. Size: 3197 \u00d7 600-token text chunks, with 100-token overlaps between chunks, approximately 1.7 million tokens","source_id":"922778ce1cb2fdd6dbab1746c8795620"},{"name":"HOTPOTQA","type":"DATASET","description":"A benchmark dataset for open-domain question answering, targeting explicit fact retrieval","source_id":"922778ce1cb2fdd6dbab1746c8795620"},{"name":"MULTIHOP-RAG","type":"DATASET","description":"A benchmark dataset comprising news articles published from September 2013 to December 2023 in a range of categories, including entertainment, business, sports, technology, health, and science\nA benchmark dataset for open-domain question answering, targeting explicit fact retrieval","source_id":"922778ce1cb2fdd6dbab1746c8795620","entity_type":"DATASET"},{"name":"MT-BENCH","type":"DATASET","description":"A benchmark dataset for open-domain question answering, targeting explicit fact retrieval","source_id":"922778ce1cb2fdd6dbab1746c8795620","entity_type":"DATASET"},{"name":"DATA SENSEMAKING","type":"PROCESS","description":"The process through which people inspect, engage with, and contextualize data within the broader scope of real-world activities","source_id":"922778ce1cb2fdd6dbab1746c8795620","entity_type":"PROCESS"},{"name":"RAG SYSTEMS","type":"TECHNOLOGY","description":"Retrieval-Augmented Generation systems used for global sensemaking tasks","source_id":"922778ce1cb2fdd6dbab1746c8795620","entity_type":"TECHNOLOGY"},{"name":"LLM","type":"TECHNOLOGY","description":"Large Language Model used to automate the generation of questions for dataset evaluation","source_id":"922778ce1cb2fdd6dbab1746c8795620","entity_type":"TECHNOLOGY"},{"name":"KEVIN SCOTT","type":"PERSON","description":"Microsoft CTO who participates in podcast conversations compiled in the dataset","source_id":"922778ce1cb2fdd6dbab1746c8795620"},{"name":"KOESTEN ET AL.","type":"AUTHORS","description":"Authors of a paper on data sensemaking behaviors","source_id":"922778ce1cb2fdd6dbab1746c8795620","entity_type":"AUTHORS"},{"name":"XU AND LAPATA","type":"AUTHORS","description":"Authors of a paper on methods for extracting latent summarization queries from source texts","source_id":"922778ce1cb2fdd6dbab1746c8795620","entity_type":"AUTHORS"},{"name":"TANG AND YANG","type":"AUTHORS","description":"Authors associated with the MultiHop-RAG dataset","source_id":"922778ce1cb2fdd6dbab1746c8795620"},{"name":"YANG ET AL.","type":"AUTHORS","description":"Authors associated with the HotPotQA dataset","source_id":"922778ce1cb2fdd6dbab1746c8795620","entity_type":"AUTHORS"},{"name":"ZHENG ET AL.","type":"AUTHORS","description":"Authors associated with the MT-Bench dataset","source_id":"922778ce1cb2fdd6dbab1746c8795620","entity_type":"AUTHORS"},{"name":"GRAPH RAG","type":"TECHNOLOGY","description":"A specific implementation of RAG using four levels of graph communities","source_id":"922778ce1cb2fdd6dbab1746c8795620","entity_type":"TECHNOLOGY"},{"name":"LATENT SUMMARIZATION QUERIES","type":"","description":"","source_id":"922778ce1cb2fdd6dbab1746c8795620"},{"name":"BEHIND THE TECH","type":"PODCAST","description":"A podcast series featuring conversations between Kevin Scott and other technology leaders","source_id":"922778ce1cb2fdd6dbab1746c8795620"},{"name":"SCOTT","type":"PERSON","description":"Kevin Scott, Microsoft CTO, who participates in the podcast conversations compiled in the dataset","source_id":"922778ce1cb2fdd6dbab1746c8795620"},{"name":"TANG","type":"PERSON","description":"An author associated with the MultiHop-RAG dataset","source_id":"922778ce1cb2fdd6dbab1746c8795620"},{"name":"YANG","type":"PERSON","description":"An author associated with the MultiHop-RAG dataset","source_id":"922778ce1cb2fdd6dbab1746c8795620"},{"name":"HOTSPOTQA","type":"DATASET","description":"A benchmark dataset for open-domain question answering, targeting explicit fact retrieval","source_id":"922778ce1cb2fdd6dbab1746c8795620"}],"entity_graph":" DATASET<\/data> Compiled transcripts of podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders. Size: 1669 × 600-token text chunks, with 100-token overlaps between chunks, approximately 1 million tokens<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/node> DATASET<\/data> Benchmark dataset comprising news articles published from September 2013 to December 2023 in a range of categories, including entertainment, business, sports, technology, health, and science. Size: 3197 × 600-token text chunks, with 100-token overlaps between chunks, approximately 1.7 million tokens<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/node> DATASET<\/data> A benchmark dataset for open-domain question answering, targeting explicit fact retrieval<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/node> DATASET<\/data> A benchmark dataset comprising news articles published from September 2013 to December 2023 in a range of categories, including entertainment, business, sports, technology, health, and scienceA benchmark dataset for open-domain question answering, targeting explicit fact retrieval<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> DATASET<\/data> <\/node> DATASET<\/data> A benchmark dataset for open-domain question answering, targeting explicit fact retrieval<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> DATASET<\/data> <\/node> PROCESS<\/data> The process through which people inspect, engage with, and contextualize data within the broader scope of real-world activities<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> PROCESS<\/data> <\/node> TECHNOLOGY<\/data> Retrieval-Augmented Generation systems used for global sensemaking tasks<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> TECHNOLOGY<\/data> <\/node> TECHNOLOGY<\/data> Large Language Model used to automate the generation of questions for dataset evaluation<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> TECHNOLOGY<\/data> <\/node> PERSON<\/data> Microsoft CTO who participates in podcast conversations compiled in the dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/node> AUTHORS<\/data> Authors of a paper on data sensemaking behaviors<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> AUTHORS<\/data> <\/node> AUTHORS<\/data> Authors of a paper on methods for extracting latent summarization queries from source texts<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> AUTHORS<\/data> <\/node> AUTHORS<\/data> Authors associated with the MultiHop-RAG dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/node> AUTHORS<\/data> Authors associated with the HotPotQA dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> AUTHORS<\/data> <\/node> AUTHORS<\/data> Authors associated with the MT-Bench dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> AUTHORS<\/data> <\/node> TECHNOLOGY<\/data> A specific implementation of RAG using four levels of graph communities<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> TECHNOLOGY<\/data> <\/node> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/node> PODCAST<\/data> A podcast series featuring conversations between Kevin Scott and other technology leaders<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/node> PERSON<\/data> Kevin Scott, Microsoft CTO, who participates in the podcast conversations compiled in the dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/node> PERSON<\/data> An author associated with the MultiHop-RAG dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/node> PERSON<\/data> An author associated with the MultiHop-RAG dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/node> DATASET<\/data> A benchmark dataset for open-domain question answering, targeting explicit fact retrieval<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/node> 2.0<\/data> Kevin Scott is a participant in the podcast conversations compiled in the Podcast Transcripts dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/edge> 2.0<\/data> RAG systems are used to evaluate the Podcast Transcripts dataset for global sensemaking tasks<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/edge> 2.0<\/data> LLM is used to generate questions for evaluating the Podcast Transcripts dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/edge> 2.0<\/data> RAG systems are used to evaluate the News Articles dataset for global sensemaking tasks<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/edge> 2.0<\/data> LLM is used to generate questions for evaluating the News Articles dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/edge> 2.0<\/data> Yang et al. are the authors associated with the HotPotQA dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/edge> 2.0<\/data> Tang and Yang are the authors associated with the MultiHop-RAG dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/edge> 2.0<\/data> Zheng et al. are the authors associated with the MT-Bench dataset<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/edge> 2.0<\/data> Koesten et al. authored a paper on data sensemaking behaviors<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/edge> 2.0<\/data> Graph RAG is a specific implementation of RAG systems<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/edge> 2.0<\/data> Xu and Lapata authored a paper on methods for extracting latent summarization queries from source texts<\/data> 922778ce1cb2fdd6dbab1746c8795620<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"973164fa90bf2b4ee267f4fd795916bf","chunk":". For our evaluation, a value of N= 5 resulted in 125 test questions\nper dataset. Table 1 shows example questions for each of the two evaluation datasets.\n63.3 Conditions\nWe compare six different conditions in our analysis, including Graph RAG using four levels of graph\ncommunities ( C0,C1,C2,C3), a text summarization method applying our map-reduce approach\ndirectly to source texts ( TS), and a na \u00a8\u0131ve \u201csemantic search\u201d RAG approach ( SS):\n\u2022CO. Uses root-level community summaries (fewest in number) to answer user queries.\n\u2022C1. Uses high-level community summaries to answer queries. These are sub-communities\nof C0, if present, otherwise C0 communities projected down.\n\u2022C2. Uses intermediate-level community summaries to answer queries. These are sub-\ncommunities of C1, if present, otherwise C1 communities projected down.\n\u2022C3. Uses low-level community summaries (greatest in number) to answer queries. These\nare sub-communities of C2, if present, otherwise C2 communities projected down.\n\u2022TS. The same method as in subsection 2.6, except source texts (rather than community\nsummaries) are shuffled and chunked for the map-reduce summarization stages.\n\u2022SS. An implementation of na \u00a8\u0131ve RAG in which text chunks are retrieved and added to the\navailable context window until the specified token limit is reached.\nThe size of the context window and the prompts used for answer generation are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts","chunk_id":"973164fa90bf2b4ee267f4fd795916bf","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"N","type":"METRIC","description":"N represents the number of test questions per dataset used in the evaluation","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"DATASET","type":"DATASET","description":"A collection of data used for evaluation, including the Podcast and News datasets","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"GRAPH RAG","type":"METHOD","description":"Graph RAG is a method using graph communities at different levels to answer user queries","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"TEXT SUMMARIZATION","type":"METHOD","description":"A method applying a map-reduce approach directly to source texts for summarization","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"SEMANTIC SEARCH RAG","type":"METHOD","description":"A na\u00a8\u0131ve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"C0","type":"CATEGORY","description":"Root-level community summaries used to answer user queries, representing the fewest number of summaries","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"C1","type":"CATEGORY","description":"High-level community summaries used to answer user queries, representing sub-communities of C0","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"C2","type":"CATEGORY","description":"Intermediate-level community summaries used to answer user queries, representing sub-communities of C1","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"C3","type":"CATEGORY","description":"Low-level community summaries used to answer user queries, representing sub-communities of C2","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"TS","type":"METHOD","description":"A text summarization method applying a map-reduce approach directly to source texts","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"SS","type":"METHOD","description":"A na\u00a8\u0131ve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"CONTEXT WINDOW","type":"CONCEPT","description":"The size of the context window used for answer generation, which is the same across all conditions","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"PROMPTS","type":"CONCEPT","description":"The prompts used for answer generation, which are the same across all conditions with minor modifications","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"GRAPH INDEX","type":"TECHNOLOGY","description":"The graph index supporting conditions C0-C3, created using generic prompts for entity and relationship extraction","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"PODCAST DATASET","type":"DATASET","description":"A dataset consisting of podcast transcripts used in the evaluation","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"NEWS DATASET","type":"DATASET","description":"A dataset consisting of news articles used in the evaluation","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"METRICS","type":"CONCEPT","description":"Metrics used to evaluate natural language generation, including reference-based metrics and qualities of generated texts","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"LLMS","type":"TECHNOLOGY","description":"Large Language Models used as evaluators of natural language generation","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"WANG ET AL., 2023A","type":"REFERENCE","description":"A reference to a study by Wang et al. in 2023, indicating the effectiveness of LLMs in evaluation","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"ZHENG ET AL., 2024","type":"REFERENCE","description":"A reference to a study by Zheng et al. in 2024, indicating the effectiveness of LLMs in evaluation","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"TABLE 1","type":"REFERENCE","description":"Table 1 shows example questions for each of the two evaluation datasets","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"CONDITIONS","type":"CONCEPT","description":"Different conditions compared in the analysis, including Graph RAG, text summarization, and semantic search RAG","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"MAP-REDUCE","type":"METHOD","description":"A method used for text summarization by applying a map-reduce approach directly to source texts","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"USER QUERIES","type":"CONCEPT","description":"Queries made by users that are answered using different methods and conditions","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"ENTITY TYPES","type":"CONCEPT","description":"Types of entities extracted during the graph indexing process","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"FEW-SHOT EXAMPLES","type":"CONCEPT","description":"Examples tailored to the domain of the data used in the graph indexing process","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"CONTEXT WINDOW SIZE","type":"METRIC","description":"The size of the context window used in the graph indexing process, set to 600 tokens","source_id":"973164fa90bf2b4ee267f4fd795916bf"},{"name":"GLEANING","type":"CONCEPT","description":"The process of extracting information, with 1 gleaning for the Podcast dataset and 0 gleanings for the News dataset","source_id":"973164fa90bf2b4ee267f4fd795916bf"}],"entity_graph":" METRIC<\/data> N represents the number of test questions per dataset used in the evaluation<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> DATASET<\/data> A collection of data used for evaluation, including the Podcast and News datasets<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> METHOD<\/data> Graph RAG is a method using graph communities at different levels to answer user queries<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> METHOD<\/data> A method applying a map-reduce approach directly to source texts for summarization<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> METHOD<\/data> A na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CATEGORY<\/data> Root-level community summaries used to answer user queries, representing the fewest number of summaries<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CATEGORY<\/data> High-level community summaries used to answer user queries, representing sub-communities of C0<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CATEGORY<\/data> Intermediate-level community summaries used to answer user queries, representing sub-communities of C1<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CATEGORY<\/data> Low-level community summaries used to answer user queries, representing sub-communities of C2<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> METHOD<\/data> A text summarization method applying a map-reduce approach directly to source texts<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> METHOD<\/data> A na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CONCEPT<\/data> The size of the context window used for answer generation, which is the same across all conditions<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CONCEPT<\/data> The prompts used for answer generation, which are the same across all conditions with minor modifications<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> TECHNOLOGY<\/data> The graph index supporting conditions C0-C3, created using generic prompts for entity and relationship extraction<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> DATASET<\/data> A dataset consisting of podcast transcripts used in the evaluation<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> DATASET<\/data> A dataset consisting of news articles used in the evaluation<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CONCEPT<\/data> Metrics used to evaluate natural language generation, including reference-based metrics and qualities of generated texts<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> TECHNOLOGY<\/data> Large Language Models used as evaluators of natural language generation<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> REFERENCE<\/data> A reference to a study by Wang et al. in 2023, indicating the effectiveness of LLMs in evaluation<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> REFERENCE<\/data> A reference to a study by Zheng et al. in 2024, indicating the effectiveness of LLMs in evaluation<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> REFERENCE<\/data> Table 1 shows example questions for each of the two evaluation datasets<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CONCEPT<\/data> Different conditions compared in the analysis, including Graph RAG, text summarization, and semantic search RAG<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> METHOD<\/data> A method used for text summarization by applying a map-reduce approach directly to source texts<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CONCEPT<\/data> Queries made by users that are answered using different methods and conditions<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CONCEPT<\/data> Types of entities extracted during the graph indexing process<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CONCEPT<\/data> Examples tailored to the domain of the data used in the graph indexing process<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> METRIC<\/data> The size of the context window used in the graph indexing process, set to 600 tokens<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> CONCEPT<\/data> The process of extracting information, with 1 gleaning for the Podcast dataset and 0 gleanings for the News dataset<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/node> 1.0<\/data> N represents the number of test questions per dataset<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Table 1 shows example questions for each of the two evaluation datasets<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Graph RAG uses root-level community summaries (C0) to answer user queries<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Graph RAG uses high-level community summaries (C1) to answer user queries<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Graph RAG uses intermediate-level community summaries (C2) to answer user queries<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Graph RAG uses low-level community summaries (C3) to answer user queries<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Graph RAG is one of the conditions compared in the analysis<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Graph RAG uses different levels of graph communities to answer user queries<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Text summarization method applies a map-reduce approach directly to source texts (TS)<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Text summarization is one of the conditions compared in the analysis<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Map-reduce is the method used in the text summarization condition<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Semantic search RAG is a na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached (SS)<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Semantic search RAG is one of the conditions compared in the analysis<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> The graph index supports condition C0<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> C0 uses root-level community summaries to answer user queries<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> The graph index supports condition C1<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> C1 uses high-level community summaries to answer user queries<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> The graph index supports condition C2<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> C2 uses intermediate-level community summaries to answer user queries<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> The graph index supports condition C3<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> C3 uses low-level community summaries to answer user queries<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> The size of the context window and the prompts used for answer generation are the same across all conditions<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> The graph indexing process used a context window size of 600 tokens with 1 gleaning for the Podcast dataset<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> The graph indexing process used a context window size of 600 tokens with 0 gleanings for the News dataset<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> The graph index was created using generic prompts for entity and relationship extraction<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Few-shot examples tailored to the domain of the data were used in the graph indexing process<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> The graph indexing process used a context window size of 600 tokens<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> The graph indexing process used 1 gleaning for the Podcast dataset<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> The graph indexing process used 0 gleanings for the News dataset<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> LLMs are used to generate metrics for evaluating natural language generation<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Wang et al. (2023) indicated the effectiveness of LLMs in evaluation<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> 1.0<\/data> Zheng et al. (2024) indicated the effectiveness of LLMs in evaluation<\/data> 973164fa90bf2b4ee267f4fd795916bf<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"322e02986c8724eedbcf3ebfa20b989c","chunk":" of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in a reference-\nfree style (Wang et al., 2023a) as well as in head-to-head comparison of competing outputs (LLM-\nas-a-judge, Zheng et al., 2024). LLMs have also shown promise at evaluating the performance of\nconventional RAG systems, automatically evaluating qualities like context relevance, faithfulness,\nand answer relevance (RAGAS, Es et al., 2023).\nGiven the multi-stage nature of our Graph RAG mechanism, the multiple conditions we wanted to\ncompare, and the lack of gold standard answers to our activity-based sensemaking questions, we\ndecided to adopt a head-to-head comparison approach using an LLM evaluator. We selected three\ntarget metrics capturing qualities that are desirable for sensemaking activities, as well as a control\nmetric (directness) used as a indicator of validity. Since directness is effectively in opposition to\ncomprehensiveness and diversity, we would not expect any method to win across all four metrics.\nOur head-to-head measures computed using an LLM evaluator are as follows:\n\u2022Comprehensiveness . How much detail does the answer provide to cover all aspects and\ndetails of the question?\n\u2022Diversity . How varied and rich is the answer in providing different perspectives and insights\non the question?\n\u2022Empowerment . How well does the answer help the reader understand and make informed\njudgements about the topic?\n\u2022Directness . How specifically and clearly does the answer address the question?\nFor our evaluation, the LLM is provided with the question, target metric, and a pair of answers, and\nasked to assess which answer is better according to the metric, as well as why. It returns the winner\nif one exists, otherwise a tie if they are fundamentally similar and the differences are negligible.\nTo account for the stochasticity of LLMs, we run each comparison five times and use mean scores.\nTable 2 shows an example of LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompass","chunk_id":"322e02986c8724eedbcf3ebfa20b989c","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"NATURAL LANGUAGE GENERATION","type":"TECHNOLOGY","description":"Natural Language Generation (NLG) is a subfield of artificial intelligence that focuses on generating human-like text from data","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"WANG ET AL., 2023A","type":"PUBLICATION","description":"A reference to a study or paper authored by Wang and others in 2023","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"ZHENG ET AL., 2024","type":"PUBLICATION","description":"A reference to a study or paper authored by Zheng and others in 2024","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"LLM-AS-A-JUDGE","type":"METHOD","description":"A method where a Large Language Model (LLM) is used to compare and evaluate competing outputs","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"RAGAS","type":"METHOD","description":"A method for evaluating the performance of Retrieval-Augmented Generation (RAG) systems, focusing on context relevance, faithfulness, and answer relevance","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"ES ET AL., 2023","type":"PUBLICATION","description":"A reference to a study or paper authored by Es and others in 2023","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"GRAPH RAG","type":"METHOD","description":"A multi-stage mechanism for Retrieval-Augmented Generation (RAG) that involves comparing multiple conditions","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"LLM EVALUATOR","type":"TOOL","description":"A Large Language Model (LLM) used to evaluate and compare generated texts based on specific metrics","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"COMPREHENSIVENESS","type":"METRIC","description":"A metric that measures how much detail an answer provides to cover all aspects and details of a question","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"DIVERSITY","type":"METRIC","description":"A metric that measures how varied and rich an answer is in providing different perspectives and insights on a question","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"EMPOWERMENT","type":"METRIC","description":"A metric that measures how well an answer helps the reader understand and make informed judgements about a topic","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"DIRECTNESS","type":"METRIC","description":"A metric that measures how specifically and clearly an answer addresses a question","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"TABLE 2","type":"DATA","description":"An example of LLM-generated assessment shown in a table format","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"QUESTION","type":"DATA","description":"A specific query used in the evaluation process","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"PUBLIC FIGURES","type":"ENTITY","description":"Individuals who are well-known in the entertainment industry and are mentioned across various articles","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"ENTERTAINMENT ARTICLES","type":"DATASET","description":"A collection of articles focused on the entertainment industry","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"ENTERTAINMENT INDUSTRY","type":"DOMAIN","description":"A sector that encompasses various forms of entertainment, including movies, music, and television","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"STATE-OF-THE-ART","type":"METRIC","description":"A metric indicating the highest level of development or achievement in a particular field","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"COMPETITIVE RESULTS","type":"METRIC","description":"A metric indicating results that are comparable to or better than those of others in the same field","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"HUMAN JUDGEMENTS","type":"METRIC","description":"A metric based on evaluations made by humans","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"REFERENCE-BASED METRICS","type":"METRIC","description":"Metrics that require a gold standard or reference answers for evaluation","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"REFERENCE-FREE STYLE","type":"METHOD","description":"An evaluation method that does not require reference answers","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"CONTEXT RELEVANCE","type":"METRIC","description":"A metric that measures how relevant the generated text is to the given context","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"FAITHFULNESS","type":"METRIC","description":"A metric that measures how accurately the generated text reflects the source information","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"ANSWER RELEVANCE","type":"METRIC","description":"A metric that measures how relevant the generated answer is to the question","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"MULTI-STAGE","type":"METHOD","description":"A method involving multiple stages or steps","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"CONDITIONS","type":"DATA","description":"Different scenarios or variables that are compared in an experiment","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"GOLD STANDARD ANSWERS","type":"DATA","description":"The correct or ideal answers used as a benchmark in evaluations","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"SENSEMAKING QUESTIONS","type":"DATA","description":"Questions designed to help understand and make sense of complex information","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"HEAD-TO-HEAD COMPARISON","type":"METHOD","description":"A method where two items are directly compared against each other","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"TARGET METRICS","type":"DATA","description":"Specific metrics that are the focus of an evaluation","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"CONTROL METRIC","type":"DATA","description":"A metric used as a baseline or standard for comparison","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"VALIDITY","type":"METRIC","description":"A metric that measures the accuracy and reliability of a method or result","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"STOCHASTICITY","type":"METRIC","description":"A metric that measures the randomness or variability in a process","source_id":"322e02986c8724eedbcf3ebfa20b989c"},{"name":"MEAN SCORES","type":"DATA","description":"The average scores obtained from multiple evaluations","source_id":"322e02986c8724eedbcf3ebfa20b989c"}],"entity_graph":" TECHNOLOGY<\/data> Natural Language Generation (NLG) is a subfield of artificial intelligence that focuses on generating human-like text from data<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> PUBLICATION<\/data> A reference to a study or paper authored by Wang and others in 2023<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> PUBLICATION<\/data> A reference to a study or paper authored by Zheng and others in 2024<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METHOD<\/data> A method where a Large Language Model (LLM) is used to compare and evaluate competing outputs<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METHOD<\/data> A method for evaluating the performance of Retrieval-Augmented Generation (RAG) systems, focusing on context relevance, faithfulness, and answer relevance<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> PUBLICATION<\/data> A reference to a study or paper authored by Es and others in 2023<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METHOD<\/data> A multi-stage mechanism for Retrieval-Augmented Generation (RAG) that involves comparing multiple conditions<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> TOOL<\/data> A Large Language Model (LLM) used to evaluate and compare generated texts based on specific metrics<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric that measures how much detail an answer provides to cover all aspects and details of a question<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric that measures how varied and rich an answer is in providing different perspectives and insights on a question<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric that measures how well an answer helps the reader understand and make informed judgements about a topic<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric that measures how specifically and clearly an answer addresses a question<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> DATA<\/data> An example of LLM-generated assessment shown in a table format<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> DATA<\/data> A specific query used in the evaluation process<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> ENTITY<\/data> Individuals who are well-known in the entertainment industry and are mentioned across various articles<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> DATASET<\/data> A collection of articles focused on the entertainment industry<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> DOMAIN<\/data> A sector that encompasses various forms of entertainment, including movies, music, and television<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric indicating the highest level of development or achievement in a particular field<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric indicating results that are comparable to or better than those of others in the same field<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric based on evaluations made by humans<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> Metrics that require a gold standard or reference answers for evaluation<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METHOD<\/data> An evaluation method that does not require reference answers<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric that measures how relevant the generated text is to the given context<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric that measures how accurately the generated text reflects the source information<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric that measures how relevant the generated answer is to the question<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METHOD<\/data> A method involving multiple stages or steps<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> DATA<\/data> Different scenarios or variables that are compared in an experiment<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> DATA<\/data> The correct or ideal answers used as a benchmark in evaluations<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> DATA<\/data> Questions designed to help understand and make sense of complex information<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METHOD<\/data> A method where two items are directly compared against each other<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> DATA<\/data> Specific metrics that are the focus of an evaluation<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> DATA<\/data> A metric used as a baseline or standard for comparison<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric that measures the accuracy and reliability of a method or result<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> METRIC<\/data> A metric that measures the randomness or variability in a process<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> DATA<\/data> The average scores obtained from multiple evaluations<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/node> 1.0<\/data> Wang et al., 2023a discusses the state-of-the-art results achieved by Natural Language Generation<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Zheng et al., 2024 discusses the competitive results achieved by Natural Language Generation<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Natural Language Generation achieves state-of-the-art results<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Natural Language Generation achieves competitive results<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Natural Language Generation is compared against human judgements<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Natural Language Generation can generate reference-based metrics<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Natural Language Generation can measure qualities in a reference-free style<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Zheng et al., 2024 discusses the LLM-as-a-judge method<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Es et al., 2023 discusses the RAGAS method<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> RAGAS evaluates context relevance<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> RAGAS evaluates faithfulness<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> RAGAS evaluates answer relevance<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The Graph RAG mechanism uses an LLM evaluator for head-to-head comparison<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Graph RAG is a multi-stage mechanism<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Graph RAG compares multiple conditions<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The LLM evaluator assesses answers based on the comprehensiveness metric<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The LLM evaluator assesses answers based on the diversity metric<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The LLM evaluator assesses answers based on the empowerment metric<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The LLM evaluator assesses answers based on the directness metric<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Table 2 shows an example of LLM-generated assessment<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The LLM evaluator uses a head-to-head comparison approach<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The LLM evaluator assesses answers based on target metrics<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The LLM evaluator uses a control metric for validity<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The LLM evaluator accounts for stochasticity<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The LLM evaluator uses mean scores from multiple comparisons<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The question asks about public figures mentioned in entertainment articles<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Public figures are repeatedly mentioned across various entertainment articles<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Entertainment articles cover topics related to the entertainment industry<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Reference-based metrics require gold standard answers<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> Gold standard answers are lacking for sensemaking questions<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> 1.0<\/data> The control metric is used as an indicator of validity<\/data> 322e02986c8724eedbcf3ebfa20b989c<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"e8c8f911135faf3ff35f24107eb3f99c","chunk":"ity of LLMs, we run each comparison five times and use mean scores.\nTable 2 shows an example of LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe public\u2019s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural\nlandscape, often becoming central figures in social discussions and public discourse.\nNa\u00a8\u0131ve\nRAGPublic figures who are repeatedly mentioned across various entertainment articles include Taylor\nSwift, Travis Kelce, Britney Spears, and Justin Timberlake. These individuals have been highlighted\nfor various reasons, ranging from their professional achievements to their personal lives.\nTaylor Swift [...]Travis Kelce [...]Britney Spears [...]Justin Timberlake [...]\nThese figures are frequently covered due to their high-profile status and the public\u2019s interest in their\ncareers and personal lives. Their activities, whether in music, sports, or personal relationships, have\nsignificant cultural and economic impacts, as evidenced by the media coverage and public reactions.\nLLM\nDecisionComprehensiveness: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a more comprehensive and detailed list of public figures from a\nwider range of entertainment sectors, including film, television, music, sports, gaming, and digital\nmedia. It also includes specific examples of their contributions and the impact they have on their\nrespective fields, as well as mentions of controversies and their implications. Answer 2, while\ndetailed in its coverage of a few individuals, is limited to a smaller number of public figures and\nfocuses primarily on their personal lives and relationships rather than a broad spectrum of their\nprofessional influence across the entertainment industry.\nDiversity: Winner=","chunk_id":"e8c8f911135faf3ff35f24107eb3f99c","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"LLM","type":"TECHNOLOGY","description":"LLM (Large Language Model) is a type of artificial intelligence used for generating and assessing text","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"RAG","type":"METHOD","description":"RAG (Retrieval-Augmented Generation) is a method that produces direct responses in text generation tasks","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"GRAPH RAG","type":"METHOD","description":"Graph RAG is a method that provides a comprehensive overview of public figures in the entertainment industry","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"NAIVE RAG","type":"METHOD","description":"Naive RAG is a method that lists public figures mentioned in entertainment articles, focusing on their professional achievements and personal lives","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"TAYLOR SWIFT","type":"PERSON","description":"Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to music and her high-profile personal life","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"TRAVIS KELCE","type":"PERSON","description":"Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to sports and his high-profile personal life","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"BRITNEY SPEARS","type":"PERSON","description":"Britney Spears is a public figure frequently mentioned in entertainment articles, known for her contributions to music and her high-profile personal life","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"JUSTIN TIMBERLAKE","type":"PERSON","description":"Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his contributions to music and his high-profile personal life","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"ENTERTAINMENT INDUSTRY","type":"SECTOR","description":"The entertainment industry encompasses film, television, music, sports, and digital media","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"ACTORS AND DIRECTORS","type":"CATEGORY","description":"A category of public figures in the entertainment industry, including those involved in film and television","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"MUSICIANS AND EXECUTIVES","type":"CATEGORY","description":"A category of public figures in the entertainment industry, including those involved in music","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"ATHLETES AND COACHES","type":"CATEGORY","description":"A category of public figures in the entertainment industry, including those involved in sports","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"INFLUENCERS AND ENTREPRENEURS","type":"CATEGORY","description":"A category of public figures in the entertainment industry, including those involved in digital media and business","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"PUBLIC FIGURES IN CONTROVERSY","type":"CATEGORY","description":"A category of public figures in the entertainment industry who are involved in controversies","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"COMPREHENSIVENESS","type":"METRIC","description":"A metric used to evaluate the comprehensiveness of the generated responses","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"DECISION","type":"METRIC","description":"A metric used to determine the winner in the comparison of generated responses","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"QUESTION","type":"METRIC","description":"A metric used to evaluate the generated responses by asking specific questions","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"ASSESSMENT","type":"METRIC","description":"A metric used to evaluate the quality of LLM-generated responses","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"ENTERTAINMENT ARTICLES","type":"DATASET","description":"A dataset consisting of articles related to the entertainment industry","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"FILM","type":"SECTOR","description":"A sector within the entertainment industry that includes movies and cinema","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"TELEVISION","type":"SECTOR","description":"A sector within the entertainment industry that includes TV shows and series","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"MUSIC","type":"SECTOR","description":"A sector within the entertainment industry that includes musical performances and recordings","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"SPORTS","type":"SECTOR","description":"A sector within the entertainment industry that includes athletic events and competitions","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"DIGITAL MEDIA","type":"SECTOR","description":"A sector within the entertainment industry that includes online content and social media","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"CULTURAL NARRATIVES","type":"CATEGORY","description":"A category within the entertainment industry that includes stories and themes that shape culture","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"TRENDS","type":"CATEGORY","description":"A category within the entertainment industry that includes popular movements and styles","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"SOCIAL DISCUSSIONS","type":"CATEGORY","description":"A category within the entertainment industry that includes public conversations and debates","source_id":"e8c8f911135faf3ff35f24107eb3f99c"},{"name":"PUBLIC DISCOURSE","type":"CATEGORY","description":"A category within the entertainment industry that includes formal discussions and communications","source_id":"e8c8f911135faf3ff35f24107eb3f99c"}],"entity_graph":" TECHNOLOGY<\/data> LLM (Large Language Model) is a type of artificial intelligence used for generating and assessing text<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> METHOD<\/data> RAG (Retrieval-Augmented Generation) is a method that produces direct responses in text generation tasks<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> METHOD<\/data> Graph RAG is a method that provides a comprehensive overview of public figures in the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> METHOD<\/data> Naive RAG is a method that lists public figures mentioned in entertainment articles, focusing on their professional achievements and personal lives<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> PERSON<\/data> Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to music and her high-profile personal life<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> PERSON<\/data> Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to sports and his high-profile personal life<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> PERSON<\/data> Britney Spears is a public figure frequently mentioned in entertainment articles, known for her contributions to music and her high-profile personal life<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> PERSON<\/data> Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his contributions to music and his high-profile personal life<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> SECTOR<\/data> The entertainment industry encompasses film, television, music, sports, and digital media<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> CATEGORY<\/data> A category of public figures in the entertainment industry, including those involved in film and television<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> CATEGORY<\/data> A category of public figures in the entertainment industry, including those involved in music<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> CATEGORY<\/data> A category of public figures in the entertainment industry, including those involved in sports<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> CATEGORY<\/data> A category of public figures in the entertainment industry, including those involved in digital media and business<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> CATEGORY<\/data> A category of public figures in the entertainment industry who are involved in controversies<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> METRIC<\/data> A metric used to evaluate the comprehensiveness of the generated responses<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> METRIC<\/data> A metric used to determine the winner in the comparison of generated responses<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> METRIC<\/data> A metric used to evaluate the generated responses by asking specific questions<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> METRIC<\/data> A metric used to evaluate the quality of LLM-generated responses<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> DATASET<\/data> A dataset consisting of articles related to the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> SECTOR<\/data> A sector within the entertainment industry that includes movies and cinema<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> SECTOR<\/data> A sector within the entertainment industry that includes TV shows and series<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> SECTOR<\/data> A sector within the entertainment industry that includes musical performances and recordings<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> SECTOR<\/data> A sector within the entertainment industry that includes athletic events and competitions<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> SECTOR<\/data> A sector within the entertainment industry that includes online content and social media<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> CATEGORY<\/data> A category within the entertainment industry that includes stories and themes that shape culture<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> CATEGORY<\/data> A category within the entertainment industry that includes popular movements and styles<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> CATEGORY<\/data> A category within the entertainment industry that includes public conversations and debates<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> CATEGORY<\/data> A category within the entertainment industry that includes formal discussions and communications<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/node> 1.0<\/data> LLM uses RAG to generate and assess text<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> LLM uses Graph RAG to provide a comprehensive overview of public figures in the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> LLM uses Naive RAG to list public figures mentioned in entertainment articles<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> LLM-generated responses are evaluated using assessment metrics<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> LLM-generated responses are evaluated using specific questions<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Graph RAG is a specific implementation of RAG<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Naive RAG is a specific implementation of RAG<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Graph RAG mentions Taylor Swift as a prominent public figure<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Graph RAG mentions Travis Kelce as a prominent public figure<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Graph RAG mentions Britney Spears as a prominent public figure<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Graph RAG mentions Justin Timberlake as a prominent public figure<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Graph RAG is evaluated for comprehensiveness<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Graph RAG is determined to be the winner based on the decision metric<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Naive RAG mentions Taylor Swift as a public figure<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Naive RAG mentions Travis Kelce as a public figure<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Naive RAG mentions Britney Spears as a public figure<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Naive RAG mentions Justin Timberlake as a public figure<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Naive RAG is evaluated for comprehensiveness<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Naive RAG is determined to be the loser based on the decision metric<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Taylor Swift is a significant figure in the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Taylor Swift is frequently mentioned in entertainment articles<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Travis Kelce is a significant figure in the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Travis Kelce is frequently mentioned in entertainment articles<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Britney Spears is a significant figure in the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Britney Spears is frequently mentioned in entertainment articles<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Justin Timberlake is a significant figure in the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Justin Timberlake is frequently mentioned in entertainment articles<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Actors and Directors are a category within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Musicians and Executives are a category within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Athletes and Coaches are a category within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Influencers and Entrepreneurs are a category within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Public Figures in Controversy are a category within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Film is a sector within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Television is a sector within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Music is a sector within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Sports is a sector within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Digital Media is a sector within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Cultural Narratives are a category within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Trends are a category within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Social Discussions are a category within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Public Discourse is a category within the entertainment industry<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> 1.0<\/data> Comprehensiveness is a metric used to determine the decision<\/data> e8c8f911135faf3ff35f24107eb3f99c<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"718017a4871c909420f84b85b8ba969d","chunk":" as mentions of controversies and their implications. Answer 2, while\ndetailed in its coverage of a few individuals, is limited to a smaller number of public figures and\nfocuses primarily on their personal lives and relationships rather than a broad spectrum of their\nprofessional influence across the entertainment industry.\nDiversity: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a more varied and rich response by covering a wide range of\npublic figures from different sectors of the entertainment industry, including film, television, music,\nsports, gaming, and digital media. It offers insights into the contributions and influence of these\nfigures, as well as controversies and their impact on public discourse. The answer also cites specific\ndata sources for each mentioned figure, indicating a diverse range of evidence to support the claims.\nIn contrast, Answer 2 focuses on a smaller group of public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1.\nDirectness: Winner=2 (Na \u00a8\u0131ve RAG)\nAnswer 2 is better because it directly lists specific public figures who are repeatedly mentioned\nacross various entertainment articles, such as Taylor Swift, Travis Kelce, Britney Spears, and Justin\nTimberlake, and provides concise explanations for their frequent mentions. Answer 1, while\ncomprehensive, includes a lot of detailed information about various figures in different sectors of\nentertainment, which, while informative, does not directly answer the question with the same level of\nconciseness and specificity as Answer 2.\nTable 2: Example question for the News article dataset, with generated answers","chunk_id":"718017a4871c909420f84b85b8ba969d","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"ANSWER 1","type":"RESPONSE","description":"Answer 1 provides a varied and rich response by covering a wide range of public figures from different sectors of the entertainment industry, including film, television, music, sports, gaming, and digital media. It offers insights into the contributions and influence of these figures, as well as controversies and their impact on public discourse. The answer also cites specific data sources for each mentioned figure, indicating a diverse range of evidence to support the claims.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"ANSWER 2","type":"RESPONSE","description":"Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports, and relies heavily on a single source for data. It provides concise explanations for the frequent mentions of specific public figures such as Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"GRAPH RAG","type":"METHOD","description":"Graph RAG is a method used to generate responses that provide a comprehensive and structured overview of public figures across various sectors of the entertainment industry.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"NA\u00cfVE RAG","type":"METHOD","description":"Na\u00efve RAG is a method used to generate responses that directly list specific public figures who are repeatedly mentioned across various entertainment articles.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"TAYLOR SWIFT","type":"PERSON","description":"Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"TRAVIS KELCE","type":"PERSON","description":"Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"BRITNEY SPEARS","type":"PERSON","description":"Britney Spears is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"JUSTIN TIMBERLAKE","type":"PERSON","description":"Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his contributions to the music industry.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"NEWS ARTICLE DATASET","type":"DATASET","description":"A dataset consisting of news articles used for generating responses to questions about public figures in the entertainment industry.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"CONTROVERSIES","type":"TOPIC","description":"Controversies are events or issues involving public figures that generate public debate and impact public discourse.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"PUBLIC FIGURES","type":"CATEGORY","description":"Public figures are individuals who have gained fame or notoriety in various sectors such as entertainment, sports, and digital media.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"FILM","type":"SECTOR","description":"The film sector includes public figures involved in the movie industry, including actors, directors, and producers.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"TELEVISION","type":"SECTOR","description":"The television sector includes public figures involved in TV shows, including actors, hosts, and producers.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"MUSIC","type":"SECTOR","description":"The music sector includes public figures involved in the music industry, including singers, musicians, and producers.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"SPORTS","type":"SECTOR","description":"The sports sector includes public figures involved in sports, including athletes, coaches, and sports commentators.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"GAMING","type":"SECTOR","description":"The gaming sector includes public figures involved in the gaming industry, including gamers, developers, and streamers.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"DIGITAL MEDIA","type":"SECTOR","description":"The digital media sector includes public figures involved in online platforms, including influencers, content creators, and digital marketers.","source_id":"718017a4871c909420f84b85b8ba969d"},{"name":"DATA SOURCES","type":"RESOURCE","description":"Data sources are references or reports used to support claims about public figures and their influence.","source_id":"718017a4871c909420f84b85b8ba969d"}],"entity_graph":" RESPONSE<\/data> Answer 1 provides a varied and rich response by covering a wide range of public figures from different sectors of the entertainment industry, including film, television, music, sports, gaming, and digital media. It offers insights into the contributions and influence of these figures, as well as controversies and their impact on public discourse. The answer also cites specific data sources for each mentioned figure, indicating a diverse range of evidence to support the claims.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> RESPONSE<\/data> Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports, and relies heavily on a single source for data. It provides concise explanations for the frequent mentions of specific public figures such as Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> METHOD<\/data> Graph RAG is a method used to generate responses that provide a comprehensive and structured overview of public figures across various sectors of the entertainment industry.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> METHOD<\/data> Naïve RAG is a method used to generate responses that directly list specific public figures who are repeatedly mentioned across various entertainment articles.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> PERSON<\/data> Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> PERSON<\/data> Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> PERSON<\/data> Britney Spears is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> PERSON<\/data> Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his contributions to the music industry.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> DATASET<\/data> A dataset consisting of news articles used for generating responses to questions about public figures in the entertainment industry.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> TOPIC<\/data> Controversies are events or issues involving public figures that generate public debate and impact public discourse.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> CATEGORY<\/data> Public figures are individuals who have gained fame or notoriety in various sectors such as entertainment, sports, and digital media.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> SECTOR<\/data> The film sector includes public figures involved in the movie industry, including actors, directors, and producers.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> SECTOR<\/data> The television sector includes public figures involved in TV shows, including actors, hosts, and producers.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> SECTOR<\/data> The music sector includes public figures involved in the music industry, including singers, musicians, and producers.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> SECTOR<\/data> The sports sector includes public figures involved in sports, including athletes, coaches, and sports commentators.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> SECTOR<\/data> The gaming sector includes public figures involved in the gaming industry, including gamers, developers, and streamers.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> SECTOR<\/data> The digital media sector includes public figures involved in online platforms, including influencers, content creators, and digital marketers.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> RESOURCE<\/data> Data sources are references or reports used to support claims about public figures and their influence.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/node> 1.0<\/data> Answer 1 was generated using the Graph RAG method, which provides a comprehensive and structured overview of public figures across various sectors of the entertainment industry.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 1 cites specific data sources from the News article dataset for each mentioned figure.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 1 provides insights into controversies involving public figures and their impact on public discourse.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 1 covers a wide range of public figures from different sectors of the entertainment industry.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 1 includes public figures from the film sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 1 includes public figures from the television sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 1 includes public figures from the music sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 1 includes public figures from the sports sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 1 includes public figures from the gaming sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 1 includes public figures from the digital media sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 1 cites specific data sources for each mentioned figure.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 2 was generated using the Naïve RAG method, which directly lists specific public figures who are repeatedly mentioned across various entertainment articles.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 2 relies heavily on a single source from the News article dataset for data.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Taylor Swift is one of the specific public figures mentioned in Answer 2.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Travis Kelce is one of the specific public figures mentioned in Answer 2.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Britney Spears is one of the specific public figures mentioned in Answer 2.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Justin Timberlake is one of the specific public figures mentioned in Answer 2.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 2 focuses on public figures primarily from the music sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 2 focuses on public figures primarily from the sports sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Answer 2 relies heavily on a single data source.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Taylor Swift is a public figure in the music sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Travis Kelce is a public figure in the sports sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Britney Spears is a public figure in the music sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Justin Timberlake is a public figure in the music sector.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> 1.0<\/data> Controversies involve public figures and impact public discourse.<\/data> 718017a4871c909420f84b85b8ba969d<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"ebf5249c888e07fedce6572a4c03f88c","chunk":" while\ncomprehensive, includes a lot of detailed information about various figures in different sectors of\nentertainment, which, while informative, does not directly answer the question with the same level of\nconciseness and specificity as Answer 2.\nTable 2: Example question for the News article dataset, with generated answers from Graph RAG\n(C2) and Na \u00a8\u0131ve RAG, as well as LLM-generated assessments.\n8Podcast transcripts\n501728252221\n835050484344\n725050535049\n755247505250\n785750485052\n795651504850SS\nTS\nC0\nC1\nC2\nC3SSTSC0C1C2C3\nComprehensiveness501823251919\n825050504346\n775050504644\n755050504445\n815754565048\n815456555250SS\nTS\nC0\nC1\nC2\nC3SSTSC0C1C2C3\nDiversity504257524951\n585059555251\n434150494748\n484551504950\n514853515051\n494952504950SS\nTS\nC0\nC1\nC2\nC3SSTSC0C1C2C3\nEmpowerment505665606060\n445055525152\n354550474848\n404853505050\n404952505050\n404852505050SS\nTS\nC0\nC1\nC2\nC3SSTSC0C1C2C3\nDirectness\nNews articles\n502028252121\n805044413836\n725650525452\n755948505855\n796246425059\n796448454150SS\nTS\nC0\nC1\nC2\nC3SSTSC0C1C2C3\nComprehensiveness503338352931\n675053454440\n624750404141\n655560505050\n715659505051\n696059504950SS\nTS\nC0\nC1\nC2\nC3SSTSC0C1C2C3\nDiversity504757495050\n535058505048\n434250424544\n515058505251\n505055485050\n505256495050SS\nTS\nC0\nC1\nC2\nC3SSTSC0C1C2C3\nEmpowerment505459555554\n465055535252","chunk_id":"ebf5249c888e07fedce6572a4c03f88c","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"NEWS ARTICLE DATASET","type":"DATASET","description":"A dataset consisting of news articles used for analysis","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"GRAPH RAG","type":"METHOD","description":"Graph RAG is a method used to generate answers for questions in the News article dataset","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"NA\u00cfVE RAG","type":"METHOD","description":"Na\u00efve RAG is a method used to generate answers for questions in the News article dataset","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"LLM-GENERATED ASSESSMENTS","type":"METHOD","description":"Assessments generated by large language models (LLMs) to evaluate the answers produced by different methods","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"PODCAST TRANSCRIPTS","type":"DATASET","description":"A dataset consisting of transcripts from podcasts used for analysis","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"C0","type":"CATEGORY","description":"A category or cluster used in the analysis, representing a specific subset of the data","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"C1","type":"CATEGORY","description":"A category or cluster used in the analysis, representing a specific subset of the data","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"C2","type":"CATEGORY","description":"A category or cluster used in the analysis, representing a specific subset of the data","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"C3","type":"CATEGORY","description":"A category or cluster used in the analysis, representing a specific subset of the data","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"TS","type":"CATEGORY","description":"A category or cluster used in the analysis, representing a specific subset of the data","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"COMPREHENSIVENESS","type":"METRIC","description":"A metric used to evaluate the thoroughness of the generated answers","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"DIVERSITY","type":"METRIC","description":"A metric used to evaluate the variety in the generated answers","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"EMPOWERMENT","type":"METRIC","description":"A metric used to evaluate how empowering the generated answers are","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"DIRECTNESS","type":"METRIC","description":"A metric used to evaluate the straightforwardness of the generated answers","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"ANSWER 2","type":"METHOD","description":"Answer 2 is a generated answer for the example question in the News article dataset","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"EXAMPLE QUESTION","type":"DATASET","description":"An example question used in the News article dataset for analysis","source_id":"ebf5249c888e07fedce6572a4c03f88c"},{"name":"SS","type":"CATEGORY","description":"A category or cluster used in the analysis, representing a specific subset of the data","source_id":"ebf5249c888e07fedce6572a4c03f88c"}],"entity_graph":" DATASET<\/data> A dataset consisting of news articles used for analysis<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> METHOD<\/data> Graph RAG is a method used to generate answers for questions in the News article dataset<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> METHOD<\/data> Naïve RAG is a method used to generate answers for questions in the News article dataset<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> METHOD<\/data> Assessments generated by large language models (LLMs) to evaluate the answers produced by different methods<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> DATASET<\/data> A dataset consisting of transcripts from podcasts used for analysis<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> CATEGORY<\/data> A category or cluster used in the analysis, representing a specific subset of the data<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> CATEGORY<\/data> A category or cluster used in the analysis, representing a specific subset of the data<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> CATEGORY<\/data> A category or cluster used in the analysis, representing a specific subset of the data<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> CATEGORY<\/data> A category or cluster used in the analysis, representing a specific subset of the data<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> CATEGORY<\/data> A category or cluster used in the analysis, representing a specific subset of the data<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> METRIC<\/data> A metric used to evaluate the thoroughness of the generated answers<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> METRIC<\/data> A metric used to evaluate the variety in the generated answers<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> METRIC<\/data> A metric used to evaluate how empowering the generated answers are<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> METRIC<\/data> A metric used to evaluate the straightforwardness of the generated answers<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> METHOD<\/data> Answer 2 is a generated answer for the example question in the News article dataset<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> DATASET<\/data> An example question used in the News article dataset for analysis<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> CATEGORY<\/data> A category or cluster used in the analysis, representing a specific subset of the data<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/node> 1.0<\/data> Graph RAG is used to generate answers for questions in the News article dataset<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> Naïve RAG is used to generate answers for questions in the News article dataset<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> LLM-generated assessments are used to evaluate the answers produced for questions in the News article dataset<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> C0 is a category used in the analysis of news articles<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> C1 is a category used in the analysis of news articles<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> C2 is a category used in the analysis of news articles<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> C3 is a category used in the analysis of news articles<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> TS is a category used in the analysis of news articles<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> Comprehensiveness is used to evaluate the thoroughness of the generated answers for news articles<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> Diversity is used to evaluate the variety in the generated answers for news articles<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> Empowerment is used to evaluate how empowering the generated answers are for news articles<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> Directness is used to evaluate the straightforwardness of the generated answers for news articles<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> Answer 2 is a generated answer for a question in the News article dataset<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> Example question is part of the News article dataset used for analysis<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> SS is a category used in the analysis of news articles<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> C0 is a category used in the analysis of podcast transcripts<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> C1 is a category used in the analysis of podcast transcripts<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> C2 is a category used in the analysis of podcast transcripts<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> C3 is a category used in the analysis of podcast transcripts<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> TS is a category used in the analysis of podcast transcripts<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> Comprehensiveness is used to evaluate the thoroughness of the generated answers for podcast transcripts<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> Diversity is used to evaluate the variety in the generated answers for podcast transcripts<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> Empowerment is used to evaluate how empowering the generated answers are for podcast transcripts<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> 1.0<\/data> SS is a category used in the analysis of podcast transcripts<\/data> ebf5249c888e07fedce6572a4c03f88c<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"4c855404ee3d3c94aa2136f1513c666f","chunk":"050\n535058505048\n434250424544\n515058505251\n505055485050\n505256495050SS\nTS\nC0\nC1\nC2\nC3SSTSC0C1C2C3\nEmpowerment505459555554\n465055535252\n414550484847\n454752504949\n454852515049\n464853515150SS\nTS\nC0\nC1\nC2\nC3SSTSC0C1C2C3\nDirectness\nFigure 4: Head-to-head win rate percentages of (row condition) over (column condition) across two\ndatasets, four metrics, and 125 questions per comparison (each repeated five times and averaged).\nThe overall winner per dataset and metric is shown in bold. Self-win rates were not computed but\nare shown as the expected 50% for reference. All Graph RAG conditions outperformed na \u00a8\u0131ve RAG\non comprehensiveness and diversity. Conditions C1-C3 also showed slight improvements in answer\ncomprehensiveness and diversity over TS (global text summarization without a graph index).\n3.5 Configuration\nThe effect of context window size on any particular task is unclear, especially for models like\ngpt-4-turbo with a large context size of 128k tokens. Given the potential for information to\nbe \u201clost in the middle\u201d of longer contexts (Kuratov et al., 2024; Liu et al., 2023), we wanted to ex-\nplore the effects of varying the context window size for our combinations of datasets, questions, and\nmetrics. In particular, our goal was to determine the optimum context size for our baseline condition\n(SS) and then use this uniformly for all query-time LLM use. To that end, we tested four context\nwindow sizes: 8k, 16k, 32k and 64k. Surprisingly, the smallest context window size tested (8k)\nwas universally better for all comparisons on comprehensiveness (average win rate of 58.1%), while\nperforming comparably with larger context sizes on diversity (average win rate = 52.4%), and em-\npowerment (average win rate = 51.3%). Given our preference for more comprehensive and diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in","chunk_id":"4c855404ee3d3c94aa2136f1513c666f","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"GRAPH RAG","type":"METHOD","description":"Graph RAG is a method that outperformed naive RAG on comprehensiveness and diversity in text generation tasks","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"NAIVE RAG","type":"METHOD","description":"Naive RAG is a baseline method used for comparison in text generation tasks","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"C0","type":"CATEGORY","description":"A category or condition used in the analysis, representing a specific subset of the data","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"C1","type":"CATEGORY","description":"A category or condition used in the analysis, representing a specific subset of the data","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"C2","type":"CATEGORY","description":"A category or condition used in the analysis, representing a specific subset of the data","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"C3","type":"CATEGORY","description":"A category or condition used in the analysis, representing a specific subset of the data","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"TS","type":"CATEGORY","description":"A category or condition used in the analysis, representing a specific subset of the data","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"SS","type":"CATEGORY","description":"A baseline condition used in the analysis, representing a specific subset of the data","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"GPT-4-TURBO","type":"MODEL","description":"GPT-4-Turbo is a model with a large context size of 128k tokens used in the analysis","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"CONTEXT WINDOW SIZE","type":"PARAMETER","description":"The size of the context window used in the analysis, tested at 8k, 16k, 32k, and 64k tokens","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"COMPREHENSIVENESS","type":"METRIC","description":"A metric used to evaluate the quality of answers in terms of their completeness","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"DIVERSITY","type":"METRIC","description":"A metric used to evaluate the variety of answers generated","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"EMPOWERMENT","type":"METRIC","description":"A metric used to evaluate the effectiveness of answers in empowering users","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"KURATOV ET AL., 2024","type":"REFERENCE","description":"A reference to a study by Kuratov et al. in 2024, discussing the potential for information to be lost in longer contexts","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"LIU ET AL., 2023","type":"REFERENCE","description":"A reference to a study by Liu et al. in 2023, discussing the potential for information to be lost in longer contexts","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"DATASETS","type":"DATA","description":"The datasets used in the analysis, consisting of various text sources","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"QUESTIONS","type":"DATA","description":"The questions used in the analysis to evaluate the performance of different methods","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"METRICS","type":"DATA","description":"The metrics used in the analysis to evaluate the performance of different methods","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"HEAD-TO-HEAD WIN RATE","type":"METRIC","description":"A metric used to compare the performance of different conditions in the analysis","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"CONDITION","type":"CATEGORY","description":"A specific setup or scenario used in the analysis, such as C0, C1, C2, C3, TS, and SS","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"WIN RATE","type":"METRIC","description":"The percentage of times a condition outperformed another in the analysis","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"OVERALL WINNER","type":"METRIC","description":"The condition that performed the best across all comparisons in the analysis","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"SELF-WIN RATE","type":"METRIC","description":"The expected win rate of a condition when compared to itself, shown as 50% for reference","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"QUERY-TIME LLM USE","type":"METHOD","description":"The use of large language models (LLMs) at the time of querying, evaluated in the analysis","source_id":"4c855404ee3d3c94aa2136f1513c666f"},{"name":"FINAL EVALUATION","type":"METHOD","description":"The last stage of the analysis where the best performing context window size was used","source_id":"4c855404ee3d3c94aa2136f1513c666f"}],"entity_graph":" METHOD<\/data> Graph RAG is a method that outperformed naive RAG on comprehensiveness and diversity in text generation tasks<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> METHOD<\/data> Naive RAG is a baseline method used for comparison in text generation tasks<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> CATEGORY<\/data> A category or condition used in the analysis, representing a specific subset of the data<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> CATEGORY<\/data> A category or condition used in the analysis, representing a specific subset of the data<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> CATEGORY<\/data> A category or condition used in the analysis, representing a specific subset of the data<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> CATEGORY<\/data> A category or condition used in the analysis, representing a specific subset of the data<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> CATEGORY<\/data> A category or condition used in the analysis, representing a specific subset of the data<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> CATEGORY<\/data> A baseline condition used in the analysis, representing a specific subset of the data<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> MODEL<\/data> GPT-4-Turbo is a model with a large context size of 128k tokens used in the analysis<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> PARAMETER<\/data> The size of the context window used in the analysis, tested at 8k, 16k, 32k, and 64k tokens<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> METRIC<\/data> A metric used to evaluate the quality of answers in terms of their completeness<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> METRIC<\/data> A metric used to evaluate the variety of answers generated<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> METRIC<\/data> A metric used to evaluate the effectiveness of answers in empowering users<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> REFERENCE<\/data> A reference to a study by Kuratov et al. in 2024, discussing the potential for information to be lost in longer contexts<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> REFERENCE<\/data> A reference to a study by Liu et al. in 2023, discussing the potential for information to be lost in longer contexts<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> DATA<\/data> The datasets used in the analysis, consisting of various text sources<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> DATA<\/data> The questions used in the analysis to evaluate the performance of different methods<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> DATA<\/data> The metrics used in the analysis to evaluate the performance of different methods<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> METRIC<\/data> A metric used to compare the performance of different conditions in the analysis<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> CATEGORY<\/data> A specific setup or scenario used in the analysis, such as C0, C1, C2, C3, TS, and SS<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> METRIC<\/data> The percentage of times a condition outperformed another in the analysis<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> METRIC<\/data> The condition that performed the best across all comparisons in the analysis<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> METRIC<\/data> The expected win rate of a condition when compared to itself, shown as 50% for reference<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> METHOD<\/data> The use of large language models (LLMs) at the time of querying, evaluated in the analysis<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> METHOD<\/data> The last stage of the analysis where the best performing context window size was used<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/node> 1.0<\/data> Graph RAG outperformed naive RAG on comprehensiveness and diversity<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> C1 showed slight improvements in answer comprehensiveness and diversity over TS<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> C2 showed slight improvements in answer comprehensiveness and diversity over TS<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> C3 showed slight improvements in answer comprehensiveness and diversity over TS<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> GPT-4-Turbo was tested with varying context window sizes<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> The smallest context window size (8k) was universally better for comprehensiveness<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> The smallest context window size (8k) performed comparably with larger context sizes on diversity<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> The smallest context window size (8k) performed comparably with larger context sizes on empowerment<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> Kuratov et al. discussed the potential for information to be lost in longer contexts<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> Liu et al. discussed the potential for information to be lost in longer contexts<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> Query-time LLM use was evaluated with different context window sizes<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> The final evaluation used a fixed context window size of 8k tokens<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> The final evaluation prioritized comprehensiveness in answers<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> The final evaluation prioritized diversity in answers<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> Datasets were used in combination with questions for the analysis<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> Datasets were evaluated using various metrics<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> Questions were evaluated using various metrics<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> Head-to-head win rate percentages were used to compare different conditions<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> Win rate percentages were used to measure the performance of different conditions<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> The overall winner per dataset and metric was determined for each condition<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> 1.0<\/data> Self-win rates were shown as the expected 50% for each condition<\/data> 4c855404ee3d3c94aa2136f1513c666f<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"36db32c37e1987e2c5863898ad882190","chunk":" (average win rate = 52.4%), and em-\npowerment (average win rate = 51.3%). Given our preference for more comprehensive and diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na \u00a8\u0131ve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na \u00a8\u0131ve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected results,\ni.e., that na \u00a8\u0131ve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsumm","chunk_id":"36db32c37e1987e2c5863898ad882190","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"WIN RATE","type":"METRIC","description":"The percentage of times a particular approach or method achieves a win in a given context","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"EMPOWERMENT","type":"CONCEPT","description":"A concept or metric used in the evaluation, with an average win rate of 51.3%","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"CONTEXT WINDOW SIZE","type":"PARAMETER","description":"The fixed size of the context window used for the final evaluation, set to 8k tokens","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"INDEXING PROCESS","type":"PROCESS","description":"The process that resulted in the creation of graphs for the Podcast and News datasets","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"GRAPH","type":"STRUCTURE","description":"A data structure consisting of nodes and edges, used to represent the Podcast and News datasets","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"PODCAST DATASET","type":"DATASET","description":"A dataset consisting of podcast transcripts, used in the analysis","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"NEWS DATASET","type":"DATASET","description":"A dataset consisting of news articles, used in the analysis","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"NODES","type":"COMPONENT","description":"The individual elements or points in a graph, with 8564 nodes for the Podcast dataset and 15754 nodes for the News dataset","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"EDGES","type":"COMPONENT","description":"The connections or links between nodes in a graph, with 20691 edges for the Podcast dataset and 19520 edges for the News dataset","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"COMMUNITY SUMMARIES","type":"DATA","description":"Summaries of different levels of each graph community hierarchy","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"GLOBAL APPROACHES","type":"METHOD","description":"Approaches that consistently outperformed the naive RAG in comprehensiveness and diversity metrics","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"NAIVE RAG","type":"METHOD","description":"A method that produces the most direct responses but is outperformed by global approaches in comprehensiveness and diversity","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"COMPREHENSIVENESS","type":"METRIC","description":"A metric used to evaluate the thoroughness of responses, with win rates between 72-83% for Podcast transcripts and 72-80% for News articles","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"DIVERSITY","type":"METRIC","description":"A metric used to evaluate the variety of responses, with win rates ranging from 75-82% for Podcast transcripts and 62-71% for News articles","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"DIRECTNESS","type":"METRIC","description":"A validity test metric used to measure the directness of responses, with naive RAG producing the most direct responses","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"C0","type":"CATEGORY","description":"A category or cluster used in the analysis, representing a specific subset of the data","source_id":"36db32c37e1987e2c5863898ad882190","entity_type":"CATEGORY"},{"name":"C1","type":"CATEGORY","description":"A category or cluster used in the analysis, representing a specific subset of the data","source_id":"36db32c37e1987e2c5863898ad882190","entity_type":"CATEGORY"},{"name":"C2","type":"CATEGORY","description":"A category or cluster used in the analysis, representing a specific subset of the data","source_id":"36db32c37e1987e2c5863898ad882190","entity_type":"CATEGORY"},{"name":"C3","type":"CATEGORY","description":"A category or cluster used in the analysis, representing a specific subset of the data","source_id":"36db32c37e1987e2c5863898ad882190","entity_type":"CATEGORY"},{"name":"TS","type":"CATEGORY","description":"A category or cluster used in the analysis, representing a specific subset of the data","source_id":"36db32c37e1987e2c5863898ad882190","entity_type":"CATEGORY"},{"name":"UNITS","type":"METRIC","description":"The number of context units, such as community summaries or text chunks, used in the analysis","source_id":"36db32c37e1987e2c5863898ad882190","entity_type":"METRIC"},{"name":"TOKENS","type":"METRIC","description":"The number of tokens, or individual words, used in the analysis","source_id":"36db32c37e1987e2c5863898ad882190","entity_type":"METRIC"},{"name":"% MAX","type":"METRIC","description":"The percentage of the maximum token count used in the analysis","source_id":"36db32c37e1987e2c5863898ad882190","entity_type":"METRIC"},{"name":"MAP-REDUCE SUMMARIZATION","type":"METHOD","description":"A summarization approach that is the most resource-intensive, requiring the highest number of context tokens","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"ROOT-LEVEL COMMUNITY SUMMARIES","type":"DATA","description":"Summaries at the root level of the community hierarchy, requiring dramatically fewer tokens per query","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"GRAPH RAG","type":"METHOD","description":"A method used to compare community summaries to source texts, generally providing a small but consistent improvement in answer comprehensiveness and diversity","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"FINAL EVALUATION","type":"","description":"","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"PODCAST TRANSCRIPTS","type":"DATASET","description":"A dataset consisting of transcripts from podcasts used for analysis","source_id":"36db32c37e1987e2c5863898ad882190"},{"name":"NEWS ARTICLES","type":"DATASET","description":"A dataset consisting of news articles used for analysis","source_id":"36db32c37e1987e2c5863898ad882190"}],"entity_graph":" METRIC<\/data> The percentage of times a particular approach or method achieves a win in a given context<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> CONCEPT<\/data> A concept or metric used in the evaluation, with an average win rate of 51.3%<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> PARAMETER<\/data> The fixed size of the context window used for the final evaluation, set to 8k tokens<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> PROCESS<\/data> The process that resulted in the creation of graphs for the Podcast and News datasets<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> STRUCTURE<\/data> A data structure consisting of nodes and edges, used to represent the Podcast and News datasets<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> DATASET<\/data> A dataset consisting of podcast transcripts, used in the analysis<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> DATASET<\/data> A dataset consisting of news articles, used in the analysis<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> COMPONENT<\/data> The individual elements or points in a graph, with 8564 nodes for the Podcast dataset and 15754 nodes for the News dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> COMPONENT<\/data> The connections or links between nodes in a graph, with 20691 edges for the Podcast dataset and 19520 edges for the News dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> DATA<\/data> Summaries of different levels of each graph community hierarchy<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> METHOD<\/data> Approaches that consistently outperformed the naive RAG in comprehensiveness and diversity metrics<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> METHOD<\/data> A method that produces the most direct responses but is outperformed by global approaches in comprehensiveness and diversity<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> METRIC<\/data> A metric used to evaluate the thoroughness of responses, with win rates between 72-83% for Podcast transcripts and 72-80% for News articles<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> METRIC<\/data> A metric used to evaluate the variety of responses, with win rates ranging from 75-82% for Podcast transcripts and 62-71% for News articles<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> METRIC<\/data> A validity test metric used to measure the directness of responses, with naive RAG producing the most direct responses<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> CATEGORY<\/data> A category or cluster used in the analysis, representing a specific subset of the data<\/data> 36db32c37e1987e2c5863898ad882190<\/data> CATEGORY<\/data> <\/node> CATEGORY<\/data> A category or cluster used in the analysis, representing a specific subset of the data<\/data> 36db32c37e1987e2c5863898ad882190<\/data> CATEGORY<\/data> <\/node> CATEGORY<\/data> A category or cluster used in the analysis, representing a specific subset of the data<\/data> 36db32c37e1987e2c5863898ad882190<\/data> CATEGORY<\/data> <\/node> CATEGORY<\/data> A category or cluster used in the analysis, representing a specific subset of the data<\/data> 36db32c37e1987e2c5863898ad882190<\/data> CATEGORY<\/data> <\/node> CATEGORY<\/data> A category or cluster used in the analysis, representing a specific subset of the data<\/data> 36db32c37e1987e2c5863898ad882190<\/data> CATEGORY<\/data> <\/node> METRIC<\/data> The number of context units, such as community summaries or text chunks, used in the analysis<\/data> 36db32c37e1987e2c5863898ad882190<\/data> METRIC<\/data> <\/node> METRIC<\/data> The number of tokens, or individual words, used in the analysis<\/data> 36db32c37e1987e2c5863898ad882190<\/data> METRIC<\/data> <\/node> METRIC<\/data> The percentage of the maximum token count used in the analysis<\/data> 36db32c37e1987e2c5863898ad882190<\/data> METRIC<\/data> <\/node> METHOD<\/data> A summarization approach that is the most resource-intensive, requiring the highest number of context tokens<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> DATA<\/data> Summaries at the root level of the community hierarchy, requiring dramatically fewer tokens per query<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> METHOD<\/data> A method used to compare community summaries to source texts, generally providing a small but consistent improvement in answer comprehensiveness and diversity<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> DATASET<\/data> A dataset consisting of transcripts from podcasts used for analysis<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> DATASET<\/data> A dataset consisting of news articles used for analysis<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/node> 1.0<\/data> Empowerment has an average win rate of 51.3%<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> A fixed context window size of 8k tokens was used for the final evaluation<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> The indexing process resulted in the creation of graphs<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> A graph was created for the Podcast dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> A graph was created for the News dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Community summaries are part of the graph community hierarchy<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> The Podcast dataset graph consists of 8564 nodes<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> The Podcast dataset graph consists of 20691 edges<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C0 is a category used in the analysis of the Podcast dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C1 is a category used in the analysis of the Podcast dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C2 is a category used in the analysis of the Podcast dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C3 is a category used in the analysis of the Podcast dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> TS is a category used in the analysis of the Podcast dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Units are used to measure the context in the Podcast dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Tokens are used to measure the word count in the Podcast dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> % Max is used to measure the percentage of maximum token count in the Podcast dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> The News dataset graph consists of 15754 nodes<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> The News dataset graph consists of 19520 edges<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C0 is a category used in the analysis of the News dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C1 is a category used in the analysis of the News dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C2 is a category used in the analysis of the News dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C3 is a category used in the analysis of the News dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> TS is a category used in the analysis of the News dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Units are used to measure the context in the News dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Tokens are used to measure the word count in the News dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> % Max is used to measure the percentage of maximum token count in the News dataset<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Graph RAG is used to compare community summaries to source texts<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Global approaches consistently outperformed the naive RAG<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Global approaches achieved higher comprehensiveness win rates<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Global approaches achieved higher diversity win rates<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Naive RAG produces the most direct responses<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C0 is a category used in the analysis of podcast transcripts<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C0 is a category used in the analysis of news articles<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C1 is a category used in the analysis of podcast transcripts<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C1 is a category used in the analysis of news articles<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C2 is a category used in the analysis of podcast transcripts<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C2 is a category used in the analysis of news articles<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C3 is a category used in the analysis of podcast transcripts<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> C3 is a category used in the analysis of news articles<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> TS is a category used in the analysis of podcast transcripts<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> TS is a category used in the analysis of news articles<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Units are used to measure the context in podcast transcripts<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Units are used to measure the context in news articles<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Map-reduce summarization requires the highest number of context tokens<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Root-level community summaries require dramatically fewer tokens per query<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Tokens are used to measure the word count in podcast transcripts<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Tokens are used to measure the word count in news articles<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> % Max is used to measure the percentage of maximum token count in podcast transcripts<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> % Max is used to measure the percentage of maximum token count in news articles<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> 1.0<\/data> Both are datasets used in the analysis<\/data> 36db32c37e1987e2c5863898ad882190<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"6f33a085ff3304e5994f7fbb86c881a4","chunk":" C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens. For a modest drop in\nperformance compared with other global methods, root-level Graph RAG offers a highly efficient\nmethod for the iterative question answering that characterizes sensemaking activity, while retaining\nadvantages in comprehensiveness (72% win rate) and diversity (62% win rate) over na \u00a8\u0131ve RAG.\nEmpowerment . Empowerment comparisons showed mixed results for both global approaches versus\nna\u00a8\u0131ve RAG ( SS) and Graph RAG approaches versus source text summarization ( TS). Ad-hoc LLM\nuse to analyze LLM reasoning for this measure indicated that the ability to provide specific exam-\nples, quotes, and citations was judged to be key to helping users reach an informed understanding.\nTuning element extraction prompts may help to retain more of these details in the Graph RAG index.\n4 Related Work\n4.1 RAG Approaches and Systems\nWhen using LLMs, RAG involves first retrieving relevant information from external data sources,\nthen adding this information to the context window of the LLM along with the original query (Ram\net al., 2023). Na \u00a8\u0131ve RAG approaches (Gao et al., 2023) do this by converting documents to text,\nsplitting text into chunks, and embedding these chunks into a vector space in which similar positions\nrepresent similar semantics. Queries are then embedded into the same vector space, with the text\nchunks of the nearest kvectors used as context. More advanced variations exist, but all solve the\nproblem of what to do when an external dataset of interest exceeds the LLM","chunk_id":"6f33a085ff3304e5994f7fbb86c881a4","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"GRAPH RAG","type":"METHOD","description":"Graph RAG is a method that uses community summaries to improve answer comprehensiveness and diversity while requiring fewer tokens compared to source text summarization","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"COMMUNITY SUMMARIES","type":"DATASET","description":"Community summaries are summaries derived from community-generated content, used in the analysis to compare with source texts","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"SOURCE TEXTS","type":"DATASET","description":"Source texts are the original texts used for comparison with community summaries in the analysis","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"PODCAST DATASET","type":"DATASET","description":"A dataset consisting of podcast transcripts used for analysis","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"NEWS DATASET","type":"DATASET","description":"A dataset consisting of news articles used for analysis","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"C0","type":"CATEGORY","description":"A category representing root-level community summaries in the analysis","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"C3","type":"CATEGORY","description":"A category representing low-level community summaries in the analysis","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"NAIVE RAG","type":"METHOD","description":"Naive RAG is a basic retrieval-augmented generation method that converts documents to text, splits them into chunks, and embeds these chunks into a vector space for query matching","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"SS","type":"CATEGORY","description":"A category representing na\u00a8\u0131ve RAG in the analysis","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"TS","type":"CATEGORY","description":"A category representing source text summarization in the analysis","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"LLM","type":"TECHNOLOGY","description":"Large Language Models (LLMs) are used to analyze and generate text based on retrieved information and queries","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"EMPOWERMENT","type":"METRIC","description":"Empowerment is a measure used to evaluate the ability of different methods to help users reach an informed understanding","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"RAM ET AL., 2023","type":"REFERENCE","description":"A reference to a paper by Ram et al. in 2023 discussing RAG approaches","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"GAO ET AL., 2023","type":"REFERENCE","description":"A reference to a paper by Gao et al. in 2023 discussing naive RAG approaches","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"RAG","type":"","description":"","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"INTERMEDIATE-LEVEL SUMMARIES","type":"CATEGORY","description":"Intermediate-level summaries are a type of community summary used in the Podcast dataset for analysis","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"LOW-LEVEL COMMUNITY SUMMARIES","type":"CATEGORY","description":"Low-level community summaries are a type of community summary used in the News dataset for analysis","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"ROOT-LEVEL SUMMARIES","type":"CATEGORY","description":"Root-level summaries are a type of community summary used in the analysis","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"ANSWER COMPREHENSIVENESS","type":"METRIC","description":"Answer comprehensiveness is a measure used to evaluate the completeness of answers provided by different methods","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"DIVERSITY","type":"METRIC","description":"Diversity is a measure used to evaluate the variety of answers provided by different methods","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"WIN RATE","type":"METRIC","description":"Win rate is a measure used to evaluate the success rate of different methods in providing comprehensive and diverse answers","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"AD-HOC LLM USE","type":"TECHNOLOGY","description":"Ad-hoc LLM use refers to the spontaneous use of large language models to analyze reasoning and provide specific examples, quotes, and citations","source_id":"6f33a085ff3304e5994f7fbb86c881a4"},{"name":"ELEMENT EXTRACTION PROMPTS","type":"TECHNOLOGY","description":"Element extraction prompts are used to extract specific details in the Graph RAG index","source_id":"6f33a085ff3304e5994f7fbb86c881a4"}],"entity_graph":" METHOD<\/data> Graph RAG is a method that uses community summaries to improve answer comprehensiveness and diversity while requiring fewer tokens compared to source text summarization<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> DATASET<\/data> Community summaries are summaries derived from community-generated content, used in the analysis to compare with source texts<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> DATASET<\/data> Source texts are the original texts used for comparison with community summaries in the analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> DATASET<\/data> A dataset consisting of podcast transcripts used for analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> DATASET<\/data> A dataset consisting of news articles used for analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> CATEGORY<\/data> A category representing root-level community summaries in the analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> CATEGORY<\/data> A category representing low-level community summaries in the analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> METHOD<\/data> Naive RAG is a basic retrieval-augmented generation method that converts documents to text, splits them into chunks, and embeds these chunks into a vector space for query matching<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> CATEGORY<\/data> A category representing na¨ıve RAG in the analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> CATEGORY<\/data> A category representing source text summarization in the analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> TECHNOLOGY<\/data> Large Language Models (LLMs) are used to analyze and generate text based on retrieved information and queries<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> METRIC<\/data> Empowerment is a measure used to evaluate the ability of different methods to help users reach an informed understanding<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> REFERENCE<\/data> A reference to a paper by Ram et al. in 2023 discussing RAG approaches<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> REFERENCE<\/data> A reference to a paper by Gao et al. in 2023 discussing naive RAG approaches<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> CATEGORY<\/data> Intermediate-level summaries are a type of community summary used in the Podcast dataset for analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> CATEGORY<\/data> Low-level community summaries are a type of community summary used in the News dataset for analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> CATEGORY<\/data> Root-level summaries are a type of community summary used in the analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> METRIC<\/data> Answer comprehensiveness is a measure used to evaluate the completeness of answers provided by different methods<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> METRIC<\/data> Diversity is a measure used to evaluate the variety of answers provided by different methods<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> METRIC<\/data> Win rate is a measure used to evaluate the success rate of different methods in providing comprehensive and diverse answers<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> TECHNOLOGY<\/data> Ad-hoc LLM use refers to the spontaneous use of large language models to analyze reasoning and provide specific examples, quotes, and citations<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> TECHNOLOGY<\/data> Element extraction prompts are used to extract specific details in the Graph RAG index<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/node> 1.0<\/data> Graph RAG uses community summaries to improve answer comprehensiveness and diversity<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Graph RAG is compared with source texts for answer comprehensiveness and diversity<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> C0 represents root-level community summaries in the Graph RAG analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> C3 represents low-level community summaries in the Graph RAG analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> TS represents source text summarization in the Graph RAG analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> LLMs are used in Graph RAG to analyze and generate text based on retrieved information and queries<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Empowerment is used to evaluate Graph RAG's ability to help users reach an informed understanding<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Root-level summaries are used in the Graph RAG analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Answer comprehensiveness is used to evaluate the performance of Graph RAG<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Diversity is used to evaluate the performance of Graph RAG<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Win rate is used to measure the success rate of Graph RAG in providing comprehensive and diverse answers<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Element extraction prompts are used in Graph RAG to retain specific details in the index<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Community summaries are derived from the Podcast dataset for analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Community summaries are derived from the News dataset for analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Intermediate-level summaries are derived from the Podcast dataset for analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Low-level community summaries are derived from the News dataset for analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> SS represents naive RAG in the analysis<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Gao et al., 2023 discusses naive RAG approaches<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Ad-hoc LLM use involves the use of large language models to analyze reasoning and provide specific examples, quotes, and citations<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> 1.0<\/data> Ram et al., 2023 discusses RAG approaches<\/data> 6f33a085ff3304e5994f7fbb86c881a4<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"f35de4d9fb65f1d5a392064b20545c19","chunk":" these chunks into a vector space in which similar positions\nrepresent similar semantics. Queries are then embedded into the same vector space, with the text\nchunks of the nearest kvectors used as context. More advanced variations exist, but all solve the\nproblem of what to do when an external dataset of interest exceeds the LLM\u2019s context window.\nAdvanced RAG systems include pre-retrieval, retrieval, post-retrieval strategies designed to over-\ncome the drawbacks of Na \u00a8\u0131ve RAG, while Modular RAG systems include patterns for iterative and\ndynamic cycles of interleaved retrieval and generation (Gao et al., 2023). Our implementation of\nGraph RAG incorporates multiple concepts related to other systems. For example, our community\nsummaries are a kind of self-memory (Selfmem, Cheng et al., 2024) for generation-augmented re-\ntrieval (GAR, Mao et al., 2020) that facilitates future generation cycles, while our parallel generation\nof community answers from these summaries is a kind of iterative (Iter-RetGen, Shao et al., 2023)\nor federated (FeB4RAG, Wang et al., 2024) retrieval-generation strategy. Other systems have also\ncombined these concepts for multi-document summarization (CAiRE-COVID, Su et al., 2020) and\nmulti-hop question answering (ITRG, Feng et al., 2023; IR-CoT, Trivedi et al., 2022; DSP, Khattab\net al., 2022). Our use of a hierarchical index and summarization also bears resemblance to further\napproaches, such as generating a hierarchical index of text chunks by clustering the vectors of text\nembeddings (RAPTOR, Sarthi et al., 2024) or generating a \u201ctree of clarifications\u201d to answer mul-\ntiple interpretations of ambiguous questions (Kim et al., 2023). However, none of these iterative or\nhierarchical approaches use the kind of self-generated graph index that enables Graph RAG.\n104.2 Graphs and LLMs\nUse of graphs in connection with LLMs and RAG is a developing research area, with multiple\ndirections already established. These include using LLMs for knowledge graph creation (Tra-\njanoska et al., 2023) and completion (Yao et al., 2023), as well as for the extraction of causal\ngraphs (Ban et","chunk_id":"f35de4d9fb65f1d5a392064b20545c19","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"VECTOR SPACE","type":"CONCEPT, TECHNOLOGY","description":"A mathematical space in which text chunks and queries are embedded to represent similar semantics","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"QUERIES","type":"CONCEPT, DATA","description":"Search inputs that are embedded into the same vector space as text chunks to find relevant context","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"LLM","type":"TECHNOLOGY, CONCEPT","description":"Large Language Model, a type of AI model with a context window that can be exceeded by external datasets","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"RAG","type":"TECHNOLOGY, METHOD","description":"Retrieval-Augmented Generation, a method that incorporates retrieval of relevant data to augment text generation","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"NA\u00cfVE RAG","type":"TECHNOLOGY, METHOD","description":"A basic form of RAG that has certain drawbacks which advanced RAG systems aim to overcome","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"MODULAR RAG","type":"TECHNOLOGY, METHOD","description":"A type of RAG system that includes patterns for iterative and dynamic cycles of interleaved retrieval and generation","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"GRAPH RAG","type":"TECHNOLOGY, METHOD","description":"An implementation of RAG that incorporates multiple concepts from other systems, including self-memory and parallel generation of community answers","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"SELF-MEMORY (SELFMEM)","type":"TECHNOLOGY, CONCEPT","description":"A concept related to generation-augmented retrieval that facilitates future generation cycles","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"GENERATION-AUGMENTED RETRIEVAL (GAR)","type":"TECHNOLOGY, METHOD","description":"A method that facilitates future generation cycles by using self-memory","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"ITERATIVE RETRIEVAL-GENERATION (ITER-RETGEN)","type":"TECHNOLOGY, METHOD","description":"A strategy for iterative retrieval and generation","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"FEDERATED RETRIEVAL-GENERATION (FEB4RAG)","type":"TECHNOLOGY, METHOD","description":"A federated strategy for retrieval and generation","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"MULTI-DOCUMENT SUMMARIZATION","type":"TECHNOLOGY, METHOD","description":"A method that combines multiple concepts for summarizing multiple documents","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"MULTI-HOP QUESTION ANSWERING","type":"TECHNOLOGY, METHOD","description":"A method for answering questions that require multiple steps or \"hops\" to gather information","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"HIERARCHICAL INDEX","type":"TECHNOLOGY, METHOD","description":"An approach that involves generating a hierarchical index of text chunks by clustering the vectors of text embeddings","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"RAPTOR","type":"TECHNOLOGY, METHOD","description":"A method for generating a hierarchical index of text chunks by clustering the vectors of text embeddings","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"TREE OF CLARIFICATIONS","type":"TECHNOLOGY, METHOD","description":"A method for answering multiple interpretations of ambiguous questions by generating a hierarchical structure","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"GRAPH INDEX","type":"TECHNOLOGY, METHOD","description":"A self-generated index that enables Graph RAG","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"KNOWLEDGE GRAPH CREATION","type":"TECHNOLOGY, METHOD","description":"A process that involves using LLMs to create knowledge graphs","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"KNOWLEDGE GRAPH COMPLETION","type":"TECHNOLOGY, METHOD","description":"A process that involves using LLMs to complete existing knowledge graphs","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"CAUSAL GRAPHS","type":"TECHNOLOGY, METHOD","description":"Graphs that represent causal relationships, which can be extracted using LLMs","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"GAO ET AL., 2023","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Gao et al. in 2023","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"CHENG ET AL., 2024","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Cheng et al. in 2024","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"MAO ET AL., 2020","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Mao et al. in 2020","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"SHAO ET AL., 2023","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Shao et al. in 2023","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"WANG ET AL., 2024","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Wang et al. in 2024","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"SU ET AL., 2020","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Su et al. in 2020","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"FENG ET AL., 2023","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Feng et al. in 2023","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"TRIVEDI ET AL., 2022","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Trivedi et al. in 2022","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"KHATTAB ET AL., 2022","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Khattab et al. in 2022","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"SARTHI ET AL., 2024","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Sarthi et al. in 2024","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"KIM ET AL., 2023","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Kim et al. in 2023","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"TRAJANOSKA ET AL., 2023","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Trajanoska et al. in 2023","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"YAO ET AL., 2023","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Yao et al. in 2023","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"BAN ET AL.","type":"REFERENCE, PUBLICATION","description":"A reference to a publication by Ban et al.","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"REFERENCE, PUBLICATION"},{"name":"COMMUNITY SUMMARIES","type":"TECHNOLOGY, METHOD","description":"Summaries that act as a kind of self-memory for generation-augmented retrieval","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"COMMUNITY ANSWERS","type":"TECHNOLOGY, METHOD","description":"Answers generated in parallel from community summaries","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"MULTI-DOCUMENT SUMMARIZATION (CAIRE-COVID)","type":"TECHNOLOGY, METHOD","description":"A system that combines multiple concepts for multi-document summarization","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"MULTI-HOP QUESTION ANSWERING (ITRG)","type":"TECHNOLOGY, METHOD","description":"A system for multi-hop question answering","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"MULTI-HOP QUESTION ANSWERING (IR-COT)","type":"TECHNOLOGY, METHOD","description":"A system for multi-hop question answering","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"MULTI-HOP QUESTION ANSWERING (DSP)","type":"TECHNOLOGY, METHOD","description":"A system for multi-hop question answering","source_id":"f35de4d9fb65f1d5a392064b20545c19","entity_type":"TECHNOLOGY, METHOD"},{"name":"TEXT CHUNKS","type":"DATA, CONCEPT","description":"Segments of text that are embedded into a vector space for analysis","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"PRE-RETRIEVAL STRATEGIES","type":"TECHNOLOGY, METHOD","description":"Strategies used before the retrieval process in advanced RAG systems","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"RETRIEVAL STRATEGIES","type":"TECHNOLOGY, METHOD","description":"Strategies used during the retrieval process in advanced RAG systems","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"POST-RETRIEVAL STRATEGIES","type":"TECHNOLOGY, METHOD","description":"Strategies used after the retrieval process in advanced RAG systems","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"INTERLEAVED RETRIEVAL AND GENERATION","type":"TECHNOLOGY, METHOD","description":"A pattern in Modular RAG systems for iterative and dynamic cycles of retrieval and generation","source_id":"f35de4d9fb65f1d5a392064b20545c19"},{"name":"GENERATION CYCLES","type":"TECHNOLOGY, METHOD","description":"Cycles of generation that are facilitated by self-memory in Graph RAG","source_id":"f35de4d9fb65f1d5a392064b20545c19"}],"entity_graph":" CONCEPT, TECHNOLOGY<\/data> A mathematical space in which text chunks and queries are embedded to represent similar semantics<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> CONCEPT, DATA<\/data> Search inputs that are embedded into the same vector space as text chunks to find relevant context<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, CONCEPT<\/data> Large Language Model, a type of AI model with a context window that can be exceeded by external datasets<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> Retrieval-Augmented Generation, a method that incorporates retrieval of relevant data to augment text generation<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A basic form of RAG that has certain drawbacks which advanced RAG systems aim to overcome<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A type of RAG system that includes patterns for iterative and dynamic cycles of interleaved retrieval and generation<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> An implementation of RAG that incorporates multiple concepts from other systems, including self-memory and parallel generation of community answers<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, CONCEPT<\/data> A concept related to generation-augmented retrieval that facilitates future generation cycles<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A method that facilitates future generation cycles by using self-memory<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A strategy for iterative retrieval and generation<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A federated strategy for retrieval and generation<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A method that combines multiple concepts for summarizing multiple documents<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A method for answering questions that require multiple steps or \"hops\" to gather information<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> An approach that involves generating a hierarchical index of text chunks by clustering the vectors of text embeddings<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A method for generating a hierarchical index of text chunks by clustering the vectors of text embeddings<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A method for answering multiple interpretations of ambiguous questions by generating a hierarchical structure<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A self-generated index that enables Graph RAG<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A process that involves using LLMs to create knowledge graphs<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A process that involves using LLMs to complete existing knowledge graphs<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> TECHNOLOGY, METHOD<\/data> Graphs that represent causal relationships, which can be extracted using LLMs<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Gao et al. in 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Cheng et al. in 2024<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Mao et al. in 2020<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Shao et al. in 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Wang et al. in 2024<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Su et al. in 2020<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Feng et al. in 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Trivedi et al. in 2022<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Khattab et al. in 2022<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Sarthi et al. in 2024<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Kim et al. in 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Trajanoska et al. in 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Yao et al. in 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> REFERENCE, PUBLICATION<\/data> A reference to a publication by Ban et al.<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> REFERENCE, PUBLICATION<\/data> <\/node> TECHNOLOGY, METHOD<\/data> Summaries that act as a kind of self-memory for generation-augmented retrieval<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> TECHNOLOGY, METHOD<\/data> Answers generated in parallel from community summaries<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A system that combines multiple concepts for multi-document summarization<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A system for multi-hop question answering<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A system for multi-hop question answering<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A system for multi-hop question answering<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> TECHNOLOGY, METHOD<\/data> <\/node> DATA, CONCEPT<\/data> Segments of text that are embedded into a vector space for analysis<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> Strategies used before the retrieval process in advanced RAG systems<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> Strategies used during the retrieval process in advanced RAG systems<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> Strategies used after the retrieval process in advanced RAG systems<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> A pattern in Modular RAG systems for iterative and dynamic cycles of retrieval and generation<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> TECHNOLOGY, METHOD<\/data> Cycles of generation that are facilitated by self-memory in Graph RAG<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/node> 2.0<\/data> Queries are embedded into the same vector space as text chunks to find relevant context<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> RAG is used to augment the capabilities of LLMs<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> LLMs are used for knowledge graph creation<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> LLMs are used for knowledge graph completion<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> LLMs are used for the extraction of causal graphs<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> LLMs are used for knowledge graph creation as per Trajanoska et al., 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> LLMs are used for knowledge graph completion as per Yao et al., 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> LLMs are used for the extraction of causal graphs as per Ban et al.<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Naïve RAG is a basic form of RAG<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Modular RAG is an advanced form of RAG<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG is an implementation of RAG<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates the concept of self-memory<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates the concept of iterative retrieval-generation<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates the concept of federated retrieval-generation<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts used in multi-document summarization<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts used in multi-hop question answering<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG uses a hierarchical index<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates the concept of a tree of clarifications<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG uses a self-generated graph index<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts from Gao et al., 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts from Cheng et al., 2024<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts from Mao et al., 2020<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts from Shao et al., 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts from Wang et al., 2024<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts from Su et al., 2020<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts from Feng et al., 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts from Trivedi et al., 2022<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts from Khattab et al., 2022<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts from Sarthi et al., 2024<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG incorporates concepts from Kim et al., 2023<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG uses community summaries as a kind of self-memory<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Graph RAG generates community answers in parallel<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> Self-memory is related to generation-augmented retrieval<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> CAiRE-COVID is a system for multi-document summarization<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> ITRG is a system for multi-hop question answering<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> IR-CoT is a system for multi-hop question answering<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> DSP is a system for multi-hop question answering<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> 2.0<\/data> RAPTOR is a method for generating a hierarchical index<\/data> f35de4d9fb65f1d5a392064b20545c19<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"92e93fc6449756c0a60200636b297f65","chunk":" RAG is a developing research area, with multiple\ndirections already established. These include using LLMs for knowledge graph creation (Tra-\njanoska et al., 2023) and completion (Yao et al., 2023), as well as for the extraction of causal\ngraphs (Ban et al., 2023; Zhang et al., 2024) from source texts. They also include forms of ad-\nvanced RAG (Gao et al., 2023) where the index is a knowledge graph (KAPING, Baek et al., 2023),\nwhere subsets of the graph structure (G-Retriever, He et al., 2024) or derived graph metrics (Graph-\nToolFormer, Zhang, 2023) are the objects of enquiry, where narrative outputs are strongly grounded\nin the facts of retrieved subgraphs (SURGE, Kang et al., 2023), where retrieved event-plot sub-\ngraphs are serialized using narrative templates (FABULA, Ranade and Joshi, 2023), and where the\nsystem supports both creation and traversal of text-relationship graphs for multi-hop question an-\nswering (Wang et al., 2023b). In terms of open-source software, a variety a graph databases are\nsupported by both the LangChain (LangChain, 2024) and LlamaIndex (LlamaIndex, 2024) libraries,\nwhile a more general class of graph-based RAG applications is also emerging, including systems that\ncan create and reason over knowledge graphs in both Neo4J (NaLLM, Neo4J, 2024) and Nebula-\nGraph (GraphRAG, NebulaGraph, 2024) formats. Unlike our Graph RAG approach, however, none\nof these systems use the natural modularity of graphs to partition data for global summarization.\n5 Discussion\nLimitations of evaluation approach . Our evaluation to date has only examined a certain class of\nsensemaking questions for two corpora in the region of 1 million tokens. More work is needed\nto understand how performance varies across different ranges of question types, data types, and\ndataset sizes, as well as to validate our sensemaking questions and target metrics with end users.\nComparison of fabrication rates, e.g., using approaches like SelfCheckGPT (Manakul et al., 2023),\nwould also improve on the current analysis.\nTrade-offs of building a graph index . We consistently observed","chunk_id":"92e93fc6449756c0a60200636b297f65","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"RAG","type":"METHOD","description":"RAG (Retrieval-Augmented Generation) is a developing research area with multiple established directions, including knowledge graph creation, completion, and extraction of causal graphs","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METHOD"},{"name":"LLMS","type":"TECHNOLOGY","description":"Large Language Models (LLMs) are used for various tasks such as knowledge graph creation and completion","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"TECHNOLOGY"},{"name":"TRAJANOSKA ET AL., 2023","type":"PUBLICATION","description":"A paper by Trajanoska et al. published in 2023, focusing on using LLMs for knowledge graph creation","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"YAO ET AL., 2023","type":"PUBLICATION","description":"A paper by Yao et al. published in 2023, focusing on using LLMs for knowledge graph completion","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"BAN ET AL., 2023","type":"PUBLICATION","description":"A paper by Ban et al. published in 2023, focusing on the extraction of causal graphs from source texts","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"ZHANG ET AL., 2024","type":"PUBLICATION","description":"A paper by Zhang et al. published in 2024, focusing on the extraction of causal graphs from source texts","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"GAO ET AL., 2023","type":"PUBLICATION","description":"A paper by Gao et al. published in 2023, focusing on advanced RAG where the index is a knowledge graph","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"KAPING","type":"METHOD","description":"A method where the index is a knowledge graph, developed by Baek et al. in 2023","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METHOD"},{"name":"BAEK ET AL., 2023","type":"PUBLICATION","description":"A paper by Baek et al. published in 2023, focusing on the KAPING method","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"G-RETRIEVER","type":"METHOD","description":"A method where subsets of the graph structure are the objects of enquiry, developed by He et al. in 2024","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METHOD"},{"name":"HE ET AL., 2024","type":"PUBLICATION","description":"A paper by He et al. published in 2024, focusing on the G-Retriever method","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"GRAPH-TOOLFORMER","type":"METHOD","description":"A method where derived graph metrics are the objects of enquiry, developed by Zhang in 2023","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METHOD"},{"name":"ZHANG, 2023","type":"PUBLICATION","description":"A paper by Zhang published in 2023, focusing on the Graph-ToolFormer method","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"SURGE","type":"METHOD","description":"A method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, developed by Kang et al. in 2023","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METHOD"},{"name":"KANG ET AL., 2023","type":"PUBLICATION","description":"A paper by Kang et al. published in 2023, focusing on the SURGE method","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"FABULA","type":"METHOD","description":"A method where retrieved event-plot subgraphs are serialized using narrative templates, developed by Ranade and Joshi in 2023","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METHOD"},{"name":"RANADE AND JOSHI, 2023","type":"PUBLICATION","description":"A paper by Ranade and Joshi published in 2023, focusing on the FABULA method","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"WANG ET AL., 2023B","type":"PUBLICATION","description":"A paper by Wang et al. published in 2023, focusing on a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"LANGCHAIN","type":"ORGANIZATION","description":"LangChain is an organization that supports a variety of graph databases","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"ORGANIZATION"},{"name":"LLAMAINDEX","type":"ORGANIZATION","description":"LlamaIndex is an organization that supports a variety of graph databases","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"ORGANIZATION"},{"name":"NEO4J","type":"TECHNOLOGY","description":"Neo4J is a graph database format supported by various RAG applications","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"TECHNOLOGY"},{"name":"NALLM","type":"METHOD","description":"A method that can create and reason over knowledge graphs in Neo4J format, developed by Neo4J in 2024","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METHOD"},{"name":"NEBULAGRAPH","type":"TECHNOLOGY","description":"NebulaGraph is a graph database format supported by various RAG applications","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"TECHNOLOGY"},{"name":"GRAPHRAG","type":"METHOD","description":"A method that can create and reason over knowledge graphs in NebulaGraph format, developed by NebulaGraph in 2024","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METHOD"},{"name":"SELFCHECKGPT","type":"METHOD","description":"A method for comparing fabrication rates, developed by Manakul et al. in 2023","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METHOD"},{"name":"MANAKUL ET AL., 2023","type":"PUBLICATION","description":"A paper by Manakul et al. published in 2023, focusing on the SelfCheckGPT method","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PUBLICATION"},{"name":"GRAPH RAG","type":"METHOD","description":"A method that uses the natural modularity of graphs to partition data for global summarization","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METHOD"},{"name":"SENSEMAKING QUESTIONS","type":"CONCEPT","description":"A class of questions used to evaluate the performance of RAG systems","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"CONCEPT"},{"name":"TOKENS","type":"METRIC","description":"The number of individual words used in the analysis, with the evaluation focusing on corpora in the region of 1 million tokens","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METRIC"},{"name":"END USERS","type":"STAKEHOLDER","description":"Individuals who validate sensemaking questions and target metrics","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"STAKEHOLDER"},{"name":"TRADE-OFFS","type":"CONCEPT","description":"Considerations and compromises involved in building a graph index","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"CONCEPT"},{"name":"GRAPH INDEX","type":"TECHNOLOGY","description":"A data structure used in RAG systems to organize and retrieve information","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"TECHNOLOGY"},{"name":"PERFORMANCE","type":"METRIC","description":"The effectiveness of RAG systems, which varies across different ranges of question types, data types, and dataset sizes","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METRIC"},{"name":"DATA TYPES","type":"CONCEPT","description":"Various forms of data used in RAG systems","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"CONCEPT"},{"name":"DATASET SIZES","type":"METRIC","description":"The scale of datasets used in RAG systems, which affects performance","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METRIC"},{"name":"EVALUATION","type":"PROCESS","description":"The process of assessing the performance of RAG systems","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"PROCESS"},{"name":"CORPORA","type":"DATASET","description":"Collections of texts used in the evaluation of RAG systems","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"DATASET"},{"name":"QUESTION TYPES","type":"CONCEPT","description":"Different categories of questions used to evaluate RAG systems","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"CONCEPT"},{"name":"TARGET METRICS","type":"METRIC","description":"Specific measures used to evaluate the performance of RAG systems","source_id":"92e93fc6449756c0a60200636b297f65","entity_type":"METRIC"}],"entity_graph":" METHOD<\/data> RAG (Retrieval-Augmented Generation) is a developing research area with multiple established directions, including knowledge graph creation, completion, and extraction of causal graphs<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METHOD<\/data> <\/node> TECHNOLOGY<\/data> Large Language Models (LLMs) are used for various tasks such as knowledge graph creation and completion<\/data> 92e93fc6449756c0a60200636b297f65<\/data> TECHNOLOGY<\/data> <\/node> PUBLICATION<\/data> A paper by Trajanoska et al. published in 2023, focusing on using LLMs for knowledge graph creation<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> PUBLICATION<\/data> A paper by Yao et al. published in 2023, focusing on using LLMs for knowledge graph completion<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> PUBLICATION<\/data> A paper by Ban et al. published in 2023, focusing on the extraction of causal graphs from source texts<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> PUBLICATION<\/data> A paper by Zhang et al. published in 2024, focusing on the extraction of causal graphs from source texts<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> PUBLICATION<\/data> A paper by Gao et al. published in 2023, focusing on advanced RAG where the index is a knowledge graph<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> METHOD<\/data> A method where the index is a knowledge graph, developed by Baek et al. in 2023<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METHOD<\/data> <\/node> PUBLICATION<\/data> A paper by Baek et al. published in 2023, focusing on the KAPING method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> METHOD<\/data> A method where subsets of the graph structure are the objects of enquiry, developed by He et al. in 2024<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METHOD<\/data> <\/node> PUBLICATION<\/data> A paper by He et al. published in 2024, focusing on the G-Retriever method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> METHOD<\/data> A method where derived graph metrics are the objects of enquiry, developed by Zhang in 2023<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METHOD<\/data> <\/node> PUBLICATION<\/data> A paper by Zhang published in 2023, focusing on the Graph-ToolFormer method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> METHOD<\/data> A method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, developed by Kang et al. in 2023<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METHOD<\/data> <\/node> PUBLICATION<\/data> A paper by Kang et al. published in 2023, focusing on the SURGE method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> METHOD<\/data> A method where retrieved event-plot subgraphs are serialized using narrative templates, developed by Ranade and Joshi in 2023<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METHOD<\/data> <\/node> PUBLICATION<\/data> A paper by Ranade and Joshi published in 2023, focusing on the FABULA method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> PUBLICATION<\/data> A paper by Wang et al. published in 2023, focusing on a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> ORGANIZATION<\/data> LangChain is an organization that supports a variety of graph databases<\/data> 92e93fc6449756c0a60200636b297f65<\/data> ORGANIZATION<\/data> <\/node> ORGANIZATION<\/data> LlamaIndex is an organization that supports a variety of graph databases<\/data> 92e93fc6449756c0a60200636b297f65<\/data> ORGANIZATION<\/data> <\/node> TECHNOLOGY<\/data> Neo4J is a graph database format supported by various RAG applications<\/data> 92e93fc6449756c0a60200636b297f65<\/data> TECHNOLOGY<\/data> <\/node> METHOD<\/data> A method that can create and reason over knowledge graphs in Neo4J format, developed by Neo4J in 2024<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METHOD<\/data> <\/node> TECHNOLOGY<\/data> NebulaGraph is a graph database format supported by various RAG applications<\/data> 92e93fc6449756c0a60200636b297f65<\/data> TECHNOLOGY<\/data> <\/node> METHOD<\/data> A method that can create and reason over knowledge graphs in NebulaGraph format, developed by NebulaGraph in 2024<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METHOD<\/data> <\/node> METHOD<\/data> A method for comparing fabrication rates, developed by Manakul et al. in 2023<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METHOD<\/data> <\/node> PUBLICATION<\/data> A paper by Manakul et al. published in 2023, focusing on the SelfCheckGPT method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PUBLICATION<\/data> <\/node> METHOD<\/data> A method that uses the natural modularity of graphs to partition data for global summarization<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METHOD<\/data> <\/node> CONCEPT<\/data> A class of questions used to evaluate the performance of RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> CONCEPT<\/data> <\/node> METRIC<\/data> The number of individual words used in the analysis, with the evaluation focusing on corpora in the region of 1 million tokens<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METRIC<\/data> <\/node> STAKEHOLDER<\/data> Individuals who validate sensemaking questions and target metrics<\/data> 92e93fc6449756c0a60200636b297f65<\/data> STAKEHOLDER<\/data> <\/node> CONCEPT<\/data> Considerations and compromises involved in building a graph index<\/data> 92e93fc6449756c0a60200636b297f65<\/data> CONCEPT<\/data> <\/node> TECHNOLOGY<\/data> A data structure used in RAG systems to organize and retrieve information<\/data> 92e93fc6449756c0a60200636b297f65<\/data> TECHNOLOGY<\/data> <\/node> METRIC<\/data> The effectiveness of RAG systems, which varies across different ranges of question types, data types, and dataset sizes<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METRIC<\/data> <\/node> CONCEPT<\/data> Various forms of data used in RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> CONCEPT<\/data> <\/node> METRIC<\/data> The scale of datasets used in RAG systems, which affects performance<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METRIC<\/data> <\/node> PROCESS<\/data> The process of assessing the performance of RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> PROCESS<\/data> <\/node> DATASET<\/data> Collections of texts used in the evaluation of RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> DATASET<\/data> <\/node> CONCEPT<\/data> Different categories of questions used to evaluate RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> CONCEPT<\/data> <\/node> METRIC<\/data> Specific measures used to evaluate the performance of RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> METRIC<\/data> <\/node> 2.0<\/data> LLMs are used in various RAG tasks such as knowledge graph creation and completion<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by Trajanoska et al. discusses using LLMs for knowledge graph creation, which is a direction in RAG<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by Yao et al. discusses using LLMs for knowledge graph completion, which is a direction in RAG<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by Ban et al. discusses the extraction of causal graphs, which is a direction in RAG<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by Zhang et al. discusses the extraction of causal graphs, which is a direction in RAG<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by Gao et al. discusses advanced RAG where the index is a knowledge graph<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> KAPING is a method where the index is a knowledge graph, which is a direction in RAG<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> G-Retriever is a method where subsets of the graph structure are the objects of enquiry, which is a direction in RAG<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Graph-ToolFormer is a method where derived graph metrics are the objects of enquiry, which is a direction in RAG<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> SURGE is a method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, which is a direction in RAG<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> FABULA is a method where retrieved event-plot subgraphs are serialized using narrative templates, which is a direction in RAG<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by Wang et al. discusses a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering, which is a direction in RAG<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Graph RAG is a method that uses the natural modularity of graphs to partition data for global summarization<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Sensemaking questions are used to evaluate the performance of RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The evaluation of RAG systems focuses on corpora in the region of 1 million tokens<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Trade-offs are considerations involved in building a graph index for RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> A graph index is a data structure used in RAG systems to organize and retrieve information<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Performance of RAG systems varies across different ranges of question types, data types, and dataset sizes<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Different data types are used in RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Dataset sizes affect the performance of RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Evaluation is the process of assessing the performance of RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Corpora are collections of texts used in the evaluation of RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Different question types are used to evaluate RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Target metrics are specific measures used to evaluate the performance of RAG systems<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by Baek et al. discusses the KAPING method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by He et al. discusses the G-Retriever method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by Zhang discusses the Graph-ToolFormer method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by Kang et al. discusses the SURGE method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by Ranade and Joshi discusses the FABULA method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> Both LangChain and LlamaIndex support a variety of graph databases<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> LangChain supports graph databases in Neo4J format<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> LangChain supports graph databases in NebulaGraph format<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> LlamaIndex supports graph databases in Neo4J format<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> LlamaIndex supports graph databases in NebulaGraph format<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> NaLLM is a method that can create and reason over knowledge graphs in Neo4J format<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> GraphRAG is a method that can create and reason over knowledge graphs in NebulaGraph format<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> The paper by Manakul et al. discusses the SelfCheckGPT method<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> 2.0<\/data> End users validate sensemaking questions and target metrics<\/data> 92e93fc6449756c0a60200636b297f65<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"e4d9b12cf2b4c691c74019eefff4fb39","chunk":"dataset sizes, as well as to validate our sensemaking questions and target metrics with end users.\nComparison of fabrication rates, e.g., using approaches like SelfCheckGPT (Manakul et al., 2023),\nwould also improve on the current analysis.\nTrade-offs of building a graph index . We consistently observed Graph RAG achieve the best head-\nto-head results against other methods, but in many cases the graph-free approach to global summa-\nrization of source texts performed competitively. The real-world decision about whether to invest in\nbuilding a graph index depends on multiple factors, including the compute budget, expected number\nof lifetime queries per dataset, and value obtained from other aspects of the graph index (including\nthe generic community summaries and the use of other graph-related RAG approaches).\nFuture work . The graph index, rich text annotations, and hierarchical community structure support-\ning the current Graph RAG approach offer many possibilities for refinement and adaptation. This\nincludes RAG approaches that operate in a more local manner, via embedding-based matching of\nuser queries and graph annotations, as well as the possibility of hybrid RAG schemes that combine\nembedding-based matching against community reports before employing our map-reduce summa-\nrization mechanisms. This \u201croll-up\u201d operation could also be extended across more levels of the\ncommunity hierarchy, as well as implemented as a more exploratory \u201cdrill down\u201d mechanism that\nfollows the information scent contained in higher-level community summaries.\n6 Conclusion\nWe have presented a global approach to Graph RAG, combining knowledge graph generation,\nretrieval-augmented generation (RAG), and query-focused summarization (QFS) to support human\nsensemaking over entire text corpora. Initial evaluations show substantial improvements over a\nna\u00a8\u0131ve RAG baseline for both the comprehensiveness and diversity of answers, as well as favorable\ncomparisons to a global but graph-free approach using map-reduce source text summarization. For\nsituations requiring many global queries over the same dataset, summaries of root-level communi-\nties in the entity-based graph index provide a data index that is both superior to na \u00a8\u0131ve RAG and\nachieves competitive performance to other global methods at a fraction of the token cost.\nAn open-source, Python-based implementation of both global and local Graph RAG approaches is\nforthcoming at https:\/\/aka .ms\/graphrag .\n11Acknowledgements\nWe would also like to thank the following people who contributed to the work: Alonso","chunk_id":"e4d9b12cf2b4c691c74019eefff4fb39","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"SELFHECKGPT","type":"METHOD","description":"SelfCheckGPT is an approach used to compare fabrication rates in text generation tasks","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"MANAKUL ET AL., 2023","type":"REFERENCE","description":"A reference to the work by Manakul and colleagues published in 2023, related to SelfCheckGPT","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"GRAPH RAG","type":"METHOD","description":"Graph RAG is a method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) to support human sensemaking over text corpora","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"GRAPH-FREE APPROACH","type":"METHOD","description":"A method for global summarization of source texts that does not use a graph index","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"COMPUTE BUDGET","type":"RESOURCE","description":"The amount of computational resources allocated for a task","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"LIFETIME QUERIES","type":"METRIC","description":"The expected number of queries over the lifetime of a dataset","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"COMMUNITY SUMMARIES","type":"DATA","description":"Summaries of root-level communities in an entity-based graph index","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"MAP-REDUCE SUMMARIZATION","type":"METHOD","description":"A method for summarizing source texts using a map-reduce approach","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"RICH TEXT ANNOTATIONS","type":"DATA","description":"Annotations that provide detailed information about the text","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"HIERARCHICAL COMMUNITY STRUCTURE","type":"DATA","description":"A structure that organizes data into a hierarchy of communities","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"EMBEDDING-BASED MATCHING","type":"METHOD","description":"A method that uses embeddings to match user queries with graph annotations","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"HYBRID RAG SCHEMES","type":"METHOD","description":"RAG schemes that combine embedding-based matching with other approaches","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"MAP-REDUCE SUMMARIZATION MECHANISMS","type":"METHOD","description":"Mechanisms used in map-reduce summarization","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"COMMUNITY HIERARCHY","type":"DATA","description":"A hierarchical organization of communities","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"GLOBAL APPROACH TO GRAPH RAG","type":"METHOD","description":"A method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) for global text summarization","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"NA\u00cfVE RAG","type":"METHOD","description":"A baseline method for retrieval-augmented generation (RAG) that does not use advanced techniques","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"TOKEN COST","type":"METRIC","description":"The cost associated with the number of tokens used in a text generation task","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"PYTHON-BASED IMPLEMENTATION","type":"TECHNOLOGY","description":"An implementation of Graph RAG approaches using the Python programming language","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"ALONSO","type":"PERSON","description":"A person who contributed to the work mentioned in the acknowledgements","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"HTTPS:\/\/AKA.MS\/GRAPHRAG","type":"URL","description":"A URL where the open-source, Python-based implementation of Graph RAG approaches will be available","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"FABRICATION RATES","type":"METRIC","description":"The rates at which fabrications occur in text generation tasks","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"SENSEMAKING QUESTIONS","type":"DATA","description":"Questions designed to validate the understanding and interpretation of data","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"TARGET METRICS","type":"METRIC","description":"Specific metrics aimed to be achieved or measured in the analysis","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"END USERS","type":"PERSON","description":"Individuals who are the final users of the system or analysis","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"GRAPH INDEX","type":"DATA","description":"An index built using a graph structure to organize and retrieve information","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"GLOBAL SUMMARIZATION","type":"METHOD","description":"A method for summarizing information on a global scale","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"SOURCE TEXTS","type":"DATA","description":"Original texts from which summaries or analyses are derived","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"LIFETIME QUERIES PER DATASET","type":"METRIC","description":"The expected number of queries over the lifetime of a specific dataset","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"VALUE FROM GRAPH INDEX","type":"METRIC","description":"The benefits or value obtained from using a graph index","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"OTHER GRAPH-RELATED RAG APPROACHES","type":"METHOD","description":"Different methods related to retrieval-augmented generation that utilize graph structures","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"LOCAL GRAPH RAG APPROACHES","type":"METHOD","description":"Graph RAG approaches that operate in a more localized manner","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"USER QUERIES","type":"DATA","description":"Queries made by users to retrieve information","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"GRAPH ANNOTATIONS","type":"DATA","description":"Annotations made on the graph to provide additional information","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"COMMUNITY REPORTS","type":"DATA","description":"Reports generated from community summaries","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"ROLL-UP OPERATION","type":"METHOD","description":"An operation that aggregates information across multiple levels of a hierarchy","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"DRILL DOWN MECHANISM","type":"METHOD","description":"A mechanism that allows for exploring detailed information by following higher-level summaries","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"INFORMATION SCENT","type":"DATA","description":"The trail of information that guides users to more detailed data","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"ROOT-LEVEL COMMUNITIES","type":"DATA","description":"The top-level communities in a hierarchical structure","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"ENTITY-BASED GRAPH INDEX","type":"DATA","description":"A graph index organized around entities","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"},{"name":"OPEN-SOURCE IMPLEMENTATION","type":"TECHNOLOGY","description":"A publicly available implementation of a technology","source_id":"e4d9b12cf2b4c691c74019eefff4fb39"}],"entity_graph":" METHOD<\/data> SelfCheckGPT is an approach used to compare fabrication rates in text generation tasks<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> REFERENCE<\/data> A reference to the work by Manakul and colleagues published in 2023, related to SelfCheckGPT<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> Graph RAG is a method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) to support human sensemaking over text corpora<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> A method for global summarization of source texts that does not use a graph index<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> RESOURCE<\/data> The amount of computational resources allocated for a task<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METRIC<\/data> The expected number of queries over the lifetime of a dataset<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> Summaries of root-level communities in an entity-based graph index<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> A method for summarizing source texts using a map-reduce approach<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> Annotations that provide detailed information about the text<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> A structure that organizes data into a hierarchy of communities<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> A method that uses embeddings to match user queries with graph annotations<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> RAG schemes that combine embedding-based matching with other approaches<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> Mechanisms used in map-reduce summarization<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> A hierarchical organization of communities<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> A method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) for global text summarization<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> A baseline method for retrieval-augmented generation (RAG) that does not use advanced techniques<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METRIC<\/data> The cost associated with the number of tokens used in a text generation task<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> TECHNOLOGY<\/data> An implementation of Graph RAG approaches using the Python programming language<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> PERSON<\/data> A person who contributed to the work mentioned in the acknowledgements<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> URL<\/data> A URL where the open-source, Python-based implementation of Graph RAG approaches will be available<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METRIC<\/data> The rates at which fabrications occur in text generation tasks<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> Questions designed to validate the understanding and interpretation of data<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METRIC<\/data> Specific metrics aimed to be achieved or measured in the analysis<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> PERSON<\/data> Individuals who are the final users of the system or analysis<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> An index built using a graph structure to organize and retrieve information<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> A method for summarizing information on a global scale<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> Original texts from which summaries or analyses are derived<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METRIC<\/data> The expected number of queries over the lifetime of a specific dataset<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METRIC<\/data> The benefits or value obtained from using a graph index<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> Different methods related to retrieval-augmented generation that utilize graph structures<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> Graph RAG approaches that operate in a more localized manner<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> Queries made by users to retrieve information<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> Annotations made on the graph to provide additional information<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> Reports generated from community summaries<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> An operation that aggregates information across multiple levels of a hierarchy<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> METHOD<\/data> A mechanism that allows for exploring detailed information by following higher-level summaries<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> The trail of information that guides users to more detailed data<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> The top-level communities in a hierarchical structure<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> DATA<\/data> A graph index organized around entities<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> TECHNOLOGY<\/data> A publicly available implementation of a technology<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/node> 1.0<\/data> SelfCheckGPT is an approach mentioned in the work by Manakul et al., 2023<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> SelfCheckGPT is used to compare fabrication rates<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG is compared to a graph-free approach for global summarization<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG uses summaries of root-level communities in an entity-based graph index<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG is compared to map-reduce summarization<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG uses rich text annotations<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG uses a hierarchical community structure<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG can operate using embedding-based matching<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG can be part of hybrid RAG schemes<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG can employ map-reduce summarization mechanisms<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG can extend operations across the community hierarchy<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Alonso contributed to the work on Graph RAG<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG uses a graph index<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG includes local graph RAG approaches<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Graph RAG uses an entity-based graph index<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Global summarization can be performed using a graph-free approach<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Summaries of root-level communities are used in Graph RAG<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Embedding-based matching is used to match user queries<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Embedding-based matching is used to match user queries with graph annotations<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Hybrid RAG schemes combine embedding-based matching against community reports<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> The roll-up operation can be extended using map-reduce summarization mechanisms<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> The drill down mechanism follows the information scent in the community hierarchy<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> The global approach to Graph RAG shows improvements over naïve RAG<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> The global approach to Graph RAG achieves competitive performance at a fraction of the token cost<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> The Python-based implementation of Graph RAG approaches will be available at this URL<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> The open-source implementation of Graph RAG approaches is Python-based<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Sensemaking questions are validated with end users<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Target metrics are validated with end users<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> The decision to build a graph index depends on the expected number of lifetime queries per dataset<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> The decision to build a graph index depends on the value obtained from it<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> The decision to build a graph index depends on the value obtained from other graph-related RAG approaches<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> Source texts are used in global summarization<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> 1.0<\/data> The drill down mechanism follows the information scent<\/data> e4d9b12cf2b4c691c74019eefff4fb39<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"086021a89900a39bcb62036981737bfa","chunk":"ieves competitive performance to other global methods at a fraction of the token cost.\nAn open-source, Python-based implementation of both global and local Graph RAG approaches is\nforthcoming at https:\/\/aka .ms\/graphrag .\n11Acknowledgements\nWe would also like to thank the following people who contributed to the work: Alonso Guevara\nFern \u00b4andez, Amber Hoak, Andr \u00b4es Morales Esquivel, Ben Cutler, Billie Rinaldi, Chris Sanchez,\nChris Trevino, Christine Caggiano, David Tittsworth, Dayenne de Souza, Douglas Orbaker, Ed\nClark, Gabriel Nieves-Ponce, Gaudy Blanco Meneses, Kate Lytvynets, Katy Smith, M \u00b4onica Carva-\njal, Nathan Evans, Richard Ortega, Rodrigo Racanicci, Sarah Smith, and Shane Solomon.\nReferences\nAchiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., Almeida, D., Al-\ntenschmidt, J., Altman, S., Anadkat, S., et al. (2023). Gpt-4 technical report. arXiv preprint\narXiv:2303.08774 .\nAnil, R., Borgeaud, S., Wu, Y ., Alayrac, J.-B., Yu, J., Soricut, R., Schalkwyk, J., Dai, A. M.,\nHauth, A., et al. (2023). Gemini: a family of highly capable multimodal models. arXiv preprint\narXiv:2312.11805 .\nBaek, J., Aji, A. F., and Saffari, A. (2023). Knowledge-augmented language model prompting for\nzero-shot knowledge graph question answering. arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:180","chunk_id":"086021a89900a39bcb62036981737bfa","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"GRAPH RAG","type":"METHOD","description":"Graph RAG (Retrieval-Augmented Generation) is a method that combines global and local approaches for efficient token usage in text generation tasks","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"PYTHON","type":"TECHNOLOGY","description":"Python is a programming language used for implementing both global and local Graph RAG approaches","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"ALONSO GUEVARA FERN\u00c1NDEZ","type":"PERSON","description":"Alonso Guevara Fern\u00e1ndez is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"AMBER HOAK","type":"PERSON","description":"Amber Hoak is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"ANDR\u00c9S MORALES ESQUIVEL","type":"PERSON","description":"Andr\u00e9s Morales Esquivel is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"BEN CUTLER","type":"PERSON","description":"Ben Cutler is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"BILLIE RINALDI","type":"PERSON","description":"Billie Rinaldi is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"CHRIS SANCHEZ","type":"PERSON","description":"Chris Sanchez is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"CHRIS TREVINO","type":"PERSON","description":"Chris Trevino is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"CHRISTINE CAGGIANO","type":"PERSON","description":"Christine Caggiano is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"DAVID TITTSWORTH","type":"PERSON","description":"David Tittsworth is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"DAYENNE DE SOUZA","type":"PERSON","description":"Dayenne de Souza is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"DOUGLAS ORBAKER","type":"PERSON","description":"Douglas Orbaker is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"ED CLARK","type":"PERSON","description":"Ed Clark is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"GABRIEL NIEVES-PONCE","type":"PERSON","description":"Gabriel Nieves-Ponce is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"GAUDY BLANCO MENESES","type":"PERSON","description":"Gaudy Blanco Meneses is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"KATE LYTVYNETS","type":"PERSON","description":"Kate Lytvynets is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"KATY SMITH","type":"PERSON","description":"Katy Smith is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"M\u00d3NICA CARVAJAL","type":"PERSON","description":"M\u00f3nica Carvajal is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"NATHAN EVANS","type":"PERSON","description":"Nathan Evans is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"RICHARD ORTEGA","type":"PERSON","description":"Richard Ortega is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"RODRIGO RACANICCI","type":"PERSON","description":"Rodrigo Racanicci is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"SARAH SMITH","type":"PERSON","description":"Sarah Smith is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"SHANE SOLOMON","type":"PERSON","description":"Shane Solomon is a contributor to the work acknowledged in the document","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"GPT-4 TECHNICAL REPORT","type":"PUBLICATION","description":"A technical report on GPT-4 published as an arXiv preprint","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"GEMINI","type":"TECHNOLOGY","description":"Gemini is a family of highly capable multimodal models described in an arXiv preprint","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"KNOWLEDGE-AUGMENTED LANGUAGE MODEL PROMPTING","type":"METHOD","description":"A method for zero-shot knowledge graph question answering described in an arXiv preprint","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"QUERY TOOLS TO CAUSAL ARCHITECTS","type":"METHOD","description":"A method for harnessing large language models for advanced causal discovery from data","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"QUERY FOCUSED ABSTRACTIVE SUMMARIZATION","type":"METHOD","description":"A method incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"J. ACHIAM","type":"PERSON","description":"J. Achiam is an author of the GPT-4 technical report","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"S. ADLER","type":"PERSON","description":"S. Adler is an author of the GPT-4 technical report","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"S. AGARWAL","type":"PERSON","description":"S. Agarwal is an author of the GPT-4 technical report","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"L. AHMAD","type":"PERSON","description":"L. Ahmad is an author of the GPT-4 technical report","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"I. AKKAYA","type":"PERSON","description":"I. Akkaya is an author of the GPT-4 technical report","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"F. L. ALEMAN","type":"PERSON","description":"F. L. Aleman is an author of the GPT-4 technical report","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"D. ALMEIDA","type":"PERSON","description":"D. Almeida is an author of the GPT-4 technical report","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"J. ALTENSCHMIDT","type":"PERSON","description":"J. Altenschmidt is an author of the GPT-4 technical report","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"S. ALTMAN","type":"PERSON","description":"S. Altman is an author of the GPT-4 technical report","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"S. ANADKAT","type":"PERSON","description":"S. Anadkat is an author of the GPT-4 technical report","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"R. ANIL","type":"PERSON","description":"R. Anil is an author of the Gemini paper","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"S. BORGEAUD","type":"PERSON","description":"S. Borgeaud is an author of the Gemini paper","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"Y. WU","type":"PERSON","description":"Y. Wu is an author of the Gemini paper","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"J.-B. ALAYRAC","type":"PERSON","description":"J.-B. Alayrac is an author of the Gemini paper","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"J. YU","type":"PERSON","description":"J. Yu is an author of the Gemini paper","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"R. SORICUT","type":"PERSON","description":"R. Soricut is an author of the Gemini paper","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"J. SCHALKWYK","type":"PERSON","description":"J. Schalkwyk is an author of the Gemini paper","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"A. M. DAI","type":"PERSON","description":"A. M. Dai is an author of the Gemini paper","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"A. HAUTH","type":"PERSON","description":"A. Hauth is an author of the Gemini paper","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"J. BAEK","type":"PERSON","description":"J. Baek is an author of the paper on knowledge-augmented language model prompting","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"A. F. AJI","type":"PERSON","description":"A. F. Aji is an author of the paper on knowledge-augmented language model prompting","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"A. SAFFARI","type":"PERSON","description":"A. Saffari is an author of the paper on knowledge-augmented language model prompting","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"T. BAN","type":"PERSON","description":"T. Ban is an author of the paper on query tools to causal architects","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"L. CHEN","type":"PERSON","description":"L. Chen is an author of the paper on query tools to causal architects","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"X. WANG","type":"PERSON","description":"X. Wang is an author of the paper on query tools to causal architects","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"H. CHEN","type":"PERSON","description":"H. Chen is an author of the paper on query tools to causal architects","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"T. BAUMEL","type":"PERSON","description":"T. Baumel is an author of the paper on query focused abstractive summarization","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"M. EYAL","type":"PERSON","description":"M. Eyal is an author of the paper on query focused abstractive summarization","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"M. ELHADAD","type":"PERSON","description":"M. Elhadad is an author of the paper on query focused abstractive summarization","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"ARXIV","type":"PUBLICATION","description":"arXiv is a repository where the mentioned papers are published as preprints","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"ARXIV:2303.08774","type":"PUBLICATION","description":"The arXiv preprint identifier for the GPT-4 technical report","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"ARXIV:2312.11805","type":"PUBLICATION","description":"The arXiv preprint identifier for the Gemini paper","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"ARXIV:2306.04136","type":"PUBLICATION","description":"The arXiv preprint identifier for the paper on knowledge-augmented language model prompting","source_id":"086021a89900a39bcb62036981737bfa"},{"name":"ARXIV:180","type":"PUBLICATION","description":"The arXiv preprint identifier for the paper on query focused abstractive summarization","source_id":"086021a89900a39bcb62036981737bfa"}],"entity_graph":" METHOD<\/data> Graph RAG (Retrieval-Augmented Generation) is a method that combines global and local approaches for efficient token usage in text generation tasks<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> TECHNOLOGY<\/data> Python is a programming language used for implementing both global and local Graph RAG approaches<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Alonso Guevara Fernández is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Amber Hoak is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Andrés Morales Esquivel is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Ben Cutler is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Billie Rinaldi is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Chris Sanchez is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Chris Trevino is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Christine Caggiano is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> David Tittsworth is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Dayenne de Souza is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Douglas Orbaker is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Ed Clark is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Gabriel Nieves-Ponce is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Gaudy Blanco Meneses is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Kate Lytvynets is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Katy Smith is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Mónica Carvajal is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Nathan Evans is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Richard Ortega is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Rodrigo Racanicci is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Sarah Smith is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Shane Solomon is a contributor to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PUBLICATION<\/data> A technical report on GPT-4 published as an arXiv preprint<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> TECHNOLOGY<\/data> Gemini is a family of highly capable multimodal models described in an arXiv preprint<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> METHOD<\/data> A method for zero-shot knowledge graph question answering described in an arXiv preprint<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> METHOD<\/data> A method for harnessing large language models for advanced causal discovery from data<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> METHOD<\/data> A method incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> J. Achiam is an author of the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> S. Adler is an author of the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> S. Agarwal is an author of the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> L. Ahmad is an author of the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> I. Akkaya is an author of the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> F. L. Aleman is an author of the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> D. Almeida is an author of the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> J. Altenschmidt is an author of the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> S. Altman is an author of the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> S. Anadkat is an author of the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> R. Anil is an author of the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> S. Borgeaud is an author of the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> Y. Wu is an author of the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> J.-B. Alayrac is an author of the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> J. Yu is an author of the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> R. Soricut is an author of the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> J. Schalkwyk is an author of the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> A. M. Dai is an author of the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> A. Hauth is an author of the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> J. Baek is an author of the paper on knowledge-augmented language model prompting<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> A. F. Aji is an author of the paper on knowledge-augmented language model prompting<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> A. Saffari is an author of the paper on knowledge-augmented language model prompting<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> T. Ban is an author of the paper on query tools to causal architects<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> L. Chen is an author of the paper on query tools to causal architects<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> X. Wang is an author of the paper on query tools to causal architects<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> H. Chen is an author of the paper on query tools to causal architects<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> T. Baumel is an author of the paper on query focused abstractive summarization<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> M. Eyal is an author of the paper on query focused abstractive summarization<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PERSON<\/data> M. Elhadad is an author of the paper on query focused abstractive summarization<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PUBLICATION<\/data> arXiv is a repository where the mentioned papers are published as preprints<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PUBLICATION<\/data> The arXiv preprint identifier for the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PUBLICATION<\/data> The arXiv preprint identifier for the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PUBLICATION<\/data> The arXiv preprint identifier for the paper on knowledge-augmented language model prompting<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> PUBLICATION<\/data> The arXiv preprint identifier for the paper on query focused abstractive summarization<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/node> 1.0<\/data> Graph RAG is implemented using Python<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Amber Hoak both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Andrés Morales Esquivel both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Ben Cutler both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Billie Rinaldi both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Chris Sanchez both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Chris Trevino both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Christine Caggiano both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and David Tittsworth both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Dayenne de Souza both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Douglas Orbaker both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Ed Clark both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Gaudy Blanco Meneses both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Kate Lytvynets both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Katy Smith both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Mónica Carvajal both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Nathan Evans both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Richard Ortega both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Rodrigo Racanicci both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Sarah Smith both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Alonso Guevara Fernández and Shane Solomon both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Andrés Morales Esquivel both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Ben Cutler both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Billie Rinaldi both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Chris Sanchez both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Chris Trevino both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Christine Caggiano both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and David Tittsworth both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Dayenne de Souza both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Douglas Orbaker both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Ed Clark both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Gaudy Blanco Meneses both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Kate Lytvynets both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Katy Smith both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Mónica Carvajal both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Nathan Evans both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Richard Ortega both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Amber Hoak and Rodrigo Racanicci both contributed to the work acknowledged in the document<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Achiam and S. Adler co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Achiam and S. Agarwal co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Achiam and L. Ahmad co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Achiam and I. Akkaya co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Achiam and F. L. Aleman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Achiam and D. Almeida co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Achiam and J. Altenschmidt co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Achiam and S. Altman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Achiam and S. Anadkat co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Adler and S. Agarwal co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Adler and L. Ahmad co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Adler and I. Akkaya co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Adler and F. L. Aleman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Adler and D. Almeida co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Adler and J. Altenschmidt co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Adler and S. Altman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Adler and S. Anadkat co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Agarwal and L. Ahmad co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Agarwal and I. Akkaya co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Agarwal and F. L. Aleman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Agarwal and D. Almeida co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Agarwal and J. Altenschmidt co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Agarwal and S. Altman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Agarwal and S. Anadkat co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> L. Ahmad and I. Akkaya co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> L. Ahmad and F. L. Aleman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> L. Ahmad and D. Almeida co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> L. Ahmad and J. Altenschmidt co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> L. Ahmad and S. Altman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> L. Ahmad and S. Anadkat co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> I. Akkaya and F. L. Aleman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> I. Akkaya and D. Almeida co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> I. Akkaya and J. Altenschmidt co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> I. Akkaya and S. Altman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> I. Akkaya and S. Anadkat co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> F. L. Aleman and D. Almeida co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> F. L. Aleman and J. Altenschmidt co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> F. L. Aleman and S. Altman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> F. L. Aleman and S. Anadkat co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> D. Almeida and J. Altenschmidt co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> D. Almeida and S. Altman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> D. Almeida and S. Anadkat co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Altenschmidt and S. Altman co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Altenschmidt and S. Anadkat co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Altman and S. Anadkat co-authored the GPT-4 technical report<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> R. Anil and S. Borgeaud co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> R. Anil and Y. Wu co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> R. Anil and J.-B. Alayrac co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> R. Anil and J. Yu co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> R. Anil and R. Soricut co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> R. Anil and J. Schalkwyk co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> R. Anil and A. M. Dai co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> R. Anil and A. Hauth co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Borgeaud and Y. Wu co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Borgeaud and J.-B. Alayrac co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Borgeaud and J. Yu co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Borgeaud and R. Soricut co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Borgeaud and J. Schalkwyk co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Borgeaud and A. M. Dai co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> S. Borgeaud and A. Hauth co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Y. Wu and J.-B. Alayrac co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Y. Wu and J. Yu co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Y. Wu and R. Soricut co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Y. Wu and J. Schalkwyk co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Y. Wu and A. M. Dai co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> Y. Wu and A. Hauth co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J.-B. Alayrac and J. Yu co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J.-B. Alayrac and R. Soricut co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J.-B. Alayrac and J. Schalkwyk co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J.-B. Alayrac and A. M. Dai co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J.-B. Alayrac and A. Hauth co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Yu and R. Soricut co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Yu and J. Schalkwyk co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Yu and A. M. Dai co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Yu and A. Hauth co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> R. Soricut and J. Schalkwyk co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> R. Soricut and A. M. Dai co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> R. Soricut and A. Hauth co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Schalkwyk and A. M. Dai co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Schalkwyk and A. Hauth co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> A. M. Dai and A. Hauth co-authored the Gemini paper<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Baek and A. F. Aji co-authored the paper on knowledge-augmented language model prompting<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> J. Baek and A. Saffari co-authored the paper on knowledge-augmented language model prompting<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> A. F. Aji and A. Saffari co-authored the paper on knowledge-augmented language model prompting<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> 1.0<\/data> T. Ban and L. Chen co-authored the paper on query tools to causal architects<\/data> 086021a89900a39bcb62036981737bfa<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"58ae80c41cfe46db39da26b6a83584e5","chunk":".\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., Neelakantan, A., Shyam,\nP., Sastry, G., Askell, A., et al. (2020). Language models are few-shot learners. Advances in\nneural information processing systems , 33:1877\u20131901.\nCheng, X., Luo, D., Chen, X., Liu, L., Zhao, D., and Yan, R. (2024). Lift yourself up: Retrieval-\naugmented text generation with self-memory. Advances in Neural Information Processing Sys-\ntems, 36.\nDang, H. T. (2006). Duc 2005: Evaluation of question-focused summarization systems. In Proceed-\nings of the Workshop on Task-Focused Summarization and Question Answering , pages 48\u201355.\nEs, S., James, J., Espinosa-Anke, L., and Schockaert, S. (2023). Ragas: Automated evaluation of\nretrieval augmented generation. arXiv preprint arXiv:2309.15217 .\nFeng, Z., Feng, X., Zhao, D., Yang, M., and Qin, B. (2023). Retrieval-generation synergy augmented\nlarge language models. arXiv preprint arXiv:2310.05149 .\nFortunato, S. (2010). Community detection in graphs. Physics reports , 486(3-5):75\u2013174.\nGao, Y ., Xiong, Y ., Gao, X., Jia, K., Pan, J., Bi, Y ., Dai, Y ., Sun, J., and Wang, H. (2023). Retrieval-\naugmented generation","chunk_id":"58ae80c41cfe46db39da26b6a83584e5","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"BAUMEL, T.","type":"PERSON","description":"Baumel, T. is an author of the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"EYAL, M.","type":"PERSON","description":"Eyal, M. is an author of the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"ELHADAD, M.","type":"PERSON","description":"Elhadad, M. is an author of the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"ARXIV","type":"PUBLICATION","description":"The platform where the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\" was published","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"BLONDEL, V. D.","type":"PERSON","description":"Blondel, V. D. is an author of the paper \"Fast unfolding of communities in large networks\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"GUILLAUME, J.-L.","type":"PERSON","description":"Guillaume, J.-L. is an author of the paper \"Fast unfolding of communities in large networks\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"LAMBIOTTE, R.","type":"PERSON","description":"Lambiotte, R. is an author of the paper \"Fast unfolding of communities in large networks\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"LEFEBVRE, E.","type":"PERSON","description":"Lefebvre, E. is an author of the paper \"Fast unfolding of communities in large networks\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"JOURNAL OF STATISTICAL MECHANICS: THEORY AND EXPERIMENT","type":"PUBLICATION","description":"The journal where the paper \"Fast unfolding of communities in large networks\" was published","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"BROWN, T.","type":"PERSON","description":"Brown, T. is an author of the paper \"Language models are few-shot learners\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"MANN, B.","type":"PERSON","description":"Mann, B. is an author of the paper \"Language models are few-shot learners\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"RYDER, N.","type":"PERSON","description":"Ryder, N. is an author of the paper \"Language models are few-shot learners\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"SUBBIAH, M.","type":"PERSON","description":"Subbiah, M. is an author of the paper \"Language models are few-shot learners\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"KAPLAN, J. D.","type":"PERSON","description":"Kaplan, J. D. is an author of the paper \"Language models are few-shot learners\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"DHARIWAL, P.","type":"PERSON","description":"Dhariwal, P. is an author of the paper \"Language models are few-shot learners\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"NEELAKANTAN, A.","type":"PERSON","description":"Neelakantan, A. is an author of the paper \"Language models are few-shot learners\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"SHYAM, P.","type":"PERSON","description":"Shyam, P. is an author of the paper \"Language models are few-shot learners\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"SASTRY, G.","type":"PERSON","description":"Sastry, G. is an author of the paper \"Language models are few-shot learners\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"ASKELL, A.","type":"PERSON","description":"Askell, A. is an author of the paper \"Language models are few-shot learners\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS","type":"PUBLICATION","description":"The conference where the paper \"Language models are few-shot learners\" was presented","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"CHENG, X.","type":"PERSON","description":"Cheng, X. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"LUO, D.","type":"PERSON","description":"Luo, D. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"CHEN, X.","type":"PERSON","description":"Chen, X. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"LIU, L.","type":"PERSON","description":"Liu, L. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"ZHAO, D.","type":"PERSON","description":"Zhao, D. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"\nZhao, D. is an author of the paper \"Retrieval-generation synergy augmented large language models\"","source_id":"58ae80c41cfe46db39da26b6a83584e5","entity_type":"PERSON"},{"name":"YAN, R.","type":"PERSON","description":"Yan, R. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"DANG, H. T.","type":"PERSON","description":"Dang, H. T. is an author of the paper \"Duc 2005: Evaluation of question-focused summarization systems\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"PROCEEDINGS OF THE WORKSHOP ON TASK-FOCUSED SUMMARIZATION AND QUESTION ANSWERING","type":"PUBLICATION","description":"The conference where the paper \"Duc 2005: Evaluation of question-focused summarization systems\" was presented","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"ES, S.","type":"PERSON","description":"Es, S. is an author of the paper \"Ragas: Automated evaluation of retrieval augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"JAMES, J.","type":"PERSON","description":"James, J. is an author of the paper \"Ragas: Automated evaluation of retrieval augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"ESPINOSA-ANKE, L.","type":"PERSON","description":"Espinosa-Anke, L. is an author of the paper \"Ragas: Automated evaluation of retrieval augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"SCHOCKAERT, S.","type":"PERSON","description":"Schockaert, S. is an author of the paper \"Ragas: Automated evaluation of retrieval augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"FENG, Z.","type":"PERSON","description":"Feng, Z. is an author of the paper \"Retrieval-generation synergy augmented large language models\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"FENG, X.","type":"PERSON","description":"Feng, X. is an author of the paper \"Retrieval-generation synergy augmented large language models\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"YANG, M.","type":"PERSON","description":"Yang, M. is an author of the paper \"Retrieval-generation synergy augmented large language models\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"QIN, B.","type":"PERSON","description":"Qin, B. is an author of the paper \"Retrieval-generation synergy augmented large language models\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"FORTUNATO, S.","type":"PERSON","description":"Fortunato, S. is an author of the paper \"Community detection in graphs\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"PHYSICS REPORTS","type":"PUBLICATION","description":"The journal where the paper \"Community detection in graphs\" was published","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"GAO, Y.","type":"PERSON","description":"Gao, Y. is an author of the paper \"Retrieval-augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"XIONG, Y.","type":"PERSON","description":"Xiong, Y. is an author of the paper \"Retrieval-augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"GAO, X.","type":"PERSON","description":"Gao, X. is an author of the paper \"Retrieval-augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"JIA, K.","type":"PERSON","description":"Jia, K. is an author of the paper \"Retrieval-augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"PAN, J.","type":"PERSON","description":"Pan, J. is an author of the paper \"Retrieval-augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"BI, Y.","type":"PERSON","description":"Bi, Y. is an author of the paper \"Retrieval-augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"DAI, Y.","type":"PERSON","description":"Dai, Y. is an author of the paper \"Retrieval-augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"SUN, J.","type":"PERSON","description":"Sun, J. is an author of the paper \"Retrieval-augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"WANG, H.","type":"PERSON","description":"Wang, H. is an author of the paper \"Retrieval-augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"ARXIV:1801.07704","type":"PUBLICATION","description":"The arXiv identifier for the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"ARXIV:2309.15217","type":"PUBLICATION","description":"The arXiv identifier for the paper \"Ragas: Automated evaluation of retrieval augmented generation\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"},{"name":"ARXIV:2310.05149","type":"PUBLICATION","description":"The arXiv identifier for the paper \"Retrieval-generation synergy augmented large language models\"","source_id":"58ae80c41cfe46db39da26b6a83584e5"}],"entity_graph":" PERSON<\/data> Baumel, T. is an author of the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Eyal, M. is an author of the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Elhadad, M. is an author of the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PUBLICATION<\/data> The platform where the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\" was published<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Blondel, V. D. is an author of the paper \"Fast unfolding of communities in large networks\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Guillaume, J.-L. is an author of the paper \"Fast unfolding of communities in large networks\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Lambiotte, R. is an author of the paper \"Fast unfolding of communities in large networks\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Lefebvre, E. is an author of the paper \"Fast unfolding of communities in large networks\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PUBLICATION<\/data> The journal where the paper \"Fast unfolding of communities in large networks\" was published<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Brown, T. is an author of the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Mann, B. is an author of the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Ryder, N. is an author of the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Subbiah, M. is an author of the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Kaplan, J. D. is an author of the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Dhariwal, P. is an author of the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Neelakantan, A. is an author of the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Shyam, P. is an author of the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Sastry, G. is an author of the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Askell, A. is an author of the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PUBLICATION<\/data> The conference where the paper \"Language models are few-shot learners\" was presented<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Cheng, X. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Luo, D. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Chen, X. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Liu, L. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Zhao, D. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"Zhao, D. is an author of the paper \"Retrieval-generation synergy augmented large language models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> PERSON<\/data> <\/node> PERSON<\/data> Yan, R. is an author of the paper \"Lift yourself up: Retrieval-augmented text generation with self-memory\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Dang, H. T. is an author of the paper \"Duc 2005: Evaluation of question-focused summarization systems\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PUBLICATION<\/data> The conference where the paper \"Duc 2005: Evaluation of question-focused summarization systems\" was presented<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Es, S. is an author of the paper \"Ragas: Automated evaluation of retrieval augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> James, J. is an author of the paper \"Ragas: Automated evaluation of retrieval augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Espinosa-Anke, L. is an author of the paper \"Ragas: Automated evaluation of retrieval augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Schockaert, S. is an author of the paper \"Ragas: Automated evaluation of retrieval augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Feng, Z. is an author of the paper \"Retrieval-generation synergy augmented large language models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Feng, X. is an author of the paper \"Retrieval-generation synergy augmented large language models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Yang, M. is an author of the paper \"Retrieval-generation synergy augmented large language models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Qin, B. is an author of the paper \"Retrieval-generation synergy augmented large language models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Fortunato, S. is an author of the paper \"Community detection in graphs\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PUBLICATION<\/data> The journal where the paper \"Community detection in graphs\" was published<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Gao, Y. is an author of the paper \"Retrieval-augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Xiong, Y. is an author of the paper \"Retrieval-augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Gao, X. is an author of the paper \"Retrieval-augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Jia, K. is an author of the paper \"Retrieval-augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Pan, J. is an author of the paper \"Retrieval-augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Bi, Y. is an author of the paper \"Retrieval-augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Dai, Y. is an author of the paper \"Retrieval-augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Sun, J. is an author of the paper \"Retrieval-augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PERSON<\/data> Wang, H. is an author of the paper \"Retrieval-augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PUBLICATION<\/data> The arXiv identifier for the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PUBLICATION<\/data> The arXiv identifier for the paper \"Ragas: Automated evaluation of retrieval augmented generation\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> PUBLICATION<\/data> The arXiv identifier for the paper \"Retrieval-generation synergy augmented large language models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/node> 1.0<\/data> Baumel, T. and Eyal, M. co-authored the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Baumel, T. and Elhadad, M. co-authored the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Baumel, T. published the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Baumel, T. is an author of the paper with arXiv identifier 1801.07704<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Eyal, M. and Elhadad, M. co-authored the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Eyal, M. published the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Eyal, M. is an author of the paper with arXiv identifier 1801.07704<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Elhadad, M. published the paper \"Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Elhadad, M. is an author of the paper with arXiv identifier 1801.07704<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Es, S. published the paper \"Ragas: Automated evaluation of retrieval augmented generation\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> James, J. published the paper \"Ragas: Automated evaluation of retrieval augmented generation\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Espinosa-Anke, L. published the paper \"Ragas: Automated evaluation of retrieval augmented generation\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Schockaert, S. published the paper \"Ragas: Automated evaluation of retrieval augmented generation\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Feng, Z. published the paper \"Retrieval-generation synergy augmented large language models\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Feng, X. published the paper \"Retrieval-generation synergy augmented large language models\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Zhao, D. published the paper \"Retrieval-generation synergy augmented large language models\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Yang, M. published the paper \"Retrieval-generation synergy augmented large language models\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Qin, B. published the paper \"Retrieval-generation synergy augmented large language models\" on arXiv<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Blondel, V. D. and Guillaume, J.-L. co-authored the paper \"Fast unfolding of communities in large networks\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Blondel, V. D. and Lambiotte, R. co-authored the paper \"Fast unfolding of communities in large networks\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Blondel, V. D. and Lefebvre, E. co-authored the paper \"Fast unfolding of communities in large networks\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Guillaume, J.-L. and Lambiotte, R. co-authored the paper \"Fast unfolding of communities in large networks\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Guillaume, J.-L. and Lefebvre, E. co-authored the paper \"Fast unfolding of communities in large networks\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Lambiotte, R. and Lefebvre, E. co-authored the paper \"Fast unfolding of communities in large networks\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Brown, T. and Mann, B. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Brown, T. and Ryder, N. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Brown, T. and Subbiah, M. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Brown, T. and Kaplan, J. D. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Brown, T. and Dhariwal, P. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Brown, T. and Neelakantan, A. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Brown, T. and Shyam, P. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Brown, T. and Sastry, G. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Brown, T. and Askell, A. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Mann, B. and Ryder, N. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Mann, B. and Subbiah, M. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Mann, B. and Kaplan, J. D. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Mann, B. and Dhariwal, P. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Mann, B. and Neelakantan, A. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Mann, B. and Shyam, P. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Mann, B. and Sastry, G. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Mann, B. and Askell, A. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Ryder, N. and Subbiah, M. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Ryder, N. and Kaplan, J. D. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Ryder, N. and Dhariwal, P. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Ryder, N. and Neelakantan, A. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Ryder, N. and Shyam, P. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Ryder, N. and Sastry, G. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Ryder, N. and Askell, A. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Subbiah, M. and Kaplan, J. D. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Subbiah, M. and Dhariwal, P. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Subbiah, M. and Neelakantan, A. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Subbiah, M. and Shyam, P. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Subbiah, M. and Sastry, G. co-authored the paper \"Language models are few-shot learners\"<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Zhao, D. is an author of the paper with arXiv identifier 2310.05149<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Es, S. is an author of the paper with arXiv identifier 2309.15217<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> James, J. is an author of the paper with arXiv identifier 2309.15217<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Espinosa-Anke, L. is an author of the paper with arXiv identifier 2309.15217<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Schockaert, S. is an author of the paper with arXiv identifier 2309.15217<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Feng, Z. is an author of the paper with arXiv identifier 2310.05149<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Feng, X. is an author of the paper with arXiv identifier 2310.05149<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Yang, M. is an author of the paper with arXiv identifier 2310.05149<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> 1.0<\/data> Qin, B. is an author of the paper with arXiv identifier 2310.05149<\/data> 58ae80c41cfe46db39da26b6a83584e5<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"00e8e4e881bd0862022f4dfc913b900b","chunk":"3-5):75\u2013174.\nGao, Y ., Xiong, Y ., Gao, X., Jia, K., Pan, J., Bi, Y ., Dai, Y ., Sun, J., and Wang, H. (2023). Retrieval-\naugmented generation for large language models: A survey. arXiv preprint arXiv:2312.10997 .\nGoodwin, T. R., Savery, M. E., and Demner-Fushman, D. (2020). Flight of the pegasus? comparing\ntransformers on few-shot and zero-shot multi-document abstractive summarization. In Proceed-\nings of COLING. International Conference on Computational Linguistics , volume 2020, page\n5640. NIH Public Access.\nHe, X., Tian, Y ., Sun, Y ., Chawla, N. V ., Laurent, T., LeCun, Y ., Bresson, X., and Hooi, B. (2024).\nG-retriever: Retrieval-augmented generation for textual graph understanding and question an-\nswering. arXiv preprint arXiv:2402.07630 .\n12Jacomy, M., Venturini, T., Heymann, S., and Bastian, M. (2014). Forceatlas2, a continuous graph\nlayout algorithm for handy network visualization designed for the gephi software. PLoS ONE\n9(6): e98679. https:\/\/doi.org\/10.1371\/journal.pone.0098679 .\nJin, D., Yu, Z., Jiao, P., Pan, S., He, D., Wu, J., Philip, S. Y ., and Zhang, W. (2021). A survey of\ncommunity detection approaches: From statistical modeling to deep learning. IEEE Transactions\non Knowledge and Data Engineering , 35(2):1149\u20131170.\nKang, M., Kwak, J. M., Baek, J., and Hwang, S. J. (2023). Knowledge graph-augmented language\nmodels for knowledge-grounded dialogue generation. arXiv preprint arXiv:2305.18846 .\nKhattab, O., Santhanam, K., Li, X. L., Hall, D., Liang, P., Potts, C., and Zaharia, M. (2022).\n","chunk_id":"00e8e4e881bd0862022f4dfc913b900b","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"GAO, Y.","type":"PERSON","description":"Gao, Y. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"XIONG, Y.","type":"PERSON","description":"Xiong, Y. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"GAO, X.","type":"PERSON","description":"Gao, X. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"JIA, K.","type":"PERSON","description":"Jia, K. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"PAN, J.","type":"PERSON","description":"Pan, J. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"BI, Y.","type":"PERSON","description":"Bi, Y. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"DAI, Y.","type":"PERSON","description":"Dai, Y. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"SUN, J.","type":"PERSON","description":"Sun, J. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"WANG, H.","type":"PERSON","description":"Wang, H. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"ARXIV","type":"PUBLICATION","description":"arXiv is the platform where the paper \"Retrieval-augmented generation for large language models: A survey\" was published\narXiv is the platform where the paper \"Knowledge graph-augmented language models for knowledge-grounded dialogue generation\" was published","source_id":"00e8e4e881bd0862022f4dfc913b900b","entity_type":"PUBLICATION"},{"name":"GOODWIN, T. R.","type":"PERSON","description":"Goodwin, T. R. is an author of the paper \"Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"SAVERY, M. E.","type":"PERSON","description":"Savery, M. E. is an author of the paper \"Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"DEMNER-FUSHMAN, D.","type":"PERSON","description":"Demner-Fushman, D. is an author of the paper \"Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"COLING","type":"CONFERENCE","description":"COLING (International Conference on Computational Linguistics) is the conference where the paper \"Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization\" was presented","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"HE, X.","type":"PERSON","description":"He, X. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"TIAN, Y.","type":"PERSON","description":"Tian, Y. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"SUN, Y.","type":"PERSON","description":"Sun, Y. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"CHAWLA, N. V.","type":"PERSON","description":"Chawla, N. V. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"LAURENT, T.","type":"PERSON","description":"Laurent, T. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"LECUN, Y.","type":"PERSON","description":"LeCun, Y. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"BRESSON, X.","type":"PERSON","description":"Bresson, X. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"HOOI, B.","type":"PERSON","description":"Hooi, B. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"JACOMY, M.","type":"PERSON","description":"Jacomy, M. is an author of the paper \"Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"VENTURINI, T.","type":"PERSON","description":"Venturini, T. is an author of the paper \"Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"HEYMANN, S.","type":"PERSON","description":"Heymann, S. is an author of the paper \"Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"BASTIAN, M.","type":"PERSON","description":"Bastian, M. is an author of the paper \"Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"PLOS ONE","type":"PUBLICATION","description":"PLOS ONE is the journal where the paper \"Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software\" was published","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"JIN, D.","type":"PERSON","description":"Jin, D. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"YU, Z.","type":"PERSON","description":"Yu, Z. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"JIAO, P.","type":"PERSON","description":"Jiao, P. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"PAN, S.","type":"PERSON","description":"Pan, S. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"HE, D.","type":"PERSON","description":"He, D. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"WU, J.","type":"PERSON","description":"Wu, J. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"PHILIP, S. Y.","type":"PERSON","description":"Philip, S. Y. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"ZHANG, W.","type":"PERSON","description":"Zhang, W. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING","type":"PUBLICATION","description":"IEEE Transactions on Knowledge and Data Engineering is the journal where the paper \"A survey of community detection approaches: From statistical modeling to deep learning\" was published","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"KANG, M.","type":"PERSON","description":"Kang, M. is an author of the paper \"Knowledge graph-augmented language models for knowledge-grounded dialogue generation\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"KWAK, J. M.","type":"PERSON","description":"Kwak, J. M. is an author of the paper \"Knowledge graph-augmented language models for knowledge-grounded dialogue generation\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"BAEK, J.","type":"PERSON","description":"Baek, J. is an author of the paper \"Knowledge graph-augmented language models for knowledge-grounded dialogue generation\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"HWANG, S. J.","type":"PERSON","description":"Hwang, S. J. is an author of the paper \"Knowledge graph-augmented language models for knowledge-grounded dialogue generation\"","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"KHATTAB, O.","type":"PERSON","description":"Khattab, O. is an author of the paper mentioned in the text","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"SANTHANAM, K.","type":"PERSON","description":"Santhanam, K. is an author of the paper mentioned in the text","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"LI, X. L.","type":"PERSON","description":"Li, X. L. is an author of the paper mentioned in the text","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"HALL, D.","type":"PERSON","description":"Hall, D. is an author of the paper mentioned in the text","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"LIANG, P.","type":"PERSON","description":"Liang, P. is an author of the paper mentioned in the text","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"POTTS, C.","type":"PERSON","description":"Potts, C. is an author of the paper mentioned in the text","source_id":"00e8e4e881bd0862022f4dfc913b900b"},{"name":"ZAHARIA, M.","type":"PERSON","description":"Zaharia, M. is an author of the paper mentioned in the text","source_id":"00e8e4e881bd0862022f4dfc913b900b"}],"entity_graph":" PERSON<\/data> Gao, Y. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Xiong, Y. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Gao, X. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Jia, K. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Pan, J. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Bi, Y. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Dai, Y. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Sun, J. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Wang, H. is an author of the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PUBLICATION<\/data> arXiv is the platform where the paper \"Retrieval-augmented generation for large language models: A survey\" was publishedarXiv is the platform where the paper \"Knowledge graph-augmented language models for knowledge-grounded dialogue generation\" was published<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> PUBLICATION<\/data> <\/node> PERSON<\/data> Goodwin, T. R. is an author of the paper \"Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Savery, M. E. is an author of the paper \"Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Demner-Fushman, D. is an author of the paper \"Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> CONFERENCE<\/data> COLING (International Conference on Computational Linguistics) is the conference where the paper \"Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization\" was presented<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> He, X. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Tian, Y. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Sun, Y. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Chawla, N. V. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Laurent, T. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> LeCun, Y. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Bresson, X. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Hooi, B. is an author of the paper \"G-retriever: Retrieval-augmented generation for textual graph understanding and question answering\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Jacomy, M. is an author of the paper \"Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Venturini, T. is an author of the paper \"Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Heymann, S. is an author of the paper \"Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Bastian, M. is an author of the paper \"Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PUBLICATION<\/data> PLOS ONE is the journal where the paper \"Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software\" was published<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Jin, D. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Yu, Z. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Jiao, P. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Pan, S. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> He, D. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Wu, J. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Philip, S. Y. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Zhang, W. is an author of the paper \"A survey of community detection approaches: From statistical modeling to deep learning\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PUBLICATION<\/data> IEEE Transactions on Knowledge and Data Engineering is the journal where the paper \"A survey of community detection approaches: From statistical modeling to deep learning\" was published<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Kang, M. is an author of the paper \"Knowledge graph-augmented language models for knowledge-grounded dialogue generation\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Kwak, J. M. is an author of the paper \"Knowledge graph-augmented language models for knowledge-grounded dialogue generation\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Baek, J. is an author of the paper \"Knowledge graph-augmented language models for knowledge-grounded dialogue generation\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Hwang, S. J. is an author of the paper \"Knowledge graph-augmented language models for knowledge-grounded dialogue generation\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Khattab, O. is an author of the paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Santhanam, K. is an author of the paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Li, X. L. is an author of the paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Hall, D. is an author of the paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Liang, P. is an author of the paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Potts, C. is an author of the paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> PERSON<\/data> Zaharia, M. is an author of the paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/node> 1.0<\/data> Gao, Y. and Xiong, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, Y. and Gao, X. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, Y. and Jia, K. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, Y. and Pan, J. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, Y. and Bi, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, Y. and Dai, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, Y. and Sun, J. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, Y. and Wang, H. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Xiong, Y. and Gao, X. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Xiong, Y. and Jia, K. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Xiong, Y. and Pan, J. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Xiong, Y. and Bi, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Xiong, Y. and Dai, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Xiong, Y. and Sun, J. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Xiong, Y. and Wang, H. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, X. and Jia, K. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, X. and Pan, J. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, X. and Bi, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, X. and Dai, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, X. and Sun, J. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Gao, X. and Wang, H. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Jia, K. and Pan, J. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Jia, K. and Bi, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Jia, K. and Dai, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Jia, K. and Sun, J. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Jia, K. and Wang, H. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Pan, J. and Bi, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Pan, J. and Dai, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Pan, J. and Sun, J. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Pan, J. and Wang, H. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Bi, Y. and Dai, Y. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Bi, Y. and Sun, J. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Bi, Y. and Wang, H. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Dai, Y. and Sun, J. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Dai, Y. and Wang, H. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Sun, J. and Wang, H. co-authored the paper \"Retrieval-augmented generation for large language models: A survey\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Goodwin, T. R. and Savery, M. E. co-authored the paper \"Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Goodwin, T. R. and Demner-Fushman, D. co-authored the paper \"Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization\"<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Khattab, O. and Santhanam, K. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Khattab, O. and Li, X. L. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Khattab, O. and Hall, D. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Khattab, O. and Liang, P. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Khattab, O. and Potts, C. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Khattab, O. and Zaharia, M. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Santhanam, K. and Li, X. L. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Santhanam, K. and Hall, D. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Santhanam, K. and Liang, P. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Santhanam, K. and Potts, C. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Santhanam, K. and Zaharia, M. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Li, X. L. and Hall, D. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Li, X. L. and Liang, P. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Li, X. L. and Potts, C. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Li, X. L. and Zaharia, M. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Hall, D. and Liang, P. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Hall, D. and Potts, C. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Hall, D. and Zaharia, M. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Liang, P. and Potts, C. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Liang, P. and Zaharia, M. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> 1.0<\/data> Potts, C. and Zaharia, M. co-authored a paper mentioned in the text<\/data> 00e8e4e881bd0862022f4dfc913b900b<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"71f6daf11e64e5273a3847d46bf228e1","chunk":" for knowledge-grounded dialogue generation. arXiv preprint arXiv:2305.18846 .\nKhattab, O., Santhanam, K., Li, X. L., Hall, D., Liang, P., Potts, C., and Zaharia, M. (2022).\nDemonstrate-search-predict: Composing retrieval and language models for knowledge-intensive\nnlp. arXiv preprint arXiv:2212.14024 .\nKim, G., Kim, S., Jeon, B., Park, J., and Kang, J. (2023). Tree of clarifications: Answering ambigu-\nous questions with retrieval-augmented large language models. arXiv preprint arXiv:2310.14696 .\nKlein, G., Moon, B., and Hoffman, R. R. (2006a). Making sense of sensemaking 1: Alternative\nperspectives. IEEE intelligent systems , 21(4):70\u201373.\nKlein, G., Moon, B., and Hoffman, R. R. (2006b). Making sense of sensemaking 2: A macrocogni-\ntive model. IEEE Intelligent systems , 21(5):88\u201392.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasets\u2013understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https:\/\/python .langchain .com\/docs\/use cases\/graph\/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: 33rd Canadian Conference on Artificial Intelligence, Canadian AI 2020,\nOttawa, ON, Canada, May 13\u201315, 2020, Proceedings 33 , pages 342\u2013348. Springer.\nLaskar, M. T. R., Hoque, E., and Huang, J.","chunk_id":"71f6daf11e64e5273a3847d46bf228e1","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"KHATTAB, O.","type":"PERSON","description":"Khattab, O. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"SANTHANAM, K.","type":"PERSON","description":"Santhanam, K. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"LI, X. L.","type":"PERSON","description":"Li, X. L. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"HALL, D.","type":"PERSON","description":"Hall, D. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"LIANG, P.","type":"PERSON","description":"Liang, P. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"POTTS, C.","type":"PERSON","description":"Potts, C. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"ZAHARIA, M.","type":"PERSON","description":"Zaharia, M. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"KIM, G.","type":"PERSON","description":"Kim, G. is an author of the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"KIM, S.","type":"PERSON","description":"Kim, S. is an author of the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"JEON, B.","type":"PERSON","description":"Jeon, B. is an author of the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"PARK, J.","type":"PERSON","description":"Park, J. is an author of the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"KANG, J.","type":"PERSON","description":"Kang, J. is an author of the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"KLEIN, G.","type":"PERSON","description":"Klein, G. is an author of the papers \"Making sense of sensemaking 1: Alternative perspectives\" and \"Making sense of sensemaking 2: A macrocognitive model\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"MOON, B.","type":"PERSON","description":"Moon, B. is an author of the papers \"Making sense of sensemaking 1: Alternative perspectives\" and \"Making sense of sensemaking 2: A macrocognitive model\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"HOFFMAN, R. R.","type":"PERSON","description":"Hoffman, R. R. is an author of the papers \"Making sense of sensemaking 1: Alternative perspectives\" and \"Making sense of sensemaking 2: A macrocognitive model\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"IEEE INTELLIGENT SYSTEMS","type":"PUBLICATION","description":"The journal where the papers \"Making sense of sensemaking 1: Alternative perspectives\" and \"Making sense of sensemaking 2: A macrocognitive model\" were published","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"KOESTEN, L.","type":"PERSON","description":"Koesten, L. is an author of the paper \"Talking datasets\u2013understanding data sensemaking behaviours\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"GREGORY, K.","type":"PERSON","description":"Gregory, K. is an author of the paper \"Talking datasets\u2013understanding data sensemaking behaviours\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"GROTH, P.","type":"PERSON","description":"Groth, P. is an author of the paper \"Talking datasets\u2013understanding data sensemaking behaviours\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"SIMPERL, E.","type":"PERSON","description":"Simperl, E. is an author of the paper \"Talking datasets\u2013understanding data sensemaking behaviours\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"INTERNATIONAL JOURNAL OF HUMAN-COMPUTER STUDIES","type":"PUBLICATION","description":"The journal where the paper \"Talking datasets\u2013understanding data sensemaking behaviours\" was published","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"KURATOV, Y.","type":"PERSON","description":"Kuratov, Y. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"BULATOV, A.","type":"PERSON","description":"Bulatov, A. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"ANOKHIN, P.","type":"PERSON","description":"Anokhin, P. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"SOROKIN, D.","type":"PERSON","description":"Sorokin, D. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"SOROKIN, A.","type":"PERSON","description":"Sorokin, A. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"BURTSEV, M.","type":"PERSON","description":"Burtsev, M. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"LANGCHAIN","type":"ORGANIZATION","description":"LangChain is an organization that developed Langchain graphs","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"LANGCHAIN GRAPHS","type":"TECHNOLOGY","description":"Langchain graphs is a technology developed by LangChain","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"LASKAR, M. T. R.","type":"PERSON","description":"Laskar, M. T. R. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"HOQUE, E.","type":"PERSON","description":"Hoque, E. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"HUANG, J.","type":"PERSON","description":"Huang, J. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"ADVANCES IN ARTIFICIAL INTELLIGENCE","type":"PUBLICATION","description":"The conference where the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\" was presented","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"ARXIV","type":"PUBLICATION","description":"arXiv is a repository of electronic preprints (known as e-prints) approved for publication after moderation, but not full peer review","source_id":"71f6daf11e64e5273a3847d46bf228e1"},{"name":"ARXIV PREPRINT","type":"PUBLICATION","description":"arXiv preprint refers to a preprint of a paper that is available on the arXiv repository","source_id":"71f6daf11e64e5273a3847d46bf228e1"}],"entity_graph":" PERSON<\/data> Khattab, O. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Santhanam, K. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Li, X. L. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Hall, D. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Liang, P. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Potts, C. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Zaharia, M. is an author of the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Kim, G. is an author of the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Kim, S. is an author of the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Jeon, B. is an author of the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Park, J. is an author of the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Kang, J. is an author of the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Klein, G. is an author of the papers \"Making sense of sensemaking 1: Alternative perspectives\" and \"Making sense of sensemaking 2: A macrocognitive model\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Moon, B. is an author of the papers \"Making sense of sensemaking 1: Alternative perspectives\" and \"Making sense of sensemaking 2: A macrocognitive model\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Hoffman, R. R. is an author of the papers \"Making sense of sensemaking 1: Alternative perspectives\" and \"Making sense of sensemaking 2: A macrocognitive model\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PUBLICATION<\/data> The journal where the papers \"Making sense of sensemaking 1: Alternative perspectives\" and \"Making sense of sensemaking 2: A macrocognitive model\" were published<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Koesten, L. is an author of the paper \"Talking datasets–understanding data sensemaking behaviours\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Gregory, K. is an author of the paper \"Talking datasets–understanding data sensemaking behaviours\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Groth, P. is an author of the paper \"Talking datasets–understanding data sensemaking behaviours\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Simperl, E. is an author of the paper \"Talking datasets–understanding data sensemaking behaviours\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PUBLICATION<\/data> The journal where the paper \"Talking datasets–understanding data sensemaking behaviours\" was published<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Kuratov, Y. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Bulatov, A. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Anokhin, P. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Sorokin, D. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Sorokin, A. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Burtsev, M. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> ORGANIZATION<\/data> LangChain is an organization that developed Langchain graphs<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> TECHNOLOGY<\/data> Langchain graphs is a technology developed by LangChain<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Laskar, M. T. R. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Hoque, E. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PERSON<\/data> Huang, J. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PUBLICATION<\/data> The conference where the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\" was presented<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PUBLICATION<\/data> arXiv is a repository of electronic preprints (known as e-prints) approved for publication after moderation, but not full peer review<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> PUBLICATION<\/data> arXiv preprint refers to a preprint of a paper that is available on the arXiv repository<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/node> 1.0<\/data> Khattab, O. and Santhanam, K. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Khattab, O. and Li, X. L. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Khattab, O. and Hall, D. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Khattab, O. and Liang, P. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Khattab, O. and Potts, C. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Khattab, O. and Zaharia, M. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Khattab, O. is an author of the arXiv preprint \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Santhanam, K. and Li, X. L. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Santhanam, K. and Hall, D. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Santhanam, K. and Liang, P. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Santhanam, K. and Potts, C. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Santhanam, K. and Zaharia, M. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Santhanam, K. is an author of the arXiv preprint \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Li, X. L. and Hall, D. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Li, X. L. and Liang, P. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Li, X. L. and Potts, C. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Li, X. L. and Zaharia, M. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Li, X. L. is an author of the arXiv preprint \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Hall, D. and Liang, P. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Hall, D. and Potts, C. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Hall, D. and Zaharia, M. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Hall, D. is an author of the arXiv preprint \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Liang, P. and Potts, C. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Liang, P. and Zaharia, M. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Liang, P. is an author of the arXiv preprint \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Potts, C. and Zaharia, M. co-authored the paper \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Potts, C. is an author of the arXiv preprint \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Zaharia, M. is an author of the arXiv preprint \"Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Kim, G. and Kim, S. co-authored the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Kim, G. and Jeon, B. co-authored the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Kim, G. and Park, J. co-authored the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Kim, G. and Kang, J. co-authored the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Kim, G. is an author of the arXiv preprint \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Kim, S. and Jeon, B. co-authored the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Kim, S. and Park, J. co-authored the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Kim, S. and Kang, J. co-authored the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Kim, S. is an author of the arXiv preprint \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Jeon, B. and Park, J. co-authored the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Jeon, B. and Kang, J. co-authored the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Jeon, B. is an author of the arXiv preprint \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Park, J. and Kang, J. co-authored the paper \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Park, J. is an author of the arXiv preprint \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Kang, J. is an author of the arXiv preprint \"Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Klein, G. and Moon, B. co-authored the papers \"Making sense of sensemaking 1: Alternative perspectives\" and \"Making sense of sensemaking 2: A macrocognitive model\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Klein, G. and Hoffman, R. R. co-authored the papers \"Making sense of sensemaking 1: Alternative perspectives\" and \"Making sense of sensemaking 2: A macrocognitive model\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Klein, G. is an author of papers published in IEEE Intelligent Systems<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Moon, B. and Hoffman, R. R. co-authored the papers \"Making sense of sensemaking 1: Alternative perspectives\" and \"Making sense of sensemaking 2: A macrocognitive model\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Moon, B. is an author of papers published in IEEE Intelligent Systems<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Hoffman, R. R. is an author of papers published in IEEE Intelligent Systems<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Koesten, L. and Gregory, K. co-authored the paper \"Talking datasets–understanding data sensemaking behaviours\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Koesten, L. and Groth, P. co-authored the paper \"Talking datasets–understanding data sensemaking behaviours\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Koesten, L. and Simperl, E. co-authored the paper \"Talking datasets–understanding data sensemaking behaviours\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Koesten, L. is an author of a paper published in International Journal of Human-Computer Studies<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Gregory, K. and Groth, P. co-authored the paper \"Talking datasets–understanding data sensemaking behaviours\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Gregory, K. and Simperl, E. co-authored the paper \"Talking datasets–understanding data sensemaking behaviours\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Gregory, K. is an author of a paper published in International Journal of Human-Computer Studies<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Groth, P. and Simperl, E. co-authored the paper \"Talking datasets–understanding data sensemaking behaviours\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Groth, P. is an author of a paper published in International Journal of Human-Computer Studies<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Simperl, E. is an author of a paper published in International Journal of Human-Computer Studies<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Kuratov, Y. is an author of the arXiv preprint \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Bulatov, A. is an author of the arXiv preprint \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Anokhin, P. is an author of the arXiv preprint \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Sorokin, D. is an author of the arXiv preprint \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Sorokin, A. is an author of the arXiv preprint \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Burtsev, M. is an author of the arXiv preprint \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> LangChain is an organization that has published on arXiv<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> LangChain developed Langchain graphs<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Laskar, M. T. R. is an author of a paper presented at Advances in Artificial Intelligence<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Laskar, M. T. R. and Hoque, E. co-authored the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Laskar, M. T. R. and Huang, J. co-authored the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Hoque, E. is an author of a paper presented at Advances in Artificial Intelligence<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Hoque, E. and Huang, J. co-authored the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> 1.0<\/data> Huang, J. is an author of a paper presented at Advances in Artificial Intelligence<\/data> 71f6daf11e64e5273a3847d46bf228e1<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"6cd82819982879bd164547d2773ba5c7","chunk":"33rd Canadian Conference on Artificial Intelligence, Canadian AI 2020,\nOttawa, ON, Canada, May 13\u201315, 2020, Proceedings 33 , pages 342\u2013348. Springer.\nLaskar, M. T. R., Hoque, E., and Huang, J. X. (2022). Domain adaptation with pre-trained transform-\ners for query-focused abstractive text summarization. Computational Linguistics , 48(2):279\u2013320.\nLewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V ., Goyal, N., K \u00a8uttler, H., Lewis, M., Yih,\nW.-t., Rockt \u00a8aschel, T., et al. (2020). Retrieval-augmented generation for knowledge-intensive nlp\ntasks. Advances in Neural Information Processing Systems , 33:9459\u20139474.\nLiu, N. F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F., and Liang, P. (2023). Lost\nin the middle: How language models use long contexts. arXiv:2307.03172.\nLiu, Y . and Lapata, M. (2019). Hierarchical transformers for multi-document summarization. arXiv\npreprint arXiv:1905.13164 .\nLlamaIndex (2024). LlamaIndex Knowledge Graph Index. https:\/\/docs .llamaindex .ai\/en\/stable\/\nexamples\/index structs\/knowledge graph\/KnowledgeGraphDemo .html.\nManakul, P., Liusie, A., and Gales, M. J. (2023). Selfcheckgpt: Zero-resource black-box hallucina-\ntion detection for generative large language models. arXiv preprint arXiv:2303.08896 .\nMao, Y ., He, P., Liu, X., Shen, Y ., Gao, J., Han, J., and Chen, W. (2020). Generation-augmented\nretrieval for open-domain question answering. arXiv preprint arXiv:2009.08553 .\nMartin, S., Brown, W. M., Klavans, R., and Boyack, K. (2011). Openord: An open-source toolbox\nfor large graph","chunk_id":"6cd82819982879bd164547d2773ba5c7","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"33RD CANADIAN CONFERENCE ON ARTIFICIAL INTELLIGENCE","type":"EVENT","description":"The 33rd Canadian Conference on Artificial Intelligence, held in Ottawa, ON, Canada, from May 13\u201315, 2020","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"CANADIAN AI 2020","type":"EVENT","description":"The 2020 edition of the Canadian Conference on Artificial Intelligence","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"SPRINGER","type":"PUBLISHER","description":"Springer is the publisher of the proceedings of the 33rd Canadian Conference on Artificial Intelligence","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LASKAR, M. T. R.","type":"PERSON","description":"Laskar, M. T. R. is an author of the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"HOQUE, E.","type":"PERSON","description":"Hoque, E. is an author of the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"HUANG, J. X.","type":"PERSON","description":"Huang, J. X. is an author of the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"COMPUTATIONAL LINGUISTICS","type":"PUBLICATION","description":"The journal where the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\" was published","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LEWIS, P.","type":"PERSON","description":"Lewis, P. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"PEREZ, E.","type":"PERSON","description":"Perez, E. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"PIKTUS, A.","type":"PERSON","description":"Piktus, A. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"PETRONI, F.","type":"PERSON","description":"Petroni, F. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"\nPetroni, F. is an author of the paper \"Lost in the middle: How language models use long contexts\"","source_id":"6cd82819982879bd164547d2773ba5c7","entity_type":"PERSON"},{"name":"KARPUKHIN, V.","type":"PERSON","description":"Karpukhin, V. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"GOYAL, N.","type":"PERSON","description":"Goyal, N. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"K\u00dcTTLER, H.","type":"PERSON","description":"K\u00fcttler, H. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LEWIS, M.","type":"PERSON","description":"Lewis, M. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"YIH, W.-T.","type":"PERSON","description":"Yih, W.-T. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"ROCKT\u00c4SCHEL, T.","type":"PERSON","description":"Rockt\u00e4schel, T. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS","type":"PUBLICATION","description":"The conference where the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\" was presented","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LIU, N. F.","type":"PERSON","description":"Liu, N. F. is an author of the paper \"Lost in the middle: How language models use long contexts\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LIN, K.","type":"PERSON","description":"Lin, K. is an author of the paper \"Lost in the middle: How language models use long contexts\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"HEWITT, J.","type":"PERSON","description":"Hewitt, J. is an author of the paper \"Lost in the middle: How language models use long contexts\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"PARANJAPE, A.","type":"PERSON","description":"Paranjape, A. is an author of the paper \"Lost in the middle: How language models use long contexts\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"BEVILACQUA, M.","type":"PERSON","description":"Bevilacqua, M. is an author of the paper \"Lost in the middle: How language models use long contexts\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LIANG, P.","type":"PERSON","description":"Liang, P. is an author of the paper \"Lost in the middle: How language models use long contexts\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"ARXIV","type":"PUBLICATION","description":"The preprint server where the paper \"Lost in the middle: How language models use long contexts\" was published","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LIU, Y.","type":"PERSON","description":"Liu, Y. is an author of the paper \"Hierarchical transformers for multi-document summarization\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LAPATA, M.","type":"PERSON","description":"Lapata, M. is an author of the paper \"Hierarchical transformers for multi-document summarization\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LLAMAINDEX","type":"ORGANIZATION","description":"LlamaIndex is an organization that developed the LlamaIndex Knowledge Graph Index","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LLAMAINDEX KNOWLEDGE GRAPH INDEX","type":"TECHNOLOGY","description":"LlamaIndex Knowledge Graph Index is a technology developed by LlamaIndex","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"MANAKUL, P.","type":"PERSON","description":"Manakul, P. is an author of the paper \"Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LIUSIE, A.","type":"PERSON","description":"Liusie, A. is an author of the paper \"Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"GALES, M. J.","type":"PERSON","description":"Gales, M. J. is an author of the paper \"Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"MAO, Y.","type":"PERSON","description":"Mao, Y. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"HE, P.","type":"PERSON","description":"He, P. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"LIU, X.","type":"PERSON","description":"Liu, X. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"SHEN, Y.","type":"PERSON","description":"Shen, Y. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"GAO, J.","type":"PERSON","description":"Gao, J. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"HAN, J.","type":"PERSON","description":"Han, J. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"CHEN, W.","type":"PERSON","description":"Chen, W. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"MARTIN, S.","type":"PERSON","description":"Martin, S. is an author of the paper \"Openord: An open-source toolbox for large graph\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"BROWN, W. M.","type":"PERSON","description":"Brown, W. M. is an author of the paper \"Openord: An open-source toolbox for large graph\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"KLAVANS, R.","type":"PERSON","description":"Klavans, R. is an author of the paper \"Openord: An open-source toolbox for large graph\"","source_id":"6cd82819982879bd164547d2773ba5c7"},{"name":"BOYACK, K.","type":"PERSON","description":"Boyack, K. is an author of the paper \"Openord: An open-source toolbox for large graph\"","source_id":"6cd82819982879bd164547d2773ba5c7"}],"entity_graph":" EVENT<\/data> The 33rd Canadian Conference on Artificial Intelligence, held in Ottawa, ON, Canada, from May 13–15, 2020<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> EVENT<\/data> The 2020 edition of the Canadian Conference on Artificial Intelligence<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PUBLISHER<\/data> Springer is the publisher of the proceedings of the 33rd Canadian Conference on Artificial Intelligence<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Laskar, M. T. R. is an author of the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Hoque, E. is an author of the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Huang, J. X. is an author of the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PUBLICATION<\/data> The journal where the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\" was published<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Lewis, P. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Perez, E. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Piktus, A. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Petroni, F. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"Petroni, F. is an author of the paper \"Lost in the middle: How language models use long contexts\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> PERSON<\/data> <\/node> PERSON<\/data> Karpukhin, V. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Goyal, N. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Küttler, H. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Lewis, M. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Yih, W.-T. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Rocktäschel, T. is an author of the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PUBLICATION<\/data> The conference where the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\" was presented<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Liu, N. F. is an author of the paper \"Lost in the middle: How language models use long contexts\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Lin, K. is an author of the paper \"Lost in the middle: How language models use long contexts\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Hewitt, J. is an author of the paper \"Lost in the middle: How language models use long contexts\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Paranjape, A. is an author of the paper \"Lost in the middle: How language models use long contexts\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Bevilacqua, M. is an author of the paper \"Lost in the middle: How language models use long contexts\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Liang, P. is an author of the paper \"Lost in the middle: How language models use long contexts\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PUBLICATION<\/data> The preprint server where the paper \"Lost in the middle: How language models use long contexts\" was published<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Liu, Y. is an author of the paper \"Hierarchical transformers for multi-document summarization\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Lapata, M. is an author of the paper \"Hierarchical transformers for multi-document summarization\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> ORGANIZATION<\/data> LlamaIndex is an organization that developed the LlamaIndex Knowledge Graph Index<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> TECHNOLOGY<\/data> LlamaIndex Knowledge Graph Index is a technology developed by LlamaIndex<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Manakul, P. is an author of the paper \"Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Liusie, A. is an author of the paper \"Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Gales, M. J. is an author of the paper \"Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Mao, Y. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> He, P. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Liu, X. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Shen, Y. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Gao, J. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Han, J. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Chen, W. is an author of the paper \"Generation-augmented retrieval for open-domain question answering\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Martin, S. is an author of the paper \"Openord: An open-source toolbox for large graph\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Brown, W. M. is an author of the paper \"Openord: An open-source toolbox for large graph\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Klavans, R. is an author of the paper \"Openord: An open-source toolbox for large graph\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> PERSON<\/data> Boyack, K. is an author of the paper \"Openord: An open-source toolbox for large graph\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/node> 1.0<\/data> The 33rd Canadian Conference on Artificial Intelligence is also known as Canadian AI 2020<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Springer published the proceedings of the 33rd Canadian Conference on Artificial Intelligence<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Laskar, M. T. R. and Hoque, E. co-authored the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Laskar, M. T. R. and Huang, J. X. co-authored the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Laskar, M. T. R. published the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\" in Computational Linguistics<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Hoque, E. and Huang, J. X. co-authored the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Hoque, E. published the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\" in Computational Linguistics<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Huang, J. X. published the paper \"Domain adaptation with pre-trained transformers for query-focused abstractive text summarization\" in Computational Linguistics<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Lewis, P. and Perez, E. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Lewis, P. and Piktus, A. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Lewis, P. and Petroni, F. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Lewis, P. and Karpukhin, V. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Lewis, P. and Goyal, N. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Lewis, P. and Küttler, H. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Lewis, P. and Lewis, M. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Lewis, P. and Yih, W.-T. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Lewis, P. and Rocktäschel, T. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Perez, E. and Piktus, A. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Perez, E. and Petroni, F. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Perez, E. and Karpukhin, V. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Perez, E. and Goyal, N. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Perez, E. and Küttler, H. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Perez, E. and Lewis, M. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Perez, E. and Yih, W.-T. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Perez, E. and Rocktäschel, T. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Piktus, A. and Petroni, F. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Piktus, A. and Karpukhin, V. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Piktus, A. and Goyal, N. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Piktus, A. and Küttler, H. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Piktus, A. and Lewis, M. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Piktus, A. and Yih, W.-T. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Piktus, A. and Rocktäschel, T. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Petroni, F. and Karpukhin, V. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Petroni, F. and Goyal, N. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Petroni, F. and Küttler, H. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Petroni, F. and Lewis, M. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Petroni, F. and Yih, W.-T. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Petroni, F. and Rocktäschel, T. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Karpukhin, V. and Goyal, N. co-authored the paper \"Retrieval-augmented generation for knowledge-intensive NLP tasks\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Martin, S. and Brown, W. M. co-authored the paper \"Openord: An open-source toolbox for large graph\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Martin, S. and Klavans, R. co-authored the paper \"Openord: An open-source toolbox for large graph\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Martin, S. and Boyack, K. co-authored the paper \"Openord: An open-source toolbox for large graph\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Brown, W. M. and Klavans, R. co-authored the paper \"Openord: An open-source toolbox for large graph\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Brown, W. M. and Boyack, K. co-authored the paper \"Openord: An open-source toolbox for large graph\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> 1.0<\/data> Klavans, R. and Boyack, K. co-authored the paper \"Openord: An open-source toolbox for large graph\"<\/data> 6cd82819982879bd164547d2773ba5c7<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"833e7d67dcd30790b26b71c9b5306f6b","chunk":"ugmented\nretrieval for open-domain question answering. arXiv preprint arXiv:2009.08553 .\nMartin, S., Brown, W. M., Klavans, R., and Boyack, K. (2011). Openord: An open-source toolbox\nfor large graph layout. SPIE Conference on Visualization and Data Analysis (VDA) .\nMicrosoft (2023). The impact of large language models on scientific discovery: a preliminary study\nusing gpt-4.\n13NebulaGraph (2024). Nebulagraph launches industry-first graph rag: Retrieval-augmented genera-\ntion with llm based on knowledge graphs. https:\/\/www .nebula-graph .io\/posts\/graph-RAG.\nNeo4J (2024). Project NaLLM. https:\/\/github .com\/neo4j\/NaLLM.\nNewman, M. E. (2006). Modularity and community structure in networks. Proceedings of the\nnational academy of sciences , 103(23):8577\u20138582.\nRam, O., Levine, Y ., Dalmedigos, I., Muhlgay, D., Shashua, A., Leyton-Brown, K., and Shoham,\nY . (2023). In-context retrieval-augmented language models. Transactions of the Association for\nComputational Linguistics , 11:1316\u20131331.\nRanade, P. and Joshi, A. (2023). Fabula: Intelligence report generation using retrieval-augmented\nnarrative construction. arXiv preprint arXiv:2310.13848 .\nSarthi, P., Abdullah, S., Tuli, A., Khanna, S., Goldie, A., and Manning, C. D. (2024). Raptor:\nRecursive abstractive processing for tree-organized retrieval. arXiv preprint arXiv:2401.18059 .\nScott, K. (2024). Behind the Tech. https:\/\/www .microsoft .com\/en-us\/behind-the-tech.\nShao, Z., Gong, Y ., Shen, Y ., Huang, M., Duan, N., and Chen, W. (2023). Enhancing retrieval-\naugmented large language models with iterative retrieval-generation synergy. arXiv preprint\narXiv:2305.15294 .\nSu, D., Xu, Y ., Yu, T., Sidd","chunk_id":"833e7d67dcd30790b26b71c9b5306f6b","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"MARTIN, S.","type":"PERSON","description":"Martin, S. is an author of the paper \"Openord: An open-source toolbox for large graph layout\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"BROWN, W. M.","type":"PERSON","description":"Brown, W. M. is an author of the paper \"Openord: An open-source toolbox for large graph layout\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"KLAVANS, R.","type":"PERSON","description":"Klavans, R. is an author of the paper \"Openord: An open-source toolbox for large graph layout\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"BOYACK, K.","type":"PERSON","description":"Boyack, K. is an author of the paper \"Openord: An open-source toolbox for large graph layout\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"SPIE CONFERENCE ON VISUALIZATION AND DATA ANALYSIS (VDA)","type":"EVENT","description":"The conference where the paper \"Openord: An open-source toolbox for large graph layout\" was presented","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"EVENT"},{"name":"MICROSOFT","type":"ORGANIZATION","description":"Microsoft is an organization that conducted a study on the impact of large language models on scientific discovery using GPT-4","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"ORGANIZATION"},{"name":"GPT-4","type":"TECHNOLOGY","description":"GPT-4 is a large language model used in Microsoft's study on scientific discovery","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"TECHNOLOGY"},{"name":"NEBULAGRAPH","type":"ORGANIZATION","description":"NebulaGraph is an organization that launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"ORGANIZATION"},{"name":"GRAPH RAG","type":"TECHNOLOGY","description":"Graph RAG is a retrieval-augmented generation technology based on knowledge graphs launched by NebulaGraph","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"TECHNOLOGY"},{"name":"NEO4J","type":"ORGANIZATION","description":"Neo4J is an organization that developed Project NaLLM","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"ORGANIZATION"},{"name":"PROJECT NALLM","type":"TECHNOLOGY","description":"Project NaLLM is a project developed by Neo4J","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"TECHNOLOGY"},{"name":"NEWMAN, M. E.","type":"PERSON","description":"Newman, M. E. is the author of the paper \"Modularity and community structure in networks\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES","type":"PUBLICATION","description":"The journal where the paper \"Modularity and community structure in networks\" was published","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PUBLICATION"},{"name":"RAM, O.","type":"PERSON","description":"Ram, O. is an author of the paper \"In-context retrieval-augmented language models\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"LEVINE, Y.","type":"PERSON","description":"Levine, Y. is an author of the paper \"In-context retrieval-augmented language models\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"DALMEDIGOS, I.","type":"PERSON","description":"Dalmedigos, I. is an author of the paper \"In-context retrieval-augmented language models\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"MUHLGAY, D.","type":"PERSON","description":"Muhlgay, D. is an author of the paper \"In-context retrieval-augmented language models\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"SHASHUA, A.","type":"PERSON","description":"Shashua, A. is an author of the paper \"In-context retrieval-augmented language models\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"LEYTON-BROWN, K.","type":"PERSON","description":"Leyton-Brown, K. is an author of the paper \"In-context retrieval-augmented language models\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"SHOHAM, Y.","type":"PERSON","description":"Shoham, Y. is an author of the paper \"In-context retrieval-augmented language models\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"TRANSACTIONS OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS","type":"PUBLICATION","description":"The journal where the paper \"In-context retrieval-augmented language models\" was published","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PUBLICATION"},{"name":"RANADE, P.","type":"PERSON","description":"Ranade, P. is an author of the paper \"Fabula: Intelligence report generation using retrieval-augmented narrative construction\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"JOSHI, A.","type":"PERSON","description":"Joshi, A. is an author of the paper \"Fabula: Intelligence report generation using retrieval-augmented narrative construction\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"SARTHI, P.","type":"PERSON","description":"Sarthi, P. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"ABDULLAH, S.","type":"PERSON","description":"Abdullah, S. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"TULI, A.","type":"PERSON","description":"Tuli, A. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"KHANNA, S.","type":"PERSON","description":"Khanna, S. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"GOLDIE, A.","type":"PERSON","description":"Goldie, A. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"MANNING, C. D.","type":"PERSON","description":"Manning, C. D. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"SCOTT, K.","type":"PERSON","description":"Scott, K. is associated with \"Behind the Tech\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"BEHIND THE TECH","type":"MEDIA","description":"Behind the Tech is a media platform associated with Scott, K.","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"MEDIA"},{"name":"SHAO, Z.","type":"PERSON","description":"Shao, Z. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"GONG, Y.","type":"PERSON","description":"Gong, Y. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"SHEN, Y.","type":"PERSON","description":"Shen, Y. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"HUANG, M.","type":"PERSON","description":"Huang, M. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"DUAN, N.","type":"PERSON","description":"Duan, N. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"CHEN, W.","type":"PERSON","description":"Chen, W. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PERSON"},{"name":"ARXIV","type":"PUBLICATION","description":"arXiv is a preprint repository where several papers mentioned in the text were published","source_id":"833e7d67dcd30790b26b71c9b5306f6b","entity_type":"PUBLICATION"}],"entity_graph":" PERSON<\/data> Martin, S. is an author of the paper \"Openord: An open-source toolbox for large graph layout\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Brown, W. M. is an author of the paper \"Openord: An open-source toolbox for large graph layout\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Klavans, R. is an author of the paper \"Openord: An open-source toolbox for large graph layout\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Boyack, K. is an author of the paper \"Openord: An open-source toolbox for large graph layout\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> EVENT<\/data> The conference where the paper \"Openord: An open-source toolbox for large graph layout\" was presented<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> EVENT<\/data> <\/node> ORGANIZATION<\/data> Microsoft is an organization that conducted a study on the impact of large language models on scientific discovery using GPT-4<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> ORGANIZATION<\/data> <\/node> TECHNOLOGY<\/data> GPT-4 is a large language model used in Microsoft's study on scientific discovery<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> TECHNOLOGY<\/data> <\/node> ORGANIZATION<\/data> NebulaGraph is an organization that launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> ORGANIZATION<\/data> <\/node> TECHNOLOGY<\/data> Graph RAG is a retrieval-augmented generation technology based on knowledge graphs launched by NebulaGraph<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> TECHNOLOGY<\/data> <\/node> ORGANIZATION<\/data> Neo4J is an organization that developed Project NaLLM<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> ORGANIZATION<\/data> <\/node> TECHNOLOGY<\/data> Project NaLLM is a project developed by Neo4J<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> TECHNOLOGY<\/data> <\/node> PERSON<\/data> Newman, M. E. is the author of the paper \"Modularity and community structure in networks\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PUBLICATION<\/data> The journal where the paper \"Modularity and community structure in networks\" was published<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PUBLICATION<\/data> <\/node> PERSON<\/data> Ram, O. is an author of the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Levine, Y. is an author of the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Dalmedigos, I. is an author of the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Muhlgay, D. is an author of the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Shashua, A. is an author of the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Leyton-Brown, K. is an author of the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Shoham, Y. is an author of the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PUBLICATION<\/data> The journal where the paper \"In-context retrieval-augmented language models\" was published<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PUBLICATION<\/data> <\/node> PERSON<\/data> Ranade, P. is an author of the paper \"Fabula: Intelligence report generation using retrieval-augmented narrative construction\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Joshi, A. is an author of the paper \"Fabula: Intelligence report generation using retrieval-augmented narrative construction\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Sarthi, P. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Abdullah, S. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Tuli, A. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Khanna, S. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Goldie, A. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Manning, C. D. is an author of the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Scott, K. is associated with \"Behind the Tech\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> MEDIA<\/data> Behind the Tech is a media platform associated with Scott, K.<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> MEDIA<\/data> <\/node> PERSON<\/data> Shao, Z. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Gong, Y. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Shen, Y. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Huang, M. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Duan, N. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PERSON<\/data> Chen, W. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PERSON<\/data> <\/node> PUBLICATION<\/data> arXiv is a preprint repository where several papers mentioned in the text were published<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> PUBLICATION<\/data> <\/node> 2.0<\/data> Martin, S. and Brown, W. M. co-authored the paper \"Openord: An open-source toolbox for large graph layout\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Martin, S. and Klavans, R. co-authored the paper \"Openord: An open-source toolbox for large graph layout\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Martin, S. and Boyack, K. co-authored the paper \"Openord: An open-source toolbox for large graph layout\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Brown, W. M. and Klavans, R. co-authored the paper \"Openord: An open-source toolbox for large graph layout\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Brown, W. M. and Boyack, K. co-authored the paper \"Openord: An open-source toolbox for large graph layout\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Klavans, R. and Boyack, K. co-authored the paper \"Openord: An open-source toolbox for large graph layout\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Microsoft conducted a study on the impact of large language models on scientific discovery using GPT-4<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> NebulaGraph launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Neo4J developed Project NaLLM<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Newman, M. E. published the paper \"Modularity and community structure in networks\" in the Proceedings of the National Academy of Sciences<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Ram, O. and Levine, Y. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Ram, O. and Dalmedigos, I. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Ram, O. and Muhlgay, D. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Ram, O. and Shashua, A. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Ram, O. and Leyton-Brown, K. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Ram, O. and Shoham, Y. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Levine, Y. and Dalmedigos, I. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Levine, Y. and Muhlgay, D. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Levine, Y. and Shashua, A. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Levine, Y. and Leyton-Brown, K. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Levine, Y. and Shoham, Y. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Dalmedigos, I. and Muhlgay, D. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Dalmedigos, I. and Shashua, A. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Dalmedigos, I. and Leyton-Brown, K. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Dalmedigos, I. and Shoham, Y. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Muhlgay, D. and Shashua, A. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Muhlgay, D. and Leyton-Brown, K. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Muhlgay, D. and Shoham, Y. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Shashua, A. and Leyton-Brown, K. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Shashua, A. and Shoham, Y. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Leyton-Brown, K. and Shoham, Y. co-authored the paper \"In-context retrieval-augmented language models\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Ranade, P. and Joshi, A. co-authored the paper \"Fabula: Intelligence report generation using retrieval-augmented narrative construction\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Sarthi, P. and Abdullah, S. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Sarthi, P. and Tuli, A. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Sarthi, P. and Khanna, S. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Sarthi, P. and Goldie, A. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Sarthi, P. and Manning, C. D. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Abdullah, S. and Tuli, A. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Abdullah, S. and Khanna, S. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Abdullah, S. and Goldie, A. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Abdullah, S. and Manning, C. D. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Tuli, A. and Khanna, S. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Tuli, A. and Goldie, A. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> 2.0<\/data> Tuli, A. and Manning, C. D. co-authored the paper \"Raptor: Recursive abstractive processing for tree-organized retrieval\"<\/data> 833e7d67dcd30790b26b71c9b5306f6b<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"8d87efac8c50cf20cdf26bf61e5e2035","chunk":", Huang, M., Duan, N., and Chen, W. (2023). Enhancing retrieval-\naugmented large language models with iterative retrieval-generation synergy. arXiv preprint\narXiv:2305.15294 .\nSu, D., Xu, Y ., Yu, T., Siddique, F. B., Barezi, E. J., and Fung, P. (2020). Caire-covid: A ques-\ntion answering and query-focused multi-document summarization system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell-connected communities. Scientific Reports , 9(1).\nTrajanoska, M., Stojanov, R., and Trajanov, D. (2023). Enhancing knowledge graph construction\nusing large language models. ArXiv , abs\/2305.04676.\nTrivedi, H., Balasubramanian, N., Khot, T., and Sabharwal, A. (2022). Interleaving retrieval\nwith chain-of-thought reasoning for knowledge-intensive multi-step questions. arXiv preprint\narXiv:2212.10509 .\nWang, J., Liang, Y ., Meng, F., Sun, Z., Shi, H., Li, Z., Xu, J., Qu, J., and Zhou, J. (2023a). Is chatgpt\na good nlg evaluator? a preliminary study. arXiv preprint arXiv:2303.04048 .\nWang, S., Khramtsova","chunk_id":"8d87efac8c50cf20cdf26bf61e5e2035","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"HUANG, M.","type":"PERSON","description":"Huang, M. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"DUAN, N.","type":"PERSON","description":"Duan, N. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"CHEN, W.","type":"PERSON","description":"Chen, W. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"ARXIV","type":"PUBLICATION","description":"arXiv is a repository where the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\" was published\narXiv is a repository where the paper \"Llama 2: Open foundation and fine-tuned chat models\" was published\narXiv is a repository where the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\" was published\narXiv is a repository where the paper \"MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries\" was published\narXiv is a repository where the paper \"Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions\" was published\narXiv is a repository where the paper \"Enhancing knowledge graph construction using large language models\" was published\narXiv is a repository where the paper \"Is chatgpt a good nlg evaluator? a preliminary study\" was published","source_id":"8d87efac8c50cf20cdf26bf61e5e2035","entity_type":"PUBLICATION"},{"name":"SU, D.","type":"PERSON","description":"Su, D. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"XU, Y.","type":"PERSON","description":"Xu, Y. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"YU, T.","type":"PERSON","description":"Yu, T. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"SIDDIQUE, F. B.","type":"PERSON","description":"Siddique, F. B. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"BAREZI, E. J.","type":"PERSON","description":"Barezi, E. J. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"FUNG, P.","type":"PERSON","description":"Fung, P. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"TANG, Y.","type":"PERSON","description":"Tang, Y. is an author of the paper \"MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"YANG, Y.","type":"PERSON","description":"Yang, Y. is an author of the paper \"MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"TOUVRON, H.","type":"PERSON","description":"Touvron, H. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"MARTIN, L.","type":"PERSON","description":"Martin, L. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"STONE, K.","type":"PERSON","description":"Stone, K. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"ALBERT, P.","type":"PERSON","description":"Albert, P. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"ALMAHAIRI, A.","type":"PERSON","description":"Almahairi, A. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"BABAEI, Y.","type":"PERSON","description":"Babaei, Y. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"BASHLYKOV, N.","type":"PERSON","description":"Bashlykov, N. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"BATRA, S.","type":"PERSON","description":"Batra, S. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"BHARGAVA, P.","type":"PERSON","description":"Bhargava, P. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"BHOSALE, S.","type":"PERSON","description":"Bhosale, S. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"TRAAG, V. A.","type":"PERSON","description":"Traag, V. A. is an author of the paper \"From Louvain to Leiden: guaranteeing well-connected communities\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"WALTMAN, L.","type":"PERSON","description":"Waltman, L. is an author of the paper \"From Louvain to Leiden: guaranteeing well-connected communities\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"VAN ECK, N. J.","type":"PERSON","description":"Van Eck, N. J. is an author of the paper \"From Louvain to Leiden: guaranteeing well-connected communities\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"SCIENTIFIC REPORTS","type":"PUBLICATION","description":"Scientific Reports is the journal where the paper \"From Louvain to Leiden: guaranteeing well-connected communities\" was published","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"TRAJANOSKA, M.","type":"PERSON","description":"Trajanoska, M. is an author of the paper \"Enhancing knowledge graph construction using large language models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"STOJANOV, R.","type":"PERSON","description":"Stojanov, R. is an author of the paper \"Enhancing knowledge graph construction using large language models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"TRAJANOV, D.","type":"PERSON","description":"Trajanov, D. is an author of the paper \"Enhancing knowledge graph construction using large language models\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"TRIVEDI, H.","type":"PERSON","description":"Trivedi, H. is an author of the paper \"Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"BALASUBRAMANIAN, N.","type":"PERSON","description":"Balasubramanian, N. is an author of the paper \"Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"KHOT, T.","type":"PERSON","description":"Khot, T. is an author of the paper \"Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"SABHARWAL, A.","type":"PERSON","description":"Sabharwal, A. is an author of the paper \"Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"WANG, J.","type":"PERSON","description":"Wang, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"LIANG, Y.","type":"PERSON","description":"Liang, Y. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"MENG, F.","type":"PERSON","description":"Meng, F. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"SUN, Z.","type":"PERSON","description":"Sun, Z. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"SHI, H.","type":"PERSON","description":"Shi, H. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"LI, Z.","type":"PERSON","description":"Li, Z. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"XU, J.","type":"PERSON","description":"Xu, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"QU, J.","type":"PERSON","description":"Qu, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"ZHOU, J.","type":"PERSON","description":"Zhou, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"WANG, S.","type":"PERSON","description":"Wang, S. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"},{"name":"KHRAMTSOVA","type":"PERSON","description":"Khramtsova is an author mentioned in the text","source_id":"8d87efac8c50cf20cdf26bf61e5e2035"}],"entity_graph":" PERSON<\/data> Huang, M. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Duan, N. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Chen, W. is an author of the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PUBLICATION<\/data> arXiv is a repository where the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\" was publishedarXiv is a repository where the paper \"Llama 2: Open foundation and fine-tuned chat models\" was publishedarXiv is a repository where the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\" was publishedarXiv is a repository where the paper \"MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries\" was publishedarXiv is a repository where the paper \"Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions\" was publishedarXiv is a repository where the paper \"Enhancing knowledge graph construction using large language models\" was publishedarXiv is a repository where the paper \"Is chatgpt a good nlg evaluator? a preliminary study\" was published<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> PUBLICATION<\/data> <\/node> PERSON<\/data> Su, D. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Xu, Y. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Yu, T. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Siddique, F. B. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Barezi, E. J. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Fung, P. is an author of the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Tang, Y. is an author of the paper \"MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Yang, Y. is an author of the paper \"MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Touvron, H. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Martin, L. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Stone, K. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Albert, P. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Almahairi, A. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Babaei, Y. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Bashlykov, N. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Batra, S. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Bhargava, P. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Bhosale, S. is an author of the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Traag, V. A. is an author of the paper \"From Louvain to Leiden: guaranteeing well-connected communities\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Waltman, L. is an author of the paper \"From Louvain to Leiden: guaranteeing well-connected communities\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Van Eck, N. J. is an author of the paper \"From Louvain to Leiden: guaranteeing well-connected communities\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PUBLICATION<\/data> Scientific Reports is the journal where the paper \"From Louvain to Leiden: guaranteeing well-connected communities\" was published<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Trajanoska, M. is an author of the paper \"Enhancing knowledge graph construction using large language models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Stojanov, R. is an author of the paper \"Enhancing knowledge graph construction using large language models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Trajanov, D. is an author of the paper \"Enhancing knowledge graph construction using large language models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Trivedi, H. is an author of the paper \"Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Balasubramanian, N. is an author of the paper \"Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Khot, T. is an author of the paper \"Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Sabharwal, A. is an author of the paper \"Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Wang, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Liang, Y. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Meng, F. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Sun, Z. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Shi, H. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Li, Z. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Xu, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Qu, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Zhou, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Wang, S. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> PERSON<\/data> Khramtsova is an author mentioned in the text<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/node> 1.0<\/data> Huang, M. and Duan, N. co-authored the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Huang, M. and Chen, W. co-authored the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Duan, N. and Chen, W. co-authored the paper \"Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Su, D. and Xu, Y. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Su, D. and Yu, T. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Su, D. and Siddique, F. B. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Su, D. and Barezi, E. J. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Su, D. and Fung, P. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Xu, Y. and Yu, T. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Xu, Y. and Siddique, F. B. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Xu, Y. and Barezi, E. J. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Xu, Y. and Fung, P. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Yu, T. and Siddique, F. B. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Yu, T. and Barezi, E. J. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Yu, T. and Fung, P. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Siddique, F. B. and Barezi, E. J. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Siddique, F. B. and Fung, P. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Barezi, E. J. and Fung, P. co-authored the paper \"Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Tang, Y. and Yang, Y. co-authored the paper \"MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Touvron, H. and Martin, L. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Touvron, H. and Stone, K. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Touvron, H. and Albert, P. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Touvron, H. and Almahairi, A. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Touvron, H. and Babaei, Y. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Touvron, H. and Bashlykov, N. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Touvron, H. and Batra, S. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Touvron, H. and Bhargava, P. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Touvron, H. and Bhosale, S. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Martin, L. and Stone, K. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Martin, L. and Albert, P. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Martin, L. and Almahairi, A. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Martin, L. and Babaei, Y. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Martin, L. and Bashlykov, N. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Martin, L. and Batra, S. co-authored the paper \"Llama 2: Open foundation and fine-tuned chat models\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Wang, J. and Liang, Y. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Wang, J. and Meng, F. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Wang, J. and Sun, Z. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Wang, J. and Shi, H. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Wang, J. and Li, Z. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Wang, J. and Xu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Wang, J. and Qu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Wang, J. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Liang, Y. and Meng, F. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Liang, Y. and Sun, Z. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Liang, Y. and Shi, H. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Liang, Y. and Li, Z. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Liang, Y. and Xu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Liang, Y. and Qu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Liang, Y. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Meng, F. and Sun, Z. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Meng, F. and Shi, H. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Meng, F. and Li, Z. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Meng, F. and Xu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Meng, F. and Qu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Meng, F. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Sun, Z. and Shi, H. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Sun, Z. and Li, Z. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Sun, Z. and Xu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Sun, Z. and Qu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Sun, Z. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Shi, H. and Li, Z. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Shi, H. and Xu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Shi, H. and Qu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Shi, H. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Li, Z. and Xu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Li, Z. and Qu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Li, Z. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Xu, J. and Qu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Xu, J. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> 1.0<\/data> Qu, J. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> 8d87efac8c50cf20cdf26bf61e5e2035<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"fc4b27d64f055b7fc30176ba110dd02e","chunk":" H., Li, Z., Xu, J., Qu, J., and Zhou, J. (2023a). Is chatgpt\na good nlg evaluator? a preliminary study. arXiv preprint arXiv:2303.04048 .\nWang, S., Khramtsova, E., Zhuang, S., and Zuccon, G. (2024). Feb4rag: Evaluating federated search\nin the context of retrieval augmented generation. arXiv preprint arXiv:2402.11891 .\nWang, Y ., Lipka, N., Rossi, R. A., Siu, A., Zhang, R., and Derr, T. (2023b). Knowledge graph\nprompting for multi-document question answering.\nXu, Y . and Lapata, M. (2021). Text summarization with latent queries. arXiv preprint\narXiv:2106.00104 .\nYang, Z., Qi, P., Zhang, S., Bengio, Y ., Cohen, W. W., Salakhutdinov, R., and Manning, C. D. (2018).\nHotpotQA: A dataset for diverse, explainable multi-hop question answering. In Conference on\nEmpirical Methods in Natural Language Processing (EMNLP) .\nYao, J.-g., Wan, X., and Xiao, J. (2017). Recent advances in document summarization. Knowledge\nand Information Systems , 53:297\u2013336.\n14Yao, L., Peng, J., Mao, C., and Luo, Y . (2023). Exploring large language models for knowledge\ngraph completion.\nZhang, J. (2023). Graph-toolformer: To empower llms with graph reasoning ability via prompt\naugmented by chatgpt. arXiv preprint arXiv:2304.11116 .\nZhang, Y ., Zhang, Y ., Gan, Y ., Yao, L., and Wang, C. (2024). Causal graph discovery with retrieval-\naugmented generation based large language models. arXiv preprint arXiv:2402.15301 .\nZheng, L., Chiang, W.-L., Sheng, Y ., Zhuang, S., Wu, Z., Zhuang, Y ., Lin, Z., Li, Z., Li, D., Xing,\nE., et al. (202","chunk_id":"fc4b27d64f055b7fc30176ba110dd02e","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":512,"entities":[{"name":"H.","type":"PERSON","description":"H. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"LI, Z.","type":"PERSON","description":"Li, Z. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"XU, J.","type":"PERSON","description":"Xu, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"QU, J.","type":"PERSON","description":"Qu, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"ZHOU, J.","type":"PERSON","description":"Zhou, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"ARXIV","type":"PUBLICATION","description":"arXiv is the platform where the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\" was published\narXiv is the platform where the paper \"Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt\" was published\narXiv is the platform where the paper \"Is chatgpt a good nlg evaluator? a preliminary study\" was published\narXiv is the platform where the paper \"Causal graph discovery with retrieval-augmented generation based large language models\" was published\narXiv is the platform where the paper \"Knowledge graph prompting for multi-document question answering\" was published\narXiv is the platform where the paper \"Text summarization with latent queries\" was published","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PUBLICATION"},{"name":"WANG, S.","type":"PERSON","description":"Wang, S. is an author of the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"KHRAMTSOVA, E.","type":"PERSON","description":"Khramtsova, E. is an author of the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"ZHUANG, S.","type":"PERSON","description":"Zhuang, S. is an author of the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"ZUCCON, G.","type":"PERSON","description":"Zuccon, G. is an author of the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"WANG, Y.","type":"PERSON","description":"Wang, Y. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"LIPKA, N.","type":"PERSON","description":"Lipka, N. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"ROSSI, R. A.","type":"PERSON","description":"Rossi, R. A. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"SIU, A.","type":"PERSON","description":"Siu, A. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"ZHANG, R.","type":"PERSON","description":"Zhang, R. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"DERR, T.","type":"PERSON","description":"Derr, T. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"XU, Y.","type":"PERSON","description":"Xu, Y. is an author of the paper \"Text summarization with latent queries\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"LAPATA, M.","type":"PERSON","description":"Lapata, M. is an author of the paper \"Text summarization with latent queries\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"YANG, Z.","type":"PERSON","description":"Yang, Z. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"QI, P.","type":"PERSON","description":"Qi, P. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"ZHANG, S.","type":"PERSON","description":"Zhang, S. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"BENGIO, Y.","type":"PERSON","description":"Bengio, Y. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"COHEN, W. W.","type":"PERSON","description":"Cohen, W. W. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"SALAKHUTDINOV, R.","type":"PERSON","description":"Salakhutdinov, R. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"MANNING, C. D.","type":"PERSON","description":"Manning, C. D. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"EMNLP","type":"CONFERENCE","description":"The conference where the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\" was presented","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"YAO, J.-G.","type":"PERSON","description":"Yao, J.-g. is an author of the paper \"Recent advances in document summarization\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"WAN, X.","type":"PERSON","description":"Wan, X. is an author of the paper \"Recent advances in document summarization\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"XIAO, J.","type":"PERSON","description":"Xiao, J. is an author of the paper \"Recent advances in document summarization\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"KNOWLEDGE AND INFORMATION SYSTEMS","type":"PUBLICATION","description":"The journal where the paper \"Recent advances in document summarization\" was published","source_id":"fc4b27d64f055b7fc30176ba110dd02e"},{"name":"YAO, L.","type":"PERSON","description":"Yao, L. is an author of the paper \"Causal graph discovery with retrieval-augmented generation based large language models\"\nYao, L. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"PENG, J.","type":"PERSON","description":"Peng, J. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"MAO, C.","type":"PERSON","description":"Mao, C. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"LUO, Y.","type":"PERSON","description":"Luo, Y. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"ZHANG, J.","type":"PERSON","description":"Zhang, J. is an author of the paper \"Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"ZHANG, Y.","type":"PERSON","description":"Zhang, Y. is an author of the paper \"Causal graph discovery with retrieval-augmented generation based large language models\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"GAN, Y.","type":"PERSON","description":"Gan, Y. is an author of the paper \"Causal graph discovery with retrieval-augmented generation based large language models\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"WANG, C.","type":"PERSON","description":"Wang, C. is an author of the paper \"Causal graph discovery with retrieval-augmented generation based large language models\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"ZHENG, L.","type":"PERSON","description":"Zheng, L. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"CHIANG, W.-L.","type":"PERSON","description":"Chiang, W.-L. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"SHENG, Y.","type":"PERSON","description":"Sheng, Y. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"WU, Z.","type":"PERSON","description":"Wu, Z. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"ZHUANG, Y.","type":"PERSON","description":"Zhuang, Y. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"LIN, Z.","type":"PERSON","description":"Lin, Z. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"LI, D.","type":"PERSON","description":"Li, D. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"},{"name":"XING, E.","type":"PERSON","description":"Xing, E. is an author of the paper \"Exploring large language models for knowledge graph completion\"","source_id":"fc4b27d64f055b7fc30176ba110dd02e","entity_type":"PERSON"}],"entity_graph":" PERSON<\/data> H. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Li, Z. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Xu, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Qu, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Zhou, J. is an author of the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PUBLICATION<\/data> arXiv is the platform where the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\" was publishedarXiv is the platform where the paper \"Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt\" was publishedarXiv is the platform where the paper \"Is chatgpt a good nlg evaluator? a preliminary study\" was publishedarXiv is the platform where the paper \"Causal graph discovery with retrieval-augmented generation based large language models\" was publishedarXiv is the platform where the paper \"Knowledge graph prompting for multi-document question answering\" was publishedarXiv is the platform where the paper \"Text summarization with latent queries\" was published<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PUBLICATION<\/data> <\/node> PERSON<\/data> Wang, S. is an author of the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Khramtsova, E. is an author of the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Zhuang, S. is an author of the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Zuccon, G. is an author of the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Wang, Y. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Lipka, N. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Rossi, R. A. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Siu, A. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Zhang, R. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Derr, T. is an author of the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Xu, Y. is an author of the paper \"Text summarization with latent queries\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Lapata, M. is an author of the paper \"Text summarization with latent queries\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Yang, Z. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Qi, P. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Zhang, S. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Bengio, Y. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Cohen, W. W. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Salakhutdinov, R. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Manning, C. D. is an author of the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> CONFERENCE<\/data> The conference where the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\" was presented<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Yao, J.-g. is an author of the paper \"Recent advances in document summarization\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Wan, X. is an author of the paper \"Recent advances in document summarization\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Xiao, J. is an author of the paper \"Recent advances in document summarization\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PUBLICATION<\/data> The journal where the paper \"Recent advances in document summarization\" was published<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/node> PERSON<\/data> Yao, L. is an author of the paper \"Causal graph discovery with retrieval-augmented generation based large language models\"Yao, L. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Peng, J. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Mao, C. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Luo, Y. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Zhang, J. is an author of the paper \"Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Zhang, Y. is an author of the paper \"Causal graph discovery with retrieval-augmented generation based large language models\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Gan, Y. is an author of the paper \"Causal graph discovery with retrieval-augmented generation based large language models\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Wang, C. is an author of the paper \"Causal graph discovery with retrieval-augmented generation based large language models\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Zheng, L. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Chiang, W.-L. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Sheng, Y. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Wu, Z. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Zhuang, Y. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Lin, Z. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Li, D. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> PERSON<\/data> Xing, E. is an author of the paper \"Exploring large language models for knowledge graph completion\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> PERSON<\/data> <\/node> 1.0<\/data> H. and Li, Z. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> H. and Xu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> H. and Qu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> H. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Li, Z. and Xu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Li, Z. and Qu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Li, Z. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Xu, J. and Qu, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Xu, J. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Qu, J. and Zhou, J. co-authored the paper \"Is chatgpt a good nlg evaluator? a preliminary study\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wang, S. published the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Khramtsova, E. published the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Zhuang, S. published the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Zuccon, G. published the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wang, Y. published the paper \"Knowledge graph prompting for multi-document question answering\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Lipka, N. published the paper \"Knowledge graph prompting for multi-document question answering\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Rossi, R. A. published the paper \"Knowledge graph prompting for multi-document question answering\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Siu, A. published the paper \"Knowledge graph prompting for multi-document question answering\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Zhang, R. published the paper \"Knowledge graph prompting for multi-document question answering\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Derr, T. published the paper \"Knowledge graph prompting for multi-document question answering\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Xu, Y. published the paper \"Text summarization with latent queries\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Lapata, M. published the paper \"Text summarization with latent queries\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Zhang, J. published the paper \"Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Zhang, Y. published the paper \"Causal graph discovery with retrieval-augmented generation based large language models\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Gan, Y. published the paper \"Causal graph discovery with retrieval-augmented generation based large language models\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Yao, L. published the paper \"Causal graph discovery with retrieval-augmented generation based large language models\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wang, C. published the paper \"Causal graph discovery with retrieval-augmented generation based large language models\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Zheng, L. published the paper \"Exploring large language models for knowledge graph completion\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Chiang, W.-L. published the paper \"Exploring large language models for knowledge graph completion\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Sheng, Y. published the paper \"Exploring large language models for knowledge graph completion\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wu, Z. published the paper \"Exploring large language models for knowledge graph completion\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Zhuang, Y. published the paper \"Exploring large language models for knowledge graph completion\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Lin, Z. published the paper \"Exploring large language models for knowledge graph completion\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Li, D. published the paper \"Exploring large language models for knowledge graph completion\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Xing, E. published the paper \"Exploring large language models for knowledge graph completion\" on arXiv<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wang, S. and Khramtsova, E. co-authored the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wang, S. and Zhuang, S. co-authored the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wang, S. and Zuccon, G. co-authored the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Khramtsova, E. and Zhuang, S. co-authored the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Khramtsova, E. and Zuccon, G. co-authored the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Zhuang, S. and Zuccon, G. co-authored the paper \"Feb4rag: Evaluating federated search in the context of retrieval augmented generation\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wang, Y. and Lipka, N. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wang, Y. and Rossi, R. A. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wang, Y. and Siu, A. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wang, Y. and Zhang, R. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Wang, Y. and Derr, T. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Lipka, N. and Rossi, R. A. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Lipka, N. and Siu, A. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Lipka, N. and Zhang, R. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Lipka, N. and Derr, T. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Rossi, R. A. and Siu, A. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Rossi, R. A. and Zhang, R. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Rossi, R. A. and Derr, T. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Siu, A. and Zhang, R. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Siu, A. and Derr, T. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Zhang, R. and Derr, T. co-authored the paper \"Knowledge graph prompting for multi-document question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Xu, Y. and Lapata, M. co-authored the paper \"Text summarization with latent queries\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Yang, Z. and Qi, P. co-authored the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Yang, Z. and Zhang, S. co-authored the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Yang, Z. and Bengio, Y. co-authored the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Yang, Z. and Cohen, W. W. co-authored the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Yang, Z. and Salakhutdinov, R. co-authored the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> 1.0<\/data> Yang, Z. and Manning, C. D. co-authored the paper \"HotpotQA: A dataset for diverse, explainable multi-hop question answering\"<\/data> fc4b27d64f055b7fc30176ba110dd02e<\/data> <\/edge> <\/graph><\/graphml>"} +{"id":"b1bbda43309e8e0e2175ea034aa88e13","chunk":"Xiv:2402.15301 .\nZheng, L., Chiang, W.-L., Sheng, Y ., Zhuang, S., Wu, Z., Zhuang, Y ., Lin, Z., Li, Z., Li, D., Xing,\nE., et al. (2024). Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in Neural\nInformation Processing Systems , 36.\n15","chunk_id":"b1bbda43309e8e0e2175ea034aa88e13","document_ids":["0668cddc5f873265ba50da5a0a06edad"],"n_tokens":95,"entities":[{"name":"ZHENG, L.","type":"PERSON","description":"Zheng, L. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"CHIANG, W.-L.","type":"PERSON","description":"Chiang, W.-L. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"SHENG, Y.","type":"PERSON","description":"Sheng, Y. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"ZHUANG, S.","type":"PERSON","description":"Zhuang, S. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"WU, Z.","type":"PERSON","description":"Wu, Z. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"ZHUANG, Y.","type":"PERSON","description":"Zhuang, Y. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"LIN, Z.","type":"PERSON","description":"Lin, Z. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"LI, Z.","type":"PERSON","description":"Li, Z. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"LI, D.","type":"PERSON","description":"Li, D. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"XING, E.","type":"PERSON","description":"Xing, E. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS","type":"PUBLICATION","description":"The journal where the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\" was published","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"MT-BENCH","type":"TECHNOLOGY","description":"MT-Bench is a benchmarking tool used in the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"},{"name":"CHATBOT ARENA","type":"TECHNOLOGY","description":"Chatbot Arena is a platform or tool used in the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"","source_id":"b1bbda43309e8e0e2175ea034aa88e13"}],"entity_graph":" PERSON<\/data> Zheng, L. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> PERSON<\/data> Chiang, W.-L. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> PERSON<\/data> Sheng, Y. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> PERSON<\/data> Zhuang, S. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> PERSON<\/data> Wu, Z. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> PERSON<\/data> Zhuang, Y. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> PERSON<\/data> Lin, Z. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> PERSON<\/data> Li, Z. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> PERSON<\/data> Li, D. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> PERSON<\/data> Xing, E. is an author of the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> PUBLICATION<\/data> The journal where the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\" was published<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> TECHNOLOGY<\/data> MT-Bench is a benchmarking tool used in the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> TECHNOLOGY<\/data> Chatbot Arena is a platform or tool used in the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/node> 1.0<\/data> Zheng, L. and Chiang, W.-L. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zheng, L. and Sheng, Y. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zheng, L. and Zhuang, S. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zheng, L. and Wu, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zheng, L. and Zhuang, Y. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zheng, L. and Lin, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zheng, L. and Li, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zheng, L. and Li, D. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zheng, L. and Xing, E. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zheng, L. is an author of the paper that discusses MT-Bench<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zheng, L. is an author of the paper that discusses Chatbot Arena<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Chiang, W.-L. and Sheng, Y. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Chiang, W.-L. and Zhuang, S. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Chiang, W.-L. and Wu, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Chiang, W.-L. and Zhuang, Y. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Chiang, W.-L. and Lin, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Chiang, W.-L. and Li, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Chiang, W.-L. and Li, D. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Chiang, W.-L. and Xing, E. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Chiang, W.-L. is an author of the paper that discusses MT-Bench<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Chiang, W.-L. is an author of the paper that discusses Chatbot Arena<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Sheng, Y. and Zhuang, S. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Sheng, Y. and Wu, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Sheng, Y. and Zhuang, Y. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Sheng, Y. and Lin, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Sheng, Y. and Li, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Sheng, Y. and Li, D. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Sheng, Y. and Xing, E. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Sheng, Y. is an author of the paper that discusses MT-Bench<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Sheng, Y. is an author of the paper that discusses Chatbot Arena<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, S. and Wu, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, S. and Zhuang, Y. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, S. and Lin, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, S. and Li, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, S. and Li, D. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, S. and Xing, E. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, S. is an author of the paper that discusses MT-Bench<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, S. is an author of the paper that discusses Chatbot Arena<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Wu, Z. and Zhuang, Y. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Wu, Z. and Lin, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Wu, Z. and Li, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Wu, Z. and Li, D. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Wu, Z. and Xing, E. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Wu, Z. is an author of the paper that discusses MT-Bench<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Wu, Z. is an author of the paper that discusses Chatbot Arena<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, Y. and Lin, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, Y. and Li, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, Y. and Li, D. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, Y. and Xing, E. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, Y. is an author of the paper that discusses MT-Bench<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Zhuang, Y. is an author of the paper that discusses Chatbot Arena<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Lin, Z. and Li, Z. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Lin, Z. and Li, D. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Lin, Z. and Xing, E. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Lin, Z. is an author of the paper that discusses MT-Bench<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Lin, Z. is an author of the paper that discusses Chatbot Arena<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Li, Z. and Li, D. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Li, Z. and Xing, E. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Li, Z. is an author of the paper that discusses MT-Bench<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Li, Z. is an author of the paper that discusses Chatbot Arena<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Li, D. and Xing, E. co-authored the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Li, D. is an author of the paper that discusses MT-Bench<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Li, D. is an author of the paper that discusses Chatbot Arena<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Xing, E. is an author of the paper that discusses MT-Bench<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> Xing, E. is an author of the paper that discusses Chatbot Arena<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> 1.0<\/data> MT-Bench and Chatbot Arena are both tools used in the paper \"Judging llm-as-a-judge with mt-bench and chatbot arena\"<\/data> b1bbda43309e8e0e2175ea034aa88e13<\/data> <\/edge> <\/graph><\/graphml>"} diff --git a/graphfleet/output/graphindex/artifacts/stats.json b/graphfleet/output/graphindex/artifacts/stats.json new file mode 100644 index 000000000..0972a6b88 --- /dev/null +++ b/graphfleet/output/graphindex/artifacts/stats.json @@ -0,0 +1,160 @@ +{ + "total_runtime": 587.0419588088989, + "num_documents": 1, + "input_load_time": 0, + "workflows": { + "create_base_text_units": { + "overall": 0.27809786796569824, + "0_orderby": 0.004933357238769531, + "1_zip": 0.002577066421508789, + "2_aggregate_override": 0.006456613540649414, + "3_chunk": 0.2312145233154297, + "4_select": 0.004644155502319336, + "5_unroll": 0.006185293197631836, + "6_rename": 0.0027365684509277344, + "7_genid": 0.003219127655029297, + "8_unzip": 0.0026962757110595703, + "9_copy": 0.0020165443420410156, + "10_filter": 0.010167360305786133 + }, + "create_base_extracted_entities": { + "overall": 200.73486065864563, + "0_entity_extract": 200.5583963394165, + "1_snapshot": 0.02570819854736328, + "2_merge_graphs": 0.1340312957763672, + "3_snapshot_rows": 0.012789011001586914 + }, + "create_final_covariates": { + "overall": 0.10694003105163574, + "0_extract_covariates": 0.08164072036743164, + "1_window": 0.0035943984985351562, + "2_genid": 0.0035076141357421875, + "3_convert": 0.008059501647949219, + "4_rename": 0.003099203109741211, + "5_select": 0.003350973129272461 + }, + "create_summarized_entities": { + "overall": 18.377408027648926, + "0_summarize_descriptions": 18.36158299446106, + "1_snapshot_rows": 0.011904716491699219 + }, + "join_text_units_to_covariate_ids": { + "overall": 0.025089025497436523, + "0_select": 0.006565093994140625, + "1_aggregate_override": 0.011329889297485352 + }, + "create_base_entity_graph": { + "overall": 4.001675128936768, + "0_cluster_graph": 0.33349084854125977, + "1_snapshot_rows": 0.019908428192138672, + "2_embed_graph": 3.40474271774292, + "3_snapshot_rows": 0.1873331069946289, + "4_select": 0.05222153663635254 + }, + "create_final_entities": { + "overall": 2.0388026237487793, + "0_unpack_graph": 0.11625313758850098, + "1_rename": 0.008584260940551758, + "2_select": 0.0059413909912109375, + "3_dedupe": 0.0067195892333984375, + "4_rename": 0.0053136348724365234, + "5_filter": 0.031118392944335938, + "6_text_split": 0.019254207611083984, + "7_drop": 0.02828812599182129, + "8_merge": 0.0956735610961914, + "9_text_embed": 1.6695220470428467, + "10_drop": 0.011039972305297852, + "11_filter": 0.02683877944946289 + }, + "create_final_nodes": { + "overall": 10.381420850753784, + "0_layout_graph": 9.700441598892212, + "1_unpack_graph": 0.2411484718322754, + "2_unpack_graph": 0.2792470455169678, + "3_filter": 0.030277252197265625, + "4_drop": 0.007983207702636719, + "5_select": 0.009097099304199219, + "6_snapshot": 0.010268211364746094, + "7_rename": 0.00806736946105957, + "8_convert": 0.03861284255981445, + "9_join": 0.0187835693359375, + "10_rename": 0.010765790939331055 + }, + "create_final_communities": { + "overall": 0.9307990074157715, + "0_unpack_graph": 0.09052109718322754, + "1_unpack_graph": 0.2796623706817627, + "2_aggregate_override": 0.011496305465698242, + "3_join": 0.0422971248626709, + "4_join": 0.0238339900970459, + "5_concat": 0.010031461715698242, + "6_filter": 0.3431220054626465, + "7_aggregate_override": 0.029775381088256836, + "8_join": 0.016811847686767578, + "9_filter": 0.020335912704467773, + "10_fill": 0.008179903030395508, + "11_merge": 0.015482664108276367, + "12_copy": 0.013491392135620117, + "13_select": 0.01628851890563965 + }, + "join_text_units_to_entity_ids": { + "overall": 0.06734323501586914, + "0_select": 0.009002208709716797, + "1_unroll": 0.03510284423828125, + "2_aggregate_override": 0.013802766799926758 + }, + "create_final_relationships": { + "overall": 0.36867332458496094, + "0_unpack_graph": 0.16987895965576172, + "1_filter": 0.04868912696838379, + "2_rename": 0.011057853698730469, + "3_filter": 0.03673553466796875, + "4_drop": 0.013250350952148438, + "5_compute_edge_combined_degree": 0.020687580108642578, + "6_convert": 0.03246474266052246, + "7_convert": 0.01535344123840332 + }, + "join_text_units_to_relationship_ids": { + "overall": 0.0804140567779541, + "0_select": 0.01375269889831543, + "1_unroll": 0.013718605041503906, + "2_aggregate_override": 0.027321338653564453, + "3_select": 0.012871265411376953 + }, + "create_final_community_reports": { + "overall": 338.77023243904114, + "0_prepare_community_reports_nodes": 0.04258394241333008, + "1_prepare_community_reports_edges": 0.02589583396911621, + "2_prepare_community_reports_claims": 0.015239715576171875, + "3_restore_community_hierarchy": 0.020676136016845703, + "4_prepare_community_reports": 0.6444807052612305, + "5_create_community_reports": 337.9709093570709, + "6_window": 0.021309614181518555 + }, + "create_final_text_units": { + "overall": 0.18959522247314453, + "0_select": 0.029093503952026367, + "1_rename": 0.02073383331298828, + "2_join": 0.038782596588134766, + "3_join": 0.02734661102294922, + "4_join": 0.017482757568359375, + "5_aggregate_override": 0.016314029693603516, + "6_select": 0.021203279495239258 + }, + "create_base_documents": { + "overall": 0.2038881778717041, + "0_unroll": 0.022261619567871094, + "1_select": 0.02303004264831543, + "2_rename": 0.012860298156738281, + "3_join": 0.02207493782043457, + "4_aggregate_override": 0.01636528968811035, + "5_join": 0.019821882247924805, + "6_rename": 0.01391744613647461, + "7_convert": 0.02918076515197754 + }, + "create_final_documents": { + "overall": 0.03125262260437012, + "0_rename": 0.01453709602355957 + } + } +} \ No newline at end of file diff --git a/graphfleet/output/graphindex/artifacts/summarized_graph.graphml b/graphfleet/output/graphindex/artifacts/summarized_graph.graphml new file mode 100644 index 000000000..829426858 --- /dev/null +++ b/graphfleet/output/graphindex/artifacts/summarized_graph.graphml @@ -0,0 +1,9715 @@ + + + + + + + + + + + PERSON + Darren Edge is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Ha Trinh is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Newman Cheng is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Joshua Bradley is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Alex Chao is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Apurva Mody is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Steven Truitt is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + PERSON + Jonathan Larson is an author of the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Research is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Strategic Missions and Technologies is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + ORGANIZATION + Microsoft Office of the CTO is an organization where some of the authors of the paper are affiliated + e8d83e6e7a7c0f57b218cef24976b745 + + + METHOD + RAG (Retrieval-Augmented Generation) is a developing research area with multiple established directions, including knowledge graph creation, completion, and extraction of causal graphs. It is a method used for generating responses in text generation tasks by retrieving relevant information from an external knowledge source to enable large language models to answer questions. This approach incorporates the retrieval of relevant data to augment text generation, producing direct responses in various text generation tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY + LLM (Large Language Model) is a type of artificial intelligence model used for a variety of tasks in the field of Natural Language Processing and Information Retrieval. These tasks include generating and assessing text, entity extraction, summarization, understanding relationships in text, and automating human-like sensemaking and reasoning over large collections of documents. LLMs are also employed to generate intermediate answers and scores for text chunks, process these chunks to extract elements of a graph index, and automate the generation of questions for dataset evaluation. Additionally, LLMs can analyze and generate text based on retrieved information and queries, and they possess a context window that can be exceeded by external datasets. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,2c6ed90897310eea2f28e33fff1c32b0,6f33a085ff3304e5994f7fbb86c881a4,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + METHOD + Graph RAG (Retrieval-Augmented Generation) is a sophisticated method that leverages the natural modularity of graphs to partition data for global summarization tasks. This approach integrates multiple stages and concepts, including knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS), to support human sensemaking over text corpora. It is particularly effective in providing comprehensive and structured overviews of public figures across various sectors of the entertainment industry, as well as generating answers for questions in the News article dataset. + +Graph RAG employs a high-level data flow and pipeline for processing and summarizing text, combining both global and local approaches to optimize token usage in text generation tasks. It uses community summaries to improve answer comprehensiveness and diversity while requiring fewer tokens compared to traditional source text summarization methods. This method has been shown to outperform naive RAG in terms of comprehensiveness and diversity in text generation tasks. + +A specific implementation of Graph RAG involves using four levels of graph communities, incorporating concepts from other systems such as self-memory and parallel generation of community answers. This multi-stage mechanism allows for the comparison of multiple conditions, enhancing the overall effectiveness of the summarization process. + +Graph RAG, launched by NebulaGraph, is a retrieval-augmented generation technology based on knowledge graphs. It combines global summarization with graph-based text indexing to answer questions over private text corpora, making it a versatile tool for various text analysis and summarization applications. + 086021a89900a39bcb62036981737bfa,21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,718017a4871c909420f84b85b8ba969d,833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19,fb3c48579608fa28be585ceb6cd2f0fe + + + METHOD + QFS (Query-Focused Summarization) is a method used to generate summaries based on specific user queries + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + Community summaries are generated summaries of data clusters or communities, used to answer queries and tailored to the domain. These summaries are pre-generated for groups of closely-related entities, particularly in the Graph RAG approach, and are derived from community-generated content to compare with source texts. They are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks. Community summaries provide report-like insights into each community within a hierarchical structure, which is useful for understanding the dataset. They are generated from modular communities in the knowledge graph and cover different levels of each graph community hierarchy, including root-level communities in an entity-based graph index. Additionally, these summaries act as a kind of self-memory for generation-augmented retrieval. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + CONCEPT + Global sensemaking questions are questions that require understanding and summarizing large datasets, often beyond the explicit content of the source texts + e8d83e6e7a7c0f57b218cef24976b745 + + + METRIC + 1 million token range refers to the scale of datasets used in the evaluation of the Graph RAG approach + e8d83e6e7a7c0f57b218cef24976b745 + + + TECHNOLOGY + Python is a programming language used for implementing both global and local Graph RAG approaches. Additionally, Python is utilized to implement the open-source version of the Graph RAG approach. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + + + URL + The URL "HTTPS://AKA.MS/GRAPHRAG" is the location where the open-source, Python-based implementation of Graph RAG approaches will be available. This URL serves as the repository for accessing the open-source implementation of the Graph RAG approach. + e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745 + + + METHOD + Query-Focused Summarization (QFS) is a method used to generate summaries that are relevant to specific user queries. This summarization technique focuses on answering specific queries by utilizing the entire corpus of information available. It is designed to provide concise and relevant information based on the specific needs of the user, ensuring that the generated summaries are directly aligned with the queries posed. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + + + CONCEPT + An external knowledge source is a repository of information that can be accessed to retrieve relevant data for answering questions + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + A text corpus is a large collection of written texts used for analysis and research + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + Source documents are the original texts from which information is extracted, retrieved, or summarized. These documents serve as the foundational input texts for various processing tasks, including the Graph RAG (Retrieval-Augmented Generation) approach. In this context, source documents are critical for extracting and analyzing information, ensuring that the data used in computational linguistics and information retrieval tasks is accurate and comprehensive. + bc9e2c9e369c4108cf4f6dd5f60960f4,e8d83e6e7a7c0f57b218cef24976b745,f0306814bf64f5c9e79603fc6a52f4ea + + + CONCEPT + A partial response is an intermediate answer generated from community summaries before being combined into a final response + e8d83e6e7a7c0f57b218cef24976b745 + + + CONCEPT + A final response is the comprehensive answer generated after combining all partial responses + e8d83e6e7a7c0f57b218cef24976b745 + + + METRIC + COMPREHENSIVENESS is a metric used to evaluate the quality of generated responses by measuring how much detail an answer provides to cover all aspects and details of a question. It assesses the completeness and thoroughness of answers, ensuring that they encompass all relevant information. This metric is particularly important in evaluating the summarization approach, focusing on the completeness of the summary. In practical applications, such as evaluating Podcast transcripts and News articles, comprehensiveness has shown win rates between 72-83% and 72-80%, respectively. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + + + METRIC + DIVERSITY is a metric used to evaluate the variety and richness of answers generated in response to a question. It measures how varied and rich an answer is in providing different perspectives and insights. This metric is particularly important in assessing the quality of summarization approaches, focusing on the variety of information included in the summary. DIVERSITY is applied to various types of content, including Podcast transcripts, where win rates range from 75-82%, and News articles, with win rates ranging from 62-71%. It is a crucial target quality for evaluating the effectiveness of different methods in generating diverse and informative responses. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745,ebf5249c888e07fedce6572a4c03f88c + + + ACTIVITY + Human endeavors refer to activities and efforts across various domains that rely on reading and reasoning about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models designed to understand and generate human-like text, playing a crucial role in automating sensemaking in complex domains. Modern language models, such as GPT, Llama, and Gemini, leverage in-context learning to effectively summarize content. These models are integral to the field of Natural Language Processing and Information Retrieval, enabling sophisticated text analysis and generation capabilities. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + + + DOMAIN + Scientific discovery is a complex domain where sensemaking is applied to understand and generate new knowledge from scientific texts + f0306814bf64f5c9e79603fc6a52f4ea + + + DOMAIN + Intelligence analysis is a complex domain where sensemaking is applied to understand and generate insights from intelligence data + f0306814bf64f5c9e79603fc6a52f4ea + + + PROCESS + SENSEMAKING is the process of understanding and making sense of complex information. It involves understanding connections among people, places, and events to anticipate their trajectories and act effectively. This process is crucial for navigating and interpreting intricate data landscapes, enabling individuals and organizations to make informed decisions based on the relationships and patterns identified within the information. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + TEXT CHUNKS are segments of text that are embedded into a vector space for analysis. These segments are extracted from source documents and are used for processing in the Graph RAG (Retrieval-Augmented Generation) approach. By embedding these text chunks into a vector space, they can be analyzed more effectively, facilitating various natural language processing and information retrieval tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + DATA + Element instances are identified and extracted instances of graph nodes and edges from text chunks. They represent individual occurrences of entities, relationships, and claims extracted from source texts. These specific pieces of information are tailored to the domain, providing a structured representation of the underlying data. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Element summaries are concise representations of element instances, tailored to the domain. They are descriptive texts created by the LLM to summarize entities, relationships, and claims extracted from source texts. These summaries provide detailed descriptions of nodes, edges, and covariates within a community, and are used to understand the structure and semantics of the dataset. In essence, element summaries serve as a tool to encapsulate and convey the intricate details of elements within a graph, facilitating a deeper comprehension of the dataset's structural dynamics and semantic relationships. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Graph communities are groups of elements, including nodes, edges, and covariates, detected within a graph index, primarily used for summarization. These communities consist of groups of nodes that exhibit stronger connections to each other than to nodes outside the group. This structural characteristic allows for the identification and analysis of densely connected subgraphs, which can be crucial for understanding the underlying relationships and dynamics within complex networks. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + COMMUNITY ANSWERS are query-focused summaries derived from community summaries. These answers are generated in parallel from the community summaries and are designed to respond to user queries effectively. Essentially, COMMUNITY ANSWERS serve as responses that synthesize information from community summaries to provide concise and relevant answers to specific questions posed by users. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea,f35de4d9fb65f1d5a392064b20545c19 + + + DATA + GLOBAL ANSWER is a comprehensive response generated from multiple community summaries to answer a user query. It is the final query-focused summary produced from all relevant community summaries. The final answer is generated by combining intermediate community answers based on their helpfulness scores. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + TIME + Indexing time refers to the time when the graph index is created and elements are summarized + f0306814bf64f5c9e79603fc6a52f4ea + + + TIME + Query time refers to the time when a query is made and the relevant summaries are generated + f0306814bf64f5c9e79603fc6a52f4ea + + + PROCESS + Graph RAG pipeline is a process using an LLM-derived graph index to detect, extract, and summarize nodes, edges, and covariates in source documents + f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + NODES are entities detected in the graph index of source documents. They represent the individual elements or points in a graph. For instance, in the Podcast dataset, there are 8,564 nodes, while the News dataset contains 15,754 nodes. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + EDGES are relationships detected in the graph index of source documents. They represent the connections or links between nodes in a graph. For instance, in the Podcast dataset, there are 20,691 edges, while the News dataset contains 19,520 edges. These edges are crucial for understanding the structural dynamics and relationships within the datasets, providing insights into how different nodes (such as topics, entities, or documents) are interconnected. + 36db32c37e1987e2c5863898ad882190,f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Covariates are additional attributes associated with extracted node instances in the graph index. They represent claims or additional information detected in the graph index of source documents. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + LEIDEN is a community detection algorithm renowned for its efficiency in recovering hierarchical community structures. It is widely used to partition graphs into modular communities, effectively grouping elements within a graph index. The algorithm's ability to identify and organize these communities makes it a valuable tool in the analysis of complex networks, particularly within the domains of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Retrieval-Augmented Generation (RAG) is an established approach in the field of Natural Language Processing and Information Retrieval, designed to answer user questions over entire datasets. This method involves retrieving relevant text regions to provide grounding for the generation task, thereby enhancing the accuracy and relevance of the generated responses. By combining retrieval and generation processes, RAG effectively synthesizes and presents pertinent information, making it a powerful tool for handling complex queries and large datasets. + f0306814bf64f5c9e79603fc6a52f4ea,fb3c48579608fa28be585ceb6cd2f0fe + + + ORGANIZATION + Microsoft is a technology company whose Chief Technology Officer, Kevin Scott, actively participates in podcast conversations. The organization is deeply involved in automating sensemaking in scientific discovery through the use of large language models (LLMs). Notably, Microsoft conducted a study examining the impact of large language models, specifically GPT-4, on scientific discovery. + 1d07b4248c2655081c7af0e373bd70c9,833e7d67dcd30790b26b71c9b5306f6b,f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Ranade is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Joshi is an author involved in research on automating sensemaking in intelligence analysis using LLMs + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Klein is an author who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Lewis is an author who contributed to the development of the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Traag is an author who contributed to the development of the Leiden community detection method + f0306814bf64f5c9e79603fc6a52f4ea + + + PUBLICATION + arXiv is a preprint repository where several significant papers in the field of Natural Language Processing and Information Retrieval have been published. It serves as a platform for electronic preprints (known as e-prints) that are approved for publication after moderation, but not full peer review. Notable papers published on arXiv include "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models," "Lost in the middle: How language models use long contexts," "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," "Llama 2: Open foundation and fine-tuned chat models," "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy," "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries," "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions," "Enhancing knowledge graph construction using large language models," "Is chatgpt a good nlg evaluator? a preliminary study," "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt," "Causal graph discovery with retrieval-augmented generation based large language models," "Knowledge graph prompting for multi-document question answering," "Text summarization with latent queries," "Retrieval-augmented generation for large language models: A survey," and "Knowledge graph-augmented language models for knowledge-grounded dialogue generation." This repository is a crucial resource for researchers to disseminate their findings rapidly and access the latest advancements in their fields. + 00e8e4e881bd0862022f4dfc913b900b,086021a89900a39bcb62036981737bfa,58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035,f0306814bf64f5c9e79603fc6a52f4ea,fc4b27d64f055b7fc30176ba110dd02e + + + PUBLICATION + Preprint refers to the version of the research paper that is under review and available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + CATEGORY + cs.CL is the category under which the research paper is classified on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + DATE + 24 Apr 2024 is the date when the research paper was submitted to arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + IDENTIFIER + 2404.16130v1 is the identifier for the research paper on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + DATA + Document collections refer to large sets of documents that are analyzed for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + TECHNOLOGY + LLM PROMPTS are specific instructions given to large language models (LLMs) to tailor their responses to the domain of the dataset. These prompts are also used to extract elements from text chunks, ensuring that the LLMs provide relevant and precise information based on the given context. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Community detection is a method used to identify groups of related elements within a graph. It involves the process of identifying communities within a graph, which are clusters of nodes that are more densely connected internally than with the rest of the network. This technique is crucial in understanding the structural dynamics and relationships within complex networks, such as those found in social networks, biological systems, and information retrieval systems. By uncovering these communities, researchers can gain insights into the underlying structure and function of the network, facilitating more effective analysis and interpretation of the data. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + Domain-tailored summarization is a method used to create summaries that are specific to the domain of the dataset + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Klein et al. are authors who defined sensemaking and discussed its importance in understanding connections among people, places, and events + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Ranade and Joshi are authors who discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Lewis et al. are authors who developed the retrieval-augmented generation (RAG) approach + f0306814bf64f5c9e79603fc6a52f4ea + + + PERSON + Traag et al. are the authors who developed the Leiden algorithm, a method renowned for its efficiency in recovering hierarchical community structures. This algorithm is widely recognized in the field of Natural Language Processing and Information Retrieval for its ability to accurately detect and map out complex community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + METHOD + QFS is a task framing that focuses on generating summaries based on specific queries, rather than just concatenating excerpts + fb3c48579608fa28be585ceb6cd2f0fe + + + METHOD + A type of summarization that generates natural language summaries based on specific queries, rather than just extracting text + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A neural network architecture that has shown substantial improvements in various summarization tasks + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A series of large language models known for their ability to perform in-context learning and summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + GEMINI is a family of highly capable multimodal models, as described in an arXiv preprint. These models are known for their ability to perform in-context learning and summarization, making them a significant advancement in the field of Natural Language Processing and Information Retrieval. + 086021a89900a39bcb62036981737bfa,fb3c48579608fa28be585ceb6cd2f0fe + + + TECHNOLOGY + A knowledge graph is a structured representation of information, utilized in the Graph RAG approach for summarization. This structured representation of knowledge is specifically employed in the Graph RAG approach for global summarization, highlighting its role in organizing and integrating information to facilitate comprehensive and coherent summaries. + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + + + REFERENCE + Authors of a paper on Retrieval-augmented generation (RAG) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Author of a paper on query-focused summarization (QFS) + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "BROWN ET AL., 2020" refers to a publication by Brown et al. in 2020, which discusses in-context learning with few-shot examples. The authors of this paper are also known for their work on the GPT series of large language models. + bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + Authors of a paper on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "KURATOV ET AL., 2024" refers to a publication by Kuratov and colleagues in 2024. The study discusses the recall degradation and potential for information loss in longer context windows of Large Language Models (LLMs). The authors explore the limitations of these extended context windows, providing insights into how the performance of LLMs can be affected when dealing with longer sequences of text. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + REFERENCE + "LIU ET AL., 2023" refers to a publication by Liu et al. in 2023, which discusses the recall degradation and potential for information loss in longer context windows of large language models (LLMs). The authors explore the limitations of LLM context windows, highlighting how extended contexts can lead to decreased recall accuracy and information retention. + 4c855404ee3d3c94aa2136f1513c666f,bc9e2c9e369c4108cf4f6dd5f60960f4,fb3c48579608fa28be585ceb6cd2f0fe + REFERENCE + + + TECHNOLOGY + COMMUNITY DETECTION ALGORITHMS are algorithms used to partition a graph into communities of nodes with stronger connections to one another. These algorithms are designed to identify modular communities of closely-related nodes within a graph, thereby revealing the underlying structure and relationships within the network. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + ALGORITHM + Louvain is a community detection algorithm used to partition graphs into modular communities + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + HOTPOTQA is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical resource for evaluating entity extraction prompts, particularly with advanced models like GPT-4-turbo. Additionally, HotPotQA is utilized to observe the behavior of text chunk extraction within the Graph RAG (Retrieval-Augmented Generation) approach, making it a versatile tool in the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,922778ce1cb2fdd6dbab1746c8795620,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNOLOGY + GPT-4-Turbo is a version of the GPT-4 model characterized by its large context size of 128k tokens, which is utilized in various analytical tasks. Specifically, GPT-4-Turbo is employed for entity extraction in evaluations, leveraging its extensive context capacity to enhance the accuracy and comprehensiveness of the analysis. This model is particularly suited for tasks within the Natural Language Processing and Information Retrieval domain, where handling large volumes of text and extracting relevant entities are critical. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + + + DATASET + The "PODCAST TRANSCRIPTS" dataset is a comprehensive collection of compiled transcripts from podcast conversations between Kevin Scott, Microsoft CTO, and other technology leaders. This dataset is used for analysis and consists of 1669 text chunks, each containing 600 tokens with 100-token overlaps between chunks, amounting to approximately 1 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620,ebf5249c888e07fedce6572a4c03f88c + + + DATASET + The "NEWS ARTICLES" dataset is a comprehensive collection of news articles used for analysis. It serves as a benchmark dataset comprising news articles published from September 2013 to December 2023. The dataset spans a range of categories, including entertainment, business, sports, technology, health, and science. It consists of 3197 text chunks, each containing 600 tokens, with a 100-token overlap between chunks, amounting to approximately 1.7 million tokens in total. + 1d07b4248c2655081c7af0e373bd70c9,21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,922778ce1cb2fdd6dbab1746c8795620 + + + METHOD + MAP-REDUCE is a method employed for text summarization by applying a map-reduce approach directly to source texts. It is particularly utilized for query-focused summarization of an entire corpus, enabling efficient processing and extraction of relevant information from large datasets. This technique leverages the map-reduce paradigm to distribute the computational workload, making it suitable for handling extensive text collections in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,973164fa90bf2b4ee267f4fd795916bf + + + METRIC + "EMPOWERMENT" is a concept and metric used in the evaluation of various methods, with an average win rate of 51.3%. It measures how well an answer helps the reader understand and make informed judgments about a topic. Specifically, it evaluates the effectiveness of generated answers in empowering users by developing their understanding of broad issues and themes. Empowerment is a target quality in summarization approaches, focusing on the ability to help users reach an informed understanding. + 21e52bc06a82796b1f4bcd73edda1f2a,322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + Naive RAG is a basic retrieval-augmented generation (RAG) method used as a baseline for comparison in text generation tasks. It converts documents to text, splits them into chunks, and embeds these chunks into a vector space for query matching. While it produces the most direct responses, it is outperformed by global approaches in terms of comprehensiveness and diversity. Naive RAG is also noted for listing public figures mentioned in entertainment articles, focusing on their professional achievements and personal lives. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c + + + METHOD + A method for summarizing source texts using a map-reduce approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Questions generated to evaluate the summarization approach, focusing on understanding activities + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + METRIC + TOKEN COSTS refer to the computational cost measured in tokens used in the summarization process. Specifically, in the context of the Graph RAG (Retrieval-Augmented Generation) approach, token costs denote the number of tokens required for processing text. This metric is crucial for evaluating the efficiency and scalability of text processing methods within the Natural Language Processing and Information Retrieval domain. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS + The high-level process of the Graph RAG approach and pipeline + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + Design parameters are key settings and configurations in the Graph RAG approach. These parameters are crucial as they influence the design of the Graph RAG approach and pipeline, determining the effectiveness and efficiency of the overall system. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + PARAMETER + + + METHOD + GLOBAL SUMMARIZATION is a method for summarizing information on a global scale. It aims to encapsulate the overall structure and semantics of a dataset, providing a comprehensive overview of information from large datasets or corpora. This technique is particularly useful in the field of Natural Language Processing and Information Retrieval, where it helps in distilling vast amounts of data into coherent and concise summaries, facilitating better understanding and analysis of the underlying information. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56,e4d9b12cf2b4c691c74019eefff4fb39 + + + ATTRIBUTE + Modularity is an inherent quality of graphs that allows them to be partitioned into communities of closely-related nodes + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Descriptions generated from modular communities in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + + + INPUT + A specific question or request for information that the summarization methods aim to answer + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + A large collection of texts or documents used for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + Intermediate answers generated from community summaries before being combined into a final global answer + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + The comprehensive answer generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + + + METHOD + A method that focuses on generating questions to understand activities from datasets + 21e52bc06a82796b1f4bcd73edda1f2a + + + INPUT + Brief descriptions of datasets used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + + + DATASET + Datasets that represent real-world information, such as podcast transcripts and news articles + 21e52bc06a82796b1f4bcd73edda1f2a + + + PARAMETER + The level of detail in community summaries used to answer queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + METHOD + A method that summarizes the original source texts directly + 21e52bc06a82796b1f4bcd73edda1f2a + + + OUTPUT + LOW-LEVEL COMMUNITY SUMMARIES are a type of community summary used in the News dataset for analysis. These summaries provide a detailed overview of the source text and are generated from lower hierarchical levels of the community in the knowledge graph. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + OUTPUT + INTERMEDIATE-LEVEL COMMUNITY SUMMARIES are summaries that provide a mid-level overview of the source text. These summaries are generated from intermediate hierarchical levels of the community in the knowledge graph, offering a balanced perspective that captures essential details without overwhelming the reader with excessive information. This approach ensures that the summaries are both informative and concise, making them valuable for understanding the structural dynamics and relationships within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + OUTPUT + Summaries generated from higher hierarchical levels of the community in the knowledge graph + 21e52bc06a82796b1f4bcd73edda1f2a + + + PROCESS, SYSTEM + The entity "PIPELINE" refers to a series of processes or steps used to analyze and summarize a dataset. Specifically, in the context of the Graph RAG approach, the pipeline denotes the sequence of steps and processes involved. This structured sequence is essential for systematically handling data, ensuring that each stage of the analysis is methodically executed to achieve accurate and comprehensive results. + 7fb7d9ce2da9c940a32afdd87d1d9e56,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA STRUCTURE, OUTPUT + The "GRAPH INDEX" is a data structure used in Retrieval-Augmented Generation (RAG) systems to organize and retrieve information. It is a self-generated index that enables Graph RAG by utilizing a graph structure to organize and retrieve data. This index is created from a graph structure and is employed for tasks such as query-focused summarization. The graph index includes various elements extracted from text chunks using Large Language Model (LLM) prompts. Additionally, it supports conditions C0-C3 and is created using generic prompts for entity and relationship extraction. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4,e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + + + DATA, UNIT + Entity references are mentions of entities within text chunks, extracted during the processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC + Recall is a metric used to measure the completeness of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC + Precision is a metric used to measure the accuracy of entity extraction from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + FEW-SHOT EXAMPLES are specialized instances provided to the Large Language Model (LLM) to improve its performance in domains with specialized knowledge such as science, medicine, and law. These examples are tailored to the domain of the data used in the graph indexing process and serve as sample inputs for in-context learning. By tailoring the extraction prompt to the document corpus domain, few-shot examples enhance the LLM's ability to understand and process domain-specific information effectively. + 2c6ed90897310eea2f28e33fff1c32b0,973164fa90bf2b4ee267f4fd795916bf,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA, UNIT + Named entities are specific types of entities such as people, places, and organizations, extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + REFERENCE, PUBLICATION + A reference to a publication by Yang et al. in 2018, introducing the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METHOD, APPROACH + Techniques refer to the specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Implementation details are specific configurations and settings used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS, METHOD + A single extraction round refers to one complete cycle of extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METRIC, ISSUE + Recall degradation refers to the decrease in recall performance when using longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + PROCESS, METHOD + The extraction process involves identifying and extracting elements from text chunks using LLM prompts + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + ATTRIBUTE, CONFIGURATION + Domain refers to the specific area of knowledge or field to which the document corpus belongs + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + DATA, INPUT + Document corpus refers to the collection of documents being processed in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + Default prompt is the standard set of instructions given to the LLM for extracting named entities + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + TECHNIQUE, METHOD + Secondary extraction prompt is an additional set of instructions given to the LLM for extracting covariates + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + METHOD + A covariate prompt is used to extract additional attributes associated with detected entities, including claims linked to the entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + Claims are statements or assertions linked to detected entities, including details such as subject, object, type, description, source text span, and start and end dates + 2c6ed90897310eea2f28e33fff1c32b0 + + + METHOD + Gleanings refer to multiple rounds of entity extraction to ensure that no entities are missed in the process + 2c6ed90897310eea2f28e33fff1c32b0 + + + TECHNIQUE + Logit bias is a technique used to force a yes/no decision from the LLM during the entity extraction process + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + An entity node is a representation of an entity in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A relationship edge is a representation of a relationship between entities in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A claim covariate is an additional attribute or variable associated with a claim in a graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + Communities of entities are groups of closely-related entities detected and summarized by the LLM + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + The "NOISY GRAPH STRUCTURE" refers to a graph structure that may contain inconsistencies or errors, making it challenging to analyze. This type of graph often includes duplicate or inconsistent entity elements due to variations in text format. These inconsistencies can arise from various sources, such as data entry errors, differing data formats, or incomplete information, which complicate the process of extracting meaningful insights and relationships from the graph. + 2c6ed90897310eea2f28e33fff1c32b0,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + DOMAIN + Science is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + DOMAIN + Medicine is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + DOMAIN + Law is a specialized domain that benefits from few-shot examples to improve LLM performance + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Source text span is an attribute of a claim that indicates the specific portion of text from which the claim was extracted + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Start date is an attribute of a claim that indicates when the event or fact described in the claim began + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + End date is an attribute of a claim that indicates when the event or fact described in the claim ended + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Description is an attribute of a claim that provides a detailed explanation of the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Subject is an attribute of a claim that indicates the main entity involved in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + ATTRIBUTE + Object is an attribute of a claim that indicates the entity that is affected by the subject in the claim + 2c6ed90897310eea2f28e33fff1c32b0 + + + CONCEPT + A concept referring to an entity that has multiple name variations but is resilient to such variations due to sufficient connectivity to closely-related entities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + TECHNOLOGY + Large Language Models (LLMs) are advanced AI models capable of understanding and generating human-like text. They are utilized for a variety of tasks, including the creation and completion of knowledge graphs, which are essential for structuring and interlinking information in a meaningful way. Additionally, LLMs serve as evaluators of natural language generation, assessing the quality and coherence of text produced by other AI systems. These models play a crucial role in the field of Natural Language Processing and Information Retrieval, contributing significantly to advancements in how machines comprehend and interact with human language. + 7fb7d9ce2da9c940a32afdd87d1d9e56,92e93fc6449756c0a60200636b297f65,973164fa90bf2b4ee267f4fd795916bf + + + TECHNOLOGY + Structured representations of knowledge in the form of triples (subject, predicate, object) used for reasoning tasks + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + Nodes in a graph that are of the same type and are described using rich descriptive text + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + Edges in a graph that represent relationships between entity nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + METRIC + Weights assigned to edges in a graph, representing the normalized counts of detected relationship instances + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The "HIERARCHICAL COMMUNITY STRUCTURE" is a framework in which communities are organized in a hierarchy, with each level providing a partition of the graph nodes. This structure organizes data into a hierarchy of communities, facilitating a multi-level clustering approach. Hierarchical community structure is utilized to generate community summaries, offering a comprehensive method for understanding the relationships and structural dynamics within specialized communities. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,e4d9b12cf2b4c691c74019eefff4fb39 + + + CONCEPT + A division of graph nodes into mutually-exclusive, collectively-exhaustive communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + TECHNOLOGY + MULTIHOP-RAG is a benchmark dataset comprising news articles published from September 2013 to December 2023, covering a range of categories including entertainment, business, sports, technology, health, and science. It is specifically designed for open-domain question answering, targeting explicit fact retrieval. Additionally, MULTIHOP-RAG represents a specific implementation or variant of Retrieval-Augmented Generation (RAG) used in the context of graph communities. This dataset is also utilized for community detection and analysis, making it a versatile tool in the field of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author who has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + PERSON + Authors who have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The entity "DATASET" refers to a collection of data used for various purposes such as analysis, summarization, and evaluation. This can include diverse types of data like podcast transcripts and news articles. Specifically, the term encompasses datasets used for evaluation purposes, including notable examples like the Podcast and News datasets. + 1d07b4248c2655081c7af0e373bd70c9,7fb7d9ce2da9c940a32afdd87d1d9e56,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + GLOBAL QUERIES refer to questions or inquiries that require comprehensive answers derived from multiple sources or datasets. These queries aim to retrieve information from a global perspective, covering the entire dataset. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + ROOT COMMUNITIES are the top-level clusters in a hierarchical community structure. These communities represent the highest level of organization within the hierarchy, serving as the primary divisions from which more specific sub-communities branch out. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + SUB-COMMUNITIES are lower-level clusters within root communities in a hierarchical community structure, providing more detailed information. These sub-communities play a crucial role in breaking down the larger, more general root communities into more specific and focused groups, thereby facilitating a deeper and more granular understanding of the overall community dynamics. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + TECHNOLOGY + Detailed documents that provide information about specific subtopics within a community + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + The division of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + A system in which elements are ranked or organized in levels + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + CONCEPT + LEVEL 0 represents the root-level communities in the hierarchical clustering with maximum modularity. It serves as the foundational layer in a hierarchical community structure, indicating the initial and most significant division of the dataset into distinct groups. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + LEVEL 1 is a sub-level in a hierarchical community structure, providing more detailed information about the internal organization. Specifically, Level 1 represents sub-communities within the root-level communities, thereby revealing the internal structure and dynamics of these larger groups. This level of granularity helps in understanding the intricate relationships and specialized interactions that occur within the broader community framework. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + A visual representation of graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + METHOD + The Leiden algorithm is a method used for detecting communities in large networks + 843fc5421e086120ffa1c75856ecf6cd + + + TOOL + OpenORD is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + + + TOOL + Force Atlas 2 is a tool used for node layout in graph visualizations + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Nodes represent entities in a graph, with size proportional to their degree + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Edges represent connections between nodes in a graph + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Covariates are variables that are linked to nodes and edges in a graph + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + The LLM context window is the token limit within which summaries are added for processing by a language model + 843fc5421e086120ffa1c75856ecf6cd + + + METHOD + Hierarchical clustering is a method of clustering data into a tree-like structure with multiple levels + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + The token limit is the maximum number of tokens that can be processed in a single context window by a language model + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Summary detail refers to the level of detail provided in a summary + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Scope refers to the range or extent of information covered in a summary + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + A "USER QUERY" is a question or inquiry posed by a user seeking information, which the system aims to answer. + 1d07b4248c2655081c7af0e373bd70c9,843fc5421e086120ffa1c75856ecf6cd + CONCEPT + + + ELEMENT + Chunks are segments of community summaries divided into pre-specified token sizes + 843fc5421e086120ffa1c75856ecf6cd + ELEMENT + + + METRIC + Prominence is a metric used to prioritize community edges based on the combined degree of source and target nodes + 843fc5421e086120ffa1c75856ecf6cd + + + METRIC + Combined source and target node degree is a metric used to measure the overall prominence of community edges + 843fc5421e086120ffa1c75856ecf6cd + + + ELEMENT + Community edges are connections between nodes within a community, prioritized based on prominence + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Sub-community summaries are shorter summaries of sub-communities used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + + + CONCEPT + Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + + + CATEGORY + Community level refers to the different levels in the hierarchical community structure used to generate summaries + 843fc5421e086120ffa1c75856ecf6cd + + + DATA + Chunks are segments of community summaries divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + + + METRIC + A score between 0-100 generated by the LLM to indicate how helpful an answer is in addressing the target question + 1d07b4248c2655081c7af0e373bd70c9 + + + USER + A user looking for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + + + USER + A user incorporating current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic discussing collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + A topic addressing the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + Insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + + + TOPIC + The importance of health literacy highlighted through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + + + OUTPUT + Answers generated for each chunk of community summaries + 1d07b4248c2655081c7af0e373bd70c9 + + + METRIC + The pre-specified size of tokens used to divide community summaries into chunks + 1d07b4248c2655081c7af0e373bd70c9 + + + TECHNOLOGY + The "CONTEXT WINDOW" refers to a window of text used to generate answers, constrained by token size. The size of the context window is consistent across all conditions, ensuring uniformity in answer generation processes. + 1d07b4248c2655081c7af0e373bd70c9,973164fa90bf2b4ee267f4fd795916bf + + + PERSON + Kevin Scott is the Chief Technology Officer (CTO) of Microsoft and actively participates in podcast conversations. His involvement in these discussions is documented and compiled in the dataset, highlighting his contributions to the field of technology and his role in shaping Microsoft's strategic direction. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + Individuals who are leaders in the technology industry and participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + + + INPUT + A specific activity or goal that the user aims to achieve using the datasets + 1d07b4248c2655081c7af0e373bd70c9 + + + INPUT + QUESTIONS refer to specific inquiries generated by the Large Language Model (LLM) based on the user's task and the target datasets. These questions are utilized in the analysis to evaluate the performance of different methods within the domain of Natural Language Processing and Information Retrieval. The generation and subsequent use of these questions are crucial for assessing the effectiveness and accuracy of various computational techniques and models. + 1d07b4248c2655081c7af0e373bd70c9,4c855404ee3d3c94aa2136f1513c666f + + + + + 1d07b4248c2655081c7af0e373bd70c9 + + + DATASET + MT-BENCH is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical tool in evaluating the performance of models in retrieving accurate and relevant information from a broad range of topics. MT-Bench is prominently featured in the academic paper titled "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," where it is utilized to assess the capabilities of large language models in a structured and rigorous manner. This dataset plays a significant role in advancing the field of Natural Language Processing and Information Retrieval by providing a standardized metric for model comparison and evaluation. + 922778ce1cb2fdd6dbab1746c8795620,b1bbda43309e8e0e2175ea034aa88e13 + DATASET + + + PROCESS + The process through which people inspect, engage with, and contextualize data within the broader scope of real-world activities + 922778ce1cb2fdd6dbab1746c8795620 + PROCESS + + + TECHNOLOGY + Retrieval-Augmented Generation systems used for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + TECHNOLOGY + + + AUTHORS + Authors of a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors of a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + AUTHORS + Authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + AUTHORS + Authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + AUTHORS + + + + + 922778ce1cb2fdd6dbab1746c8795620 + + + PODCAST + "BEHIND THE TECH" is a podcast series featuring conversations between Kevin Scott and other technology leaders. It serves as a media platform associated with Kevin Scott, providing insights and discussions on various technological advancements and industry trends. + 833e7d67dcd30790b26b71c9b5306f6b,922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + Kevin Scott, Microsoft CTO, who participates in the podcast conversations compiled in the dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + PERSON + An author associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + DATASET + A benchmark dataset for open-domain question answering, targeting explicit fact retrieval + 922778ce1cb2fdd6dbab1746c8795620 + + + METRIC + N represents the number of test questions per dataset used in the evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + METHOD + A method applying a map-reduce approach directly to source texts for summarization + 973164fa90bf2b4ee267f4fd795916bf + + + METHOD + A na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached + 973164fa90bf2b4ee267f4fd795916bf + + + CATEGORY + C0 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a root-level community summary, which is utilized to answer user queries by providing the fewest number of summaries. This category is essential for understanding the structural dynamics within the community, particularly in the domain of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C1 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a high-level community summary used to answer user queries, effectively representing sub-communities of C0. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C2 is a category or condition used in the analysis, representing a specific subset of the data. It functions as an intermediate-level community summary used to answer user queries, representing sub-communities of C1. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CATEGORY + C3 is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a category or condition that encapsulates low-level community summaries, which are instrumental in answering user queries. These summaries represent sub-communities of C2, providing detailed insights into the structural dynamics and relationships within the broader community. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + TS, or "Text Summarization," is a category or condition used in the analysis, representing a specific subset of the data. It is particularly focused on source text summarization within the analysis. TS employs a text summarization method that applies a map-reduce approach directly to source texts, facilitating efficient and scalable summarization processes. This category is integral to understanding and processing large volumes of text data, making it a crucial component in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + "SS" is a category or cluster used in the analysis, representing a specific subset of the data. It serves as a baseline condition and is associated with a na¨ıve RAG (Retrieval-Augmented Generation) approach. In this context, text chunks are retrieved and added to the context window until the token limit is reached. + 4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf,ebf5249c888e07fedce6572a4c03f88c + + + CONCEPT + The prompts used for answer generation, which are the same across all conditions with minor modifications + 973164fa90bf2b4ee267f4fd795916bf + + + DATASET + The "PODCAST DATASET" is a collection of podcast transcripts utilized for both analysis and evaluation purposes. This dataset is specifically designed to support various analytical tasks, providing a rich source of textual data for researchers and practitioners in the field of Natural Language Processing and Information Retrieval. The transcripts within the dataset offer valuable insights and serve as a critical resource for evaluating different computational models and techniques. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + DATASET + The "NEWS DATASET" is a collection of news articles utilized for both analysis and evaluation purposes. This dataset serves as a valuable resource for examining and assessing various aspects of news content, making it an essential tool in the field of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + METRICS in the context of Natural Language Processing and Information Retrieval are essential tools used to evaluate the performance of natural language generation. These metrics include both reference-based metrics, which compare generated texts to a set of reference texts, and qualities of the generated texts themselves. They are crucial in the analysis to assess the effectiveness of different methods in generating natural language, ensuring that the outputs are both accurate and of high quality. + 4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + "WANG ET AL., 2023A" refers to a study conducted by Wang and colleagues in 2023, which highlights the effectiveness of Large Language Models (LLMs) in evaluation. This study is a significant contribution to the field, providing insights into the capabilities and performance of LLMs in various evaluative tasks. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + "ZHENG ET AL., 2024" refers to a study conducted by Zheng and colleagues in 2024. This study highlights the effectiveness of Large Language Models (LLMs) in evaluation processes. The research, authored by Zheng et al., provides significant insights into the capabilities and applications of LLMs within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + REFERENCE + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + The entity "CONDITIONS" refers to the different scenarios or variables that are compared in an experiment. Specifically, in the context of the analysis, these conditions include Graph RAG, text summarization, and semantic search RAG. These conditions are used to evaluate and compare various aspects of performance and effectiveness within the domain of Natural Language Processing and Information Retrieval. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + USER QUERIES refer to the inquiries made by users to retrieve information. These queries are answered using different methods and conditions, depending on the context and the specific requirements of the information retrieval process. + 973164fa90bf2b4ee267f4fd795916bf,e4d9b12cf2b4c691c74019eefff4fb39 + + + CONCEPT + Types of entities extracted during the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + + + METRIC + The "CONTEXT WINDOW SIZE" refers to the fixed size of the context window used in various stages of natural language processing and information retrieval tasks. For the final evaluation, the context window size is set to 8k tokens. During the analysis phase, different context window sizes are tested, including 8k, 16k, 32k, and 64k tokens. Additionally, in the graph indexing process, the context window size is set to 600 tokens. This variability in context window sizes highlights the importance of adapting the window size to the specific requirements of different tasks within the domain. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,973164fa90bf2b4ee267f4fd795916bf + + + CONCEPT + The process of extracting information, with 1 gleaning for the Podcast dataset and 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + TECHNOLOGY + Natural Language Generation (NLG) is a subfield of artificial intelligence that focuses on generating human-like text from data + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method where a Large Language Model (LLM) is used to compare and evaluate competing outputs + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method for evaluating the performance of Retrieval-Augmented Generation (RAG) systems, focusing on context relevance, faithfulness, and answer relevance + 322e02986c8724eedbcf3ebfa20b989c + + + PUBLICATION + A reference to a study or paper authored by Es and others in 2023 + 322e02986c8724eedbcf3ebfa20b989c + + + TOOL + A Large Language Model (LLM) used to evaluate and compare generated texts based on specific metrics + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + DIRECTNESS is a metric that measures how specifically and clearly an answer addresses a question. It is used to evaluate the straightforwardness of the generated answers. Additionally, it serves as a validity test metric to measure the directness of responses, with naive RAG (Retrieval-Augmented Generation) producing the most direct responses. + 322e02986c8724eedbcf3ebfa20b989c,36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + DATA + An example of LLM-generated assessment shown in a table format + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The entity "QUESTION" refers to a specific query used in the evaluation process, particularly as a metric to evaluate the generated responses by asking specific questions. This approach is commonly employed in the domain of Natural Language Processing and Information Retrieval to assess the quality and relevance of responses generated by various models or systems. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + ENTITY + Public figures are individuals who have gained fame or notoriety in various sectors such as entertainment, sports, and digital media. These individuals are well-known in the entertainment industry and are frequently mentioned across various articles. Their prominence in public discourse spans multiple domains, reflecting their influence and recognition in society. + 322e02986c8724eedbcf3ebfa20b989c,718017a4871c909420f84b85b8ba969d + + + DATASET + ENTERTAINMENT ARTICLES is a collection of articles focused on the entertainment industry. This dataset consists of articles related to various aspects of the entertainment sector, providing a comprehensive resource for understanding trends, developments, and key topics within this field. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + DOMAIN + The **ENTERTAINMENT INDUSTRY** is a multifaceted sector that encompasses various forms of entertainment, including movies, music, television, sports, and digital media. This industry is characterized by its diverse range of content and mediums, which collectively contribute to its broad appeal and significant cultural impact. The entertainment industry plays a crucial role in shaping public opinion, trends, and cultural norms through its extensive reach and influence across different platforms and genres. + 322e02986c8724eedbcf3ebfa20b989c,e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric indicating the highest level of development or achievement in a particular field + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric indicating results that are comparable to or better than those of others in the same field + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric based on evaluations made by humans + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + Metrics that require a gold standard or reference answers for evaluation + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + An evaluation method that does not require reference answers + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how relevant the generated text is to the given context + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how accurately the generated text reflects the source information + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures how relevant the generated answer is to the question + 322e02986c8724eedbcf3ebfa20b989c + + + METHOD + A method involving multiple stages or steps + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The correct or ideal answers used as a benchmark in evaluations + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + "SENSEMAKING QUESTIONS" are a class of questions used to evaluate the performance of Retrieval-Augmented Generation (RAG) systems. These questions are specifically designed to help users understand and make sense of complex information, as well as to validate the understanding and interpretation of data. By employing sensemaking questions, researchers and practitioners can assess how effectively a RAG system can retrieve and generate relevant information, thereby ensuring that the system aids in the comprehension and accurate interpretation of intricate datasets. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method where two items are directly compared against each other + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + TARGET METRICS are specific measures used to evaluate the performance of RAG systems. These metrics are aimed to be achieved or measured in the analysis and are the focus of an evaluation. + 322e02986c8724eedbcf3ebfa20b989c,92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A metric used as a baseline or standard for comparison + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures the accuracy and reliability of a method or result + 322e02986c8724eedbcf3ebfa20b989c + + + METRIC + A metric that measures the randomness or variability in a process + 322e02986c8724eedbcf3ebfa20b989c + + + DATA + The average scores obtained from multiple evaluations + 322e02986c8724eedbcf3ebfa20b989c + + + PERSON + Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Britney Spears is a public figure frequently mentioned in entertainment articles, known for her significant contributions to the music industry and her high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + PERSON + Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his significant contributions to the music industry and his high-profile personal life. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in film and television + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in music + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in sports + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry, including those involved in digital media and business + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category of public figures in the entertainment industry who are involved in controversies + e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric used to determine the winner in the comparison of generated responses + e8c8f911135faf3ff35f24107eb3f99c + + + METRIC + A metric used to evaluate the quality of LLM-generated responses + e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "FILM" refers to a sector within the entertainment industry that encompasses movies and cinema. This sector includes public figures involved in the movie industry, such as actors, directors, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "TELEVISION" refers to a sector within the entertainment industry that encompasses TV shows and series. This sector includes public figures involved in TV shows, such as actors, hosts, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + MUSIC is a sector within the entertainment industry that encompasses musical performances and recordings. This sector includes public figures involved in the music industry, such as singers, musicians, and producers. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + The entity "SPORTS" refers to a sector within the entertainment industry that encompasses athletic events and competitions. This sector includes public figures involved in sports, such as athletes, coaches, and sports commentators. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + SECTOR + DIGITAL MEDIA is a sector within the entertainment industry that encompasses online content and social media. This sector includes public figures involved in online platforms, such as influencers, content creators, and digital marketers. These individuals play a significant role in shaping digital landscapes through their engagement with audiences and their ability to leverage various online tools and platforms for content dissemination and marketing purposes. + 718017a4871c909420f84b85b8ba969d,e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes stories and themes that shape culture + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes popular movements and styles + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes public conversations and debates + e8c8f911135faf3ff35f24107eb3f99c + + + CATEGORY + A category within the entertainment industry that includes formal discussions and communications + e8c8f911135faf3ff35f24107eb3f99c + + + RESPONSE + Answer 1 provides a varied and rich response by covering a wide range of public figures from different sectors of the entertainment industry, including film, television, music, sports, gaming, and digital media. It offers insights into the contributions and influence of these figures, as well as controversies and their impact on public discourse. The answer also cites specific data sources for each mentioned figure, indicating a diverse range of evidence to support the claims. + 718017a4871c909420f84b85b8ba969d + + + RESPONSE + "ANSWER 2" is a generated answer for the example question in the News article dataset. It focuses on a smaller group of public figures, primarily from the music industry and sports, and relies heavily on a single source for data. "ANSWER 2" provides concise explanations for the frequent mentions of specific public figures such as Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + METHOD + Naïve RAG is a baseline method for retrieval-augmented generation (RAG) that does not use advanced techniques. It is a basic form of RAG with certain drawbacks that advanced RAG systems aim to overcome. Naïve RAG is used to generate answers for questions in the News article dataset and to generate responses that directly list specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d,e4d9b12cf2b4c691c74019eefff4fb39,ebf5249c888e07fedce6572a4c03f88c,f35de4d9fb65f1d5a392064b20545c19 + + + DATASET + The "NEWS ARTICLE DATASET" is a collection of news articles utilized for various analytical purposes. This dataset is specifically employed for generating responses to questions about public figures in the entertainment industry, making it a valuable resource for both analysis and information retrieval tasks within this domain. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + TOPIC + Controversies are events or issues involving public figures that generate public debate and impact public discourse. + 718017a4871c909420f84b85b8ba969d + + + SECTOR + The gaming sector includes public figures involved in the gaming industry, including gamers, developers, and streamers. + 718017a4871c909420f84b85b8ba969d + + + RESOURCE + Data sources are references or reports used to support claims about public figures and their influence. + 718017a4871c909420f84b85b8ba969d + + + METHOD + Assessments generated by large language models (LLMs) to evaluate the answers produced by different methods + ebf5249c888e07fedce6572a4c03f88c + + + DATASET + An example question used in the News article dataset for analysis + ebf5249c888e07fedce6572a4c03f88c + + + DATA + The datasets used in the analysis, consisting of various text sources + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + A metric used to compare the performance of different conditions in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + CATEGORY + A specific setup or scenario used in the analysis, such as C0, C1, C2, C3, TS, and SS + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + WIN RATE is a measure used to evaluate the success rate of different methods in providing comprehensive and diverse answers. It represents the percentage of times a particular approach or method achieves a win in a given context. Specifically, it quantifies the percentage of times a condition outperformed another in the analysis. This metric is crucial in assessing the effectiveness of various strategies within the domain of Natural Language Processing and Information Retrieval, offering insights into the comparative performance of different techniques. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f,6f33a085ff3304e5994f7fbb86c881a4 + + + METRIC + The condition that performed the best across all comparisons in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + METRIC + The expected win rate of a condition when compared to itself, shown as 50% for reference + 4c855404ee3d3c94aa2136f1513c666f + + + METHOD + The use of large language models (LLMs) at the time of querying, evaluated in the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + METHOD + The "FINAL EVALUATION" is the last stage of the analysis where the best performing context window size was used. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + + + PROCESS + The process that resulted in the creation of graphs for the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + + + STRUCTURE + A data structure consisting of nodes and edges, used to represent the Podcast and News datasets + 36db32c37e1987e2c5863898ad882190 + + + METHOD + Approaches that consistently outperformed the naive RAG in comprehensiveness and diversity metrics + 36db32c37e1987e2c5863898ad882190 + + + METRIC + The number of context units, such as community summaries or text chunks, used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + + + METRIC + The term "TOKENS" refers to the number of individual words used in the analysis. The evaluation typically focuses on corpora in the region of 1 million tokens. This metric is crucial for understanding the scope and scale of the text data being analyzed, particularly in the fields of Natural Language Processing and Information Retrieval. + 36db32c37e1987e2c5863898ad882190,92e93fc6449756c0a60200636b297f65 + METRIC + + + METRIC + The percentage of the maximum token count used in the analysis + 36db32c37e1987e2c5863898ad882190 + METRIC + + + METHOD + MAP-REDUCE SUMMARIZATION is a method for summarizing source texts using a map-reduce approach. This summarization technique is notably resource-intensive, necessitating the highest number of context tokens compared to other methods. The map-reduce framework, originally popularized for its efficiency in processing large-scale data, is adapted here to handle the complexities of text summarization, ensuring comprehensive and accurate extraction of key information from extensive source texts. + 36db32c37e1987e2c5863898ad882190,e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Summaries at the root level of the community hierarchy, requiring dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + + + DATASET + SOURCE TEXTS are the original texts from which summaries or analyses are derived. These texts serve as the foundational material used for comparison with community summaries in the analysis. + 6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39 + + + REFERENCE + A reference to a paper by Ram et al. in 2023 discussing RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + REFERENCE + "GAO ET AL., 2023" is a paper published in 2023 by Gao et al. that delves into advanced Retrieval-Augmented Generation (RAG) techniques, specifically where the index is a knowledge graph. The publication also touches upon naive RAG approaches, providing a comprehensive examination of both advanced and basic methodologies within the domain of Natural Language Processing and Information Retrieval. + 6f33a085ff3304e5994f7fbb86c881a4,92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + + + CATEGORY + Intermediate-level summaries are a type of community summary used in the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + CATEGORY + Root-level summaries are a type of community summary used in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + METRIC + Answer comprehensiveness is a measure used to evaluate the completeness of answers provided by different methods + 6f33a085ff3304e5994f7fbb86c881a4 + + + TECHNOLOGY + Ad-hoc LLM use refers to the spontaneous use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + + + TECHNOLOGY + Element extraction prompts are used to extract specific details in the Graph RAG index + 6f33a085ff3304e5994f7fbb86c881a4 + + + CONCEPT, TECHNOLOGY + A mathematical space in which text chunks and queries are embedded to represent similar semantics + f35de4d9fb65f1d5a392064b20545c19 + + + CONCEPT, DATA + Search inputs that are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A type of RAG system that includes patterns for iterative and dynamic cycles of interleaved retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, CONCEPT + A concept related to generation-augmented retrieval that facilitates future generation cycles + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method that facilitates future generation cycles by using self-memory + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A strategy for iterative retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A federated strategy for retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method that combines multiple concepts for summarizing multiple documents + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A method for answering questions that require multiple steps or "hops" to gather information + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + An approach that involves generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A method for generating a hierarchical index of text chunks by clustering the vectors of text embeddings + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A method for answering multiple interpretations of ambiguous questions by generating a hierarchical structure + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A process that involves using LLMs to create knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A process that involves using LLMs to complete existing knowledge graphs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + Graphs that represent causal relationships, which can be extracted using LLMs + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + REFERENCE, PUBLICATION + A reference to a publication by Cheng et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Mao et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Shao et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Wang et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Su et al. in 2020 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Feng et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Trivedi et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Khattab et al. in 2022 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Sarthi et al. in 2024 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Kim et al. in 2023 + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + "TRAJANOSKA ET AL., 2023" refers to a paper by Trajanoska et al. published in 2023, which focuses on using Large Language Models (LLMs) for knowledge graph creation. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting innovative methodologies for leveraging advanced language models to construct and enhance knowledge graphs. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + "Yao et al., 2023" refers to a paper published by Yao and colleagues in 2023. The study focuses on the application of large language models (LLMs) for the task of knowledge graph completion. This publication is a significant contribution to the field of Natural Language Processing and Information Retrieval, highlighting the potential of advanced LLMs to enhance the accuracy and efficiency of knowledge graph completion processes. + 92e93fc6449756c0a60200636b297f65,f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + REFERENCE, PUBLICATION + A reference to a publication by Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + REFERENCE, PUBLICATION + + + TECHNOLOGY, METHOD + A system that combines multiple concepts for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + A system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + TECHNOLOGY, METHOD + + + TECHNOLOGY, METHOD + Strategies used before the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Strategies used during the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Strategies used after the retrieval process in advanced RAG systems + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + A pattern in Modular RAG systems for iterative and dynamic cycles of retrieval and generation + f35de4d9fb65f1d5a392064b20545c19 + + + TECHNOLOGY, METHOD + Cycles of generation that are facilitated by self-memory in Graph RAG + f35de4d9fb65f1d5a392064b20545c19 + + + PUBLICATION + A paper by Ban et al. published in 2023, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + PUBLICATION + A paper by Zhang et al. published in 2024, focusing on the extraction of causal graphs from source texts + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where the index is a knowledge graph, developed by Baek et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Baek et al. published in 2023, focusing on the KAPING method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where subsets of the graph structure are the objects of enquiry, developed by He et al. in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by He et al. published in 2024, focusing on the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where derived graph metrics are the objects of enquiry, developed by Zhang in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Zhang published in 2023, focusing on the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, developed by Kang et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Kang et al. published in 2023, focusing on the SURGE method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + METHOD + A method where retrieved event-plot subgraphs are serialized using narrative templates, developed by Ranade and Joshi in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + A paper by Ranade and Joshi published in 2023, focusing on the FABULA method + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + PUBLICATION + A paper by Wang et al. published in 2023, focusing on a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering + 92e93fc6449756c0a60200636b297f65 + PUBLICATION + + + ORGANIZATION + LangChain is an organization that developed Langchain graphs and supports a variety of graph databases. + 71f6daf11e64e5273a3847d46bf228e1,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + + + ORGANIZATION + LlamaIndex is an organization that developed the LlamaIndex Knowledge Graph Index and supports a variety of graph databases. + 6cd82819982879bd164547d2773ba5c7,92e93fc6449756c0a60200636b297f65 + ORGANIZATION + + + TECHNOLOGY + Neo4J is both a graph database format supported by various Retrieval-Augmented Generation (RAG) applications and an organization that developed Project NaLLM. The graph database format of Neo4J is widely recognized for its efficiency in handling complex relationships and structures, making it a valuable tool in the field of Natural Language Processing and Information Retrieval. As an organization, Neo4J has contributed significantly to the advancement of these domains through innovative projects like NaLLM, which further underscores its pivotal role in the community. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + + + METHOD + A method that can create and reason over knowledge graphs in Neo4J format, developed by Neo4J in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + TECHNOLOGY + NebulaGraph is both a graph database format and an organization that has made significant contributions to the field of graph databases and retrieval-augmented generation (RAG) applications. As a graph database format, NebulaGraph is supported by various RAG applications, facilitating the efficient handling and querying of complex graph data structures. Additionally, NebulaGraph, as an organization, has pioneered the industry-first graph RAG, which integrates retrieval-augmented generation with large language models (LLMs) based on knowledge graphs. This innovation underscores NebulaGraph's role in advancing the capabilities of knowledge graph-based applications and enhancing the performance of LLMs in generating contextually relevant information. + 833e7d67dcd30790b26b71c9b5306f6b,92e93fc6449756c0a60200636b297f65 + TECHNOLOGY + + + METHOD + A method that can create and reason over knowledge graphs in NebulaGraph format, developed by NebulaGraph in 2024 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + METHOD + A method for comparing fabrication rates, developed by Manakul et al. in 2023 + 92e93fc6449756c0a60200636b297f65 + METHOD + + + PUBLICATION + "MANAKUL ET AL., 2023" refers to a paper by Manakul et al. published in 2023, which focuses on the SelfCheckGPT method. This work by Manakul and colleagues is centered around the development and application of SelfCheckGPT, a technique likely aimed at enhancing the performance and reliability of GPT models. The paper contributes to the field of Natural Language Processing and Information Retrieval by addressing specific challenges and proposing innovative solutions through the SelfCheckGPT method. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + PUBLICATION + + + STAKEHOLDER + END USERS are individuals who are the final users of the system or analysis. They play a crucial role in validating sensemaking questions and target metrics, ensuring that the system or analysis meets the intended objectives and provides meaningful insights. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + STAKEHOLDER + + + CONCEPT + Considerations and compromises involved in building a graph index + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METRIC + The effectiveness of RAG systems, which varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + METRIC + + + CONCEPT + Various forms of data used in RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METRIC + The scale of datasets used in RAG systems, which affects performance + 92e93fc6449756c0a60200636b297f65 + METRIC + + + PROCESS + The process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + PROCESS + + + DATASET + Collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + DATASET + + + CONCEPT + Different categories of questions used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + CONCEPT + + + METHOD + SelfCheckGPT is an approach used to compare fabrication rates in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method for global summarization of source texts that does not use a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + RESOURCE + The amount of computational resources allocated for a task + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The expected number of queries over the lifetime of a dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Annotations that provide detailed information about the text + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method that uses embeddings to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + RAG schemes that combine embedding-based matching with other approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Mechanisms used in map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A hierarchical organization of communities + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A method that combines knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS) for global text summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The cost associated with the number of tokens used in a text generation task + e4d9b12cf2b4c691c74019eefff4fb39 + + + TECHNOLOGY + An implementation of Graph RAG approaches using the Python programming language + e4d9b12cf2b4c691c74019eefff4fb39 + + + PERSON + A person who contributed to the work mentioned in the acknowledgements + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The rates at which fabrications occur in text generation tasks + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The expected number of queries over the lifetime of a specific dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + METRIC + The benefits or value obtained from using a graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Different methods related to retrieval-augmented generation that utilize graph structures + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + Graph RAG approaches that operate in a more localized manner + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Annotations made on the graph to provide additional information + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + Reports generated from community summaries + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + An operation that aggregates information across multiple levels of a hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + METHOD + A mechanism that allows for exploring detailed information by following higher-level summaries + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + The trail of information that guides users to more detailed data + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + The top-level communities in a hierarchical structure + e4d9b12cf2b4c691c74019eefff4fb39 + + + DATA + A graph index organized around entities + e4d9b12cf2b4c691c74019eefff4fb39 + + + TECHNOLOGY + A publicly available implementation of a technology + e4d9b12cf2b4c691c74019eefff4fb39 + + + PERSON + Alonso Guevara Fernández is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Amber Hoak is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Andrés Morales Esquivel is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Ben Cutler is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Billie Rinaldi is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Chris Sanchez is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Chris Trevino is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Christine Caggiano is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + David Tittsworth is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Dayenne de Souza is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Douglas Orbaker is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Ed Clark is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Gabriel Nieves-Ponce is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Gaudy Blanco Meneses is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Kate Lytvynets is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Katy Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Mónica Carvajal is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Nathan Evans is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Richard Ortega is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Rodrigo Racanicci is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Sarah Smith is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PERSON + Shane Solomon is a contributor to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + A technical report on GPT-4 published as an arXiv preprint + 086021a89900a39bcb62036981737bfa + + + METHOD + A method for zero-shot knowledge graph question answering described in an arXiv preprint + 086021a89900a39bcb62036981737bfa + + + METHOD + A method for harnessing large language models for advanced causal discovery from data + 086021a89900a39bcb62036981737bfa + + + METHOD + A method incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Achiam is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Adler is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Agarwal is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + L. Ahmad is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + I. Akkaya is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + F. L. Aleman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + D. Almeida is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Altenschmidt is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Altman is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Anadkat is an author of the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PERSON + R. Anil is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + S. Borgeaud is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + Y. Wu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J.-B. Alayrac is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Yu is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + R. Soricut is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Schalkwyk is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + A. M. Dai is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + A. Hauth is an author of the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PERSON + J. Baek is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + A. F. Aji is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + A. Saffari is an author of the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PERSON + T. Ban is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + L. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + X. Wang is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + H. Chen is an author of the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + PERSON + T. Baumel is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + M. Eyal is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + M. Elhadad is an author of the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the Gemini paper + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + PUBLICATION + The arXiv preprint identifier for the paper on query focused abstractive summarization + 086021a89900a39bcb62036981737bfa + + + PERSON + Baumel, T. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Eyal, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Elhadad, M. is an author of the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Blondel, V. D. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Guillaume, J.-L. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Lambiotte, R. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Lefebvre, E. is an author of the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The journal where the paper "Fast unfolding of communities in large networks" was published + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Brown, T. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Mann, B. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Ryder, N. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Subbiah, M. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Kaplan, J. D. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dhariwal, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Neelakantan, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Shyam, P. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Sastry, G. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Askell, A. is an author of the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + "ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS" is a prominent conference where significant papers in the field of Natural Language Processing and Information Retrieval are presented. Notable papers presented at this conference include "Language models are few-shot learners" and "Retrieval-augmented generation for knowledge-intensive NLP tasks." Additionally, it is also the journal where the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" was published. This conference and journal serve as key platforms for disseminating cutting-edge research in neural information processing systems. + 58ae80c41cfe46db39da26b6a83584e5,6cd82819982879bd164547d2773ba5c7,b1bbda43309e8e0e2175ea034aa88e13 + + + PERSON + Cheng, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Luo, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Chen, X. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Liu, L. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Zhao, D. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory"Zhao, D. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + PERSON + + + PERSON + Yan, R. is an author of the paper "Lift yourself up: Retrieval-augmented text generation with self-memory" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dang, H. T. is an author of the paper "Duc 2005: Evaluation of question-focused summarization systems" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The conference where the paper "Duc 2005: Evaluation of question-focused summarization systems" was presented + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Es, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + James, J. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Espinosa-Anke, L. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Schockaert, S. is an author of the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Feng, Z. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Feng, X. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Yang, M. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Qin, B. is an author of the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Fortunato, S. is an author of the paper "Community detection in graphs" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The journal where the paper "Community detection in graphs" was published + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Gao, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Xiong, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models. The paper provides a comprehensive survey of the methodologies and applications of retrieval-augmented generation, highlighting its significance in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Gao, X. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Jia, K. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant area of research within the domains of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Pan, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Bi, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Dai, Y. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." These works focus on the integration of retrieval mechanisms with generative models to enhance the performance and capabilities of large language models, a significant area of research within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Sun, J. is an author of the paper "Retrieval-augmented generation for large language models: A survey" and also contributed to the paper titled "Retrieval-augmented generation." + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Wang, H. is an author of the paper titled "Retrieval-augmented generation for large language models: A survey." This work delves into the integration of retrieval mechanisms with generative models to enhance the performance of large language models, a significant topic within the fields of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Ragas: Automated evaluation of retrieval augmented generation" + 58ae80c41cfe46db39da26b6a83584e5 + + + PUBLICATION + The arXiv identifier for the paper "Retrieval-generation synergy augmented large language models" + 58ae80c41cfe46db39da26b6a83584e5 + + + PERSON + Goodwin, T. R. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Savery, M. E. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Demner-Fushman, D. is an author of the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + CONFERENCE + COLING (International Conference on Computational Linguistics) is the conference where the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" was presented + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + He, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Tian, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Sun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Chawla, N. V. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Laurent, T. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + LeCun, Y. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Bresson, X. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Hooi, B. is an author of the paper "G-retriever: Retrieval-augmented generation for textual graph understanding and question answering" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jacomy, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Venturini, T. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Heymann, S. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Bastian, M. is an author of the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" + 00e8e4e881bd0862022f4dfc913b900b + + + PUBLICATION + PLOS ONE is the journal where the paper "Forceatlas2, a continuous graph layout algorithm for handy network visualization designed for the gephi software" was published + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jin, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Yu, Z. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Jiao, P. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Pan, S. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + He, D. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Wu, J. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Philip, S. Y. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Zhang, W. is an author of the paper "A survey of community detection approaches: From statistical modeling to deep learning" + 00e8e4e881bd0862022f4dfc913b900b + + + PUBLICATION + IEEE Transactions on Knowledge and Data Engineering is the journal where the paper "A survey of community detection approaches: From statistical modeling to deep learning" was published + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Kang, M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Kwak, J. M. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Baek, J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Hwang, S. J. is an author of the paper "Knowledge graph-augmented language models for knowledge-grounded dialogue generation" + 00e8e4e881bd0862022f4dfc913b900b + + + PERSON + Khattab, O. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Santhanam, K. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Li, X. L. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hall, D. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text, indicating its relevance within the domain of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Liang, P. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Liang, P. contributed to the paper "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP," which explores the integration of retrieval and language models to enhance knowledge-intensive tasks in NLP. Additionally, Liang, P. authored the paper "Lost in the middle: How language models use long contexts," which investigates the utilization of extended contexts by language models. These contributions highlight Liang, P.'s significant role in advancing the understanding and application of language models in complex NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Potts, C. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the provided text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Zaharia, M. is an author of the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is referenced in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kim, G. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kim, S. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Jeon, B. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Park, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kang, J. is an author of the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Klein, G. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Moon, B. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hoffman, R. R. is an author of the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The journal where the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" were published + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Koesten, L. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Gregory, K. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Groth, P. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Simperl, E. is an author of the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The journal where the paper "Talking datasets–understanding data sensemaking behaviours" was published + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Kuratov, Y. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Bulatov, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Anokhin, P. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Sorokin, D. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Sorokin, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Burtsev, M. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + TECHNOLOGY + Langchain graphs is a technology developed by LangChain + 71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Laskar, M. T. R. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" and also contributed to the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models." Both works highlight Laskar's expertise in leveraging transformer models and transfer learning techniques to enhance the performance of query-focused abstractive text summarization, demonstrating a significant contribution to the field of Natural Language Processing and Information Retrieval. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Hoque, E. is an author of two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning to improve the effectiveness of transformer models in query-focused abstractive summarization tasks. Both works contribute to advancing the understanding and application of transformer models in specialized summarization contexts. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + PERSON + Huang, J. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + The conference where the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" was presented + 71f6daf11e64e5273a3847d46bf228e1 + + + PUBLICATION + arXiv preprint refers to a preprint of a paper that is available on the arXiv repository + 71f6daf11e64e5273a3847d46bf228e1 + + + EVENT + The 33rd Canadian Conference on Artificial Intelligence, held in Ottawa, ON, Canada, from May 13–15, 2020 + 6cd82819982879bd164547d2773ba5c7 + + + EVENT + The 2020 edition of the Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + PUBLISHER + Springer is the publisher of the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Huang, J. X. is an author of the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + PUBLICATION + The journal where the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" was published + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lewis, P. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Perez, E. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Piktus, A. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Petroni, F. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks"Petroni, F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + PERSON + + + PERSON + Karpukhin, V. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Goyal, N. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Küttler, H. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lewis, M. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Yih, W.-T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Rocktäschel, T. is an author of the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, N. F. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lin, K. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Hewitt, J. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Paranjape, A. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Bevilacqua, M. is an author of the paper "Lost in the middle: How language models use long contexts" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, Y. is an author of the paper "Hierarchical transformers for multi-document summarization" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Lapata, M. is an author known for significant contributions to the field of Natural Language Processing and Information Retrieval. Notably, Lapata, M. has authored the paper "Hierarchical transformers for multi-document summarization," which explores advanced techniques in summarizing information from multiple documents using hierarchical transformer models. Additionally, Lapata, M. has contributed to the paper "Text summarization with latent queries," which delves into innovative methods for summarizing text by leveraging latent query representations. These works highlight Lapata, M.'s expertise and active research in the domain of text summarization, showcasing a focus on developing sophisticated models and methodologies to enhance the efficiency and accuracy of summarization tasks. + 6cd82819982879bd164547d2773ba5c7,fc4b27d64f055b7fc30176ba110dd02e + + + TECHNOLOGY + LlamaIndex Knowledge Graph Index is a technology developed by LlamaIndex + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Manakul, P. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liusie, A. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Gales, M. J. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Mao, Y. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + He, P. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Liu, X. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Shen, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Shen, Y.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Gao, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Han, J. is an author of the paper "Generation-augmented retrieval for open-domain question answering" + 6cd82819982879bd164547d2773ba5c7 + + + PERSON + Chen, W. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" and also contributed to the paper "Generation-augmented retrieval for open-domain question answering." These works indicate Chen, W.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the integration of retrieval and generation techniques to improve the performance of large language models and open-domain question answering systems. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Martin, S. is an author of the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing a comprehensive, open-source solution for the layout of large graphs, which is a critical task in the visualization and analysis of complex networks. The toolbox aims to facilitate the understanding and interpretation of large-scale graph data, making it a valuable resource for researchers and practitioners in fields such as computational linguistics, information retrieval, and data science. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Brown, W. M. is an author of the paper "Openord: An open-source toolbox for large graph layout." + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + KLAVANS, R. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + PERSON + Boyack, K. is an author of the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on the development and application of Openord, a comprehensive open-source toolbox designed for the layout of large graphs. The paper likely discusses the methodologies, algorithms, and practical implementations of the toolbox, contributing to the fields of graph theory and data visualization. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + EVENT + The conference where the paper "Openord: An open-source toolbox for large graph layout" was presented + 833e7d67dcd30790b26b71c9b5306f6b + EVENT + + + TECHNOLOGY + GPT-4 is a large language model used in Microsoft's study on scientific discovery + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + + + TECHNOLOGY + Project NaLLM is a project developed by Neo4J + 833e7d67dcd30790b26b71c9b5306f6b + TECHNOLOGY + + + PERSON + Newman, M. E. is the author of the paper "Modularity and community structure in networks" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PUBLICATION + The journal where the paper "Modularity and community structure in networks" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + + + PERSON + Ram, O. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Levine, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Dalmedigos, I. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Muhlgay, D. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shashua, A. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Leyton-Brown, K. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shoham, Y. is an author of the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PUBLICATION + The journal where the paper "In-context retrieval-augmented language models" was published + 833e7d67dcd30790b26b71c9b5306f6b + PUBLICATION + + + PERSON + Ranade, P. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Joshi, A. is an author of the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Sarthi, P. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Abdullah, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Tuli, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Khanna, S. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Goldie, A. is an author of the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Manning, C. D. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" and the paper "Raptor: Recursive abstractive processing for tree-organized retrieval". These contributions highlight Manning's involvement in advancing the fields of Natural Language Processing and Information Retrieval, particularly in the areas of multi-hop question answering and recursive abstractive processing. + 833e7d67dcd30790b26b71c9b5306f6b,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Scott, K. is associated with "Behind the Tech" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Shao, Z. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Gong, Y. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b + PERSON + + + PERSON + Huang, M. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + + + PERSON + Duan, N. is an author of the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 833e7d67dcd30790b26b71c9b5306f6b,8d87efac8c50cf20cdf26bf61e5e2035 + PERSON + + + PERSON + Su, D. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Xu, Y. is an author of multiple academic papers in the field of Natural Language Processing and Information Retrieval. Notably, Xu, Y. contributed to the paper titled "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management," which addresses the management of scholarly information related to COVID-19 through advanced question answering and summarization techniques. Additionally, Xu, Y. co-authored the paper "Text summarization with latent queries," which explores innovative methods for text summarization by leveraging latent queries. These contributions highlight Xu, Y.'s expertise and active involvement in developing sophisticated systems for information retrieval and summarization. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yu, T. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Siddique, F. B. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Barezi, E. J. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Fung, P. is an author of the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Tang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Yang, Y. is an author of the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Touvron, H. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Martin, L. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Stone, K. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Albert, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Almahairi, A. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Babaei, Y. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bashlykov, N. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Batra, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bhargava, P. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Bhosale, S. is an author of the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Traag, V. A. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Waltman, L. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Van Eck, N. J. is an author of the paper "From Louvain to Leiden: guaranteeing well-connected communities" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PUBLICATION + Scientific Reports is the journal where the paper "From Louvain to Leiden: guaranteeing well-connected communities" was published + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trajanoska, M. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Stojanov, R. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trajanov, D. is an author of the paper "Enhancing knowledge graph construction using large language models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Trivedi, H. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Balasubramanian, N. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Khot, T. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Sabharwal, A. is an author of the paper "Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Wang, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Liang, Y. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Meng, F. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Sun, Z. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Shi, H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + Li, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through their work on evaluating language models. Specifically, Li, Z. has co-authored the paper titled "Is ChatGPT a Good NLG Evaluator? A Preliminary Study," which explores the effectiveness of ChatGPT as a natural language generation evaluator. Additionally, Li, Z. has co-authored another paper, "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which examines the performance of large language models in evaluative roles using specific benchmarking tools. These contributions highlight Li, Z.'s active involvement in advancing the understanding and assessment of language models within the academic community. + 8d87efac8c50cf20cdf26bf61e5e2035,b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Xu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Qu, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhou, J. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wang, S. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" and also contributed to the paper "Is chatgpt a good nlg evaluator? a preliminary study." These works indicate Wang, S.'s involvement in cutting-edge research within the fields of federated search, retrieval augmented generation, and natural language generation evaluation, showcasing a focus on both the technical and evaluative aspects of Natural Language Processing and Information Retrieval. + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Khramtsova is an author mentioned in the text + 8d87efac8c50cf20cdf26bf61e5e2035 + + + PERSON + H. is an author of the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Khramtsova, E. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhuang, S. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through multiple academic papers. Notably, Zhuang, S. co-authored the paper titled "Feb4rag: Evaluating federated search in the context of retrieval augmented generation," which explores the evaluation of federated search systems within the framework of retrieval-augmented generation. Additionally, Zhuang, S. co-authored another significant paper, "Judging llm-as-a-judge with mt-bench and chatbot arena," which delves into the assessment of large language models (LLMs) using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Zhuang, S.'s active involvement in advancing research in federated search and the evaluation of LLMs. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zuccon, G. is an author of the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wang, Y. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Lipka, N. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Rossi, R. A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Siu, A. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhang, R. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Derr, T. is an author of the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yang, Z. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Qi, P. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Zhang, S. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Bengio, Y. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Cohen, W. W. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Salakhutdinov, R. is an author of the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + CONFERENCE + The conference where the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" was presented + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yao, J.-g. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Wan, X. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Xiao, J. is an author of the paper "Recent advances in document summarization" + fc4b27d64f055b7fc30176ba110dd02e + + + PUBLICATION + The journal where the paper "Recent advances in document summarization" was published + fc4b27d64f055b7fc30176ba110dd02e + + + PERSON + Yao, L. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models"Yao, L. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Peng, J. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Mao, C. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Luo, Y. is an author of the paper "Exploring large language models for knowledge graph completion" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhang, J. is an author of the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhang, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Gan, Y. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Wang, C. is an author of the paper "Causal graph discovery with retrieval-augmented generation based large language models" + fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zheng, L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zheng, L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Zheng, L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools like MT-Bench and Chatbot Arena. These contributions highlight Zheng, L.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR domains. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Chiang, W.-L. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Chiang, W.-L. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Chiang, W.-L. has also authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Chiang, W.-L.'s active involvement in advancing the understanding and capabilities of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Sheng, Y. is an author known for contributing to the field of Natural Language Processing and Information Retrieval. Notably, Sheng, Y. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Sheng, Y. has contributed to the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Sheng, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic and technical community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Wu, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Wu, Z. co-authored the paper titled "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Wu, Z. is also credited with co-authoring the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Wu, Z.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP and IR communities. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Zhuang, Y. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant academic work. Notably, Zhuang, Y. has authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness of knowledge graphs. Additionally, Zhuang, Y. has also authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Zhuang, Y.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the domain. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Lin, Z. is an author of the paper "Exploring large language models for knowledge graph completion" and also contributed to the paper "Judging LLM-as-a-judge with MT-Bench and Chatbot Arena." These works indicate Lin, Z.'s involvement in advancing the field of Natural Language Processing and Information Retrieval, particularly focusing on the application of large language models for tasks such as knowledge graph completion and the evaluation of language models in judgment tasks. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Li, D. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through significant research. Notably, Li, D. has co-authored the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models in enhancing the completeness of knowledge graphs. Additionally, Li, D. has also co-authored the paper "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena," which evaluates the performance of large language models in judgment tasks using specific benchmarking tools. These contributions highlight Li, D.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the academic community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + PERSON + Xing, E. is an author of multiple influential papers in the field of Natural Language Processing and Information Retrieval. Notably, Xing, E. contributed to the paper "Exploring large language models for knowledge graph completion," which delves into the application of large language models to enhance the completeness and accuracy of knowledge graphs. Additionally, Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena," which evaluates the performance of large language models in judgment tasks using the MT-Bench and Chatbot Arena frameworks. These contributions highlight Xing, E.'s active involvement in advancing the capabilities and evaluation methodologies of large language models within the NLP community. + b1bbda43309e8e0e2175ea034aa88e13,fc4b27d64f055b7fc30176ba110dd02e + PERSON + + + TECHNOLOGY + Chatbot Arena is a platform or tool used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Darren Edge and Ha Trinh co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Darren Edge is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Newman Cheng co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Ha Trinh is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Joshua Bradley co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Newman Cheng is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Alex Chao co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Joshua Bradley is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Apurva Mody co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Alex Chao is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody and Steven Truitt co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Apurva Mody is affiliated with Microsoft Office of the CTO + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Steven Truitt and Jonathan Larson co-authored the paper "From Local to Global: A Graph RAG Approach to Query-Focused Summarization" + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Steven Truitt is affiliated with Microsoft Strategic Missions and Technologies + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Jonathan Larson is affiliated with Microsoft Research + e8d83e6e7a7c0f57b218cef24976b745 + + + 4.0 + RAG (Retrieval-Augmented Generation) and LLM (Large Language Models) are closely intertwined in the domain of Natural Language Processing and Information Retrieval. RAG is employed to enhance the capabilities of LLMs by enabling them to retrieve pertinent information from external knowledge sources. This symbiotic relationship allows LLMs to generate and assess text more effectively. Specifically, RAG leverages the power of LLMs to access and utilize relevant data, thereby augmenting the overall performance and accuracy of text generation tasks. + e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 7.0 + Graph RAG is a specific implementation of RAG that combines the strengths of RAG with graph-based text indexing. This method leverages the natural modularity of graphs to partition data, facilitating global summarization. As a specialized approach within the RAG framework, Graph RAG enhances the capabilities of RAG by integrating graph structures to improve the efficiency and effectiveness of text data processing and summarization. + 21e52bc06a82796b1f4bcd73edda1f2a,92e93fc6449756c0a60200636b297f65,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Query-Focused Summarization is a task that RAG fails to address effectively + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + RAG retrieves relevant information from an external knowledge source + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Naive RAG is a specific implementation of RAG + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Ram et al., 2023 discusses RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Naïve RAG is a basic form of RAG + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Modular RAG is an advanced form of RAG + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used in various RAG tasks such as knowledge graph creation and completion + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Trajanoska et al. discusses using LLMs for knowledge graph creation, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Yao et al. discusses using LLMs for knowledge graph completion, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Ban et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Zhang et al. discusses the extraction of causal graphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Gao et al. discusses advanced RAG where the index is a knowledge graph + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + KAPING is a method where the index is a knowledge graph, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + G-Retriever is a method where subsets of the graph structure are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Graph-ToolFormer is a method where derived graph metrics are the objects of enquiry, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + SURGE is a method where narrative outputs are strongly grounded in the facts of retrieved subgraphs, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + FABULA is a method where retrieved event-plot subgraphs are serialized using narrative templates, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Wang et al. discusses a system that supports both creation and traversal of text-relationship graphs for multi-hop question answering, which is a direction in RAG + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Sensemaking questions are used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The evaluation of RAG systems focuses on corpora in the region of 1 million tokens + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Trade-offs are considerations involved in building a graph index for RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + A graph index is a data structure used in RAG systems to organize and retrieve information + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Performance of RAG systems varies across different ranges of question types, data types, and dataset sizes + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Different data types are used in RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Dataset sizes affect the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Evaluation is the process of assessing the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Corpora are collections of texts used in the evaluation of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Different question types are used to evaluate RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Target metrics are specific measures used to evaluate the performance of RAG systems + 92e93fc6449756c0a60200636b297f65 + + + 4.0 + Graph RAG utilizes Large Language Models (LLMs) to construct a graph-based text index, enabling the generation of summaries and the answering of queries. In this approach, LLMs play a crucial role in analyzing and generating text based on the information retrieved through the graph structure. Additionally, LLMs leverage the Graph RAG framework to provide comprehensive overviews of public figures in the entertainment industry. This integration of LLMs within Graph RAG enhances the system's ability to process and synthesize complex text data, making it a powerful tool for information retrieval and natural language processing tasks. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Text chunks are processed using LLM to extract elements of a graph index + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM is used to extract elements of a graph index from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + LLM (Large Language Model) and Few-Shot Examples are closely related in the context of Natural Language Processing and Information Retrieval. Few-shot examples are provided to the LLM for in-context learning, which helps tailor the extraction prompt. This technique is particularly useful for improving the performance of the LLM in specialized domains. By leveraging a small number of examples, the LLM can better understand and adapt to specific tasks, thereby enhancing its overall effectiveness in extracting and processing information within those specialized areas. + 2c6ed90897310eea2f28e33fff1c32b0,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM extracts named entities from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Kuratov et al. (2024) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Liu et al. (2023) discuss the recall degradation of longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + LLM prompts are instructions given to the LLM for extracting elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Recall degradation occurs with longer LLM context windows + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + The extraction process involves using LLM to identify and extract elements from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Default prompt is the standard set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Secondary extraction prompt is an additional set of instructions given to the LLM + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + The LLM uses covariate prompts to extract additional attributes associated with detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM uses multiple rounds of gleanings to ensure no entities are missed + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Logit bias is used to force a yes/no decision from the LLM during entity extraction + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM extracts element instances from source texts + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + The LLM detects and summarizes communities of entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + LLM generates intermediate answers and scores for each chunk + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + LLM generates a helpfulness score for each answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + LLM is used to generate questions for evaluating the Podcast Transcripts dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + LLM is used to generate questions for evaluating the News Articles dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + LLM uses Naive RAG to list public figures mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + LLM-generated responses are evaluated using assessment metrics + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + LLM-generated responses are evaluated using specific questions + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Ad-hoc LLM use involves the use of large language models to analyze reasoning and provide specific examples, quotes, and citations + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + LLMs are used for knowledge graph creation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph completion + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for the extraction of causal graphs + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph creation as per Trajanoska et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for knowledge graph completion as per Yao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + LLMs are used for the extraction of causal graphs as per Ban et al. + f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is proposed as a method to combine the strengths of RAG and QFS + e8d83e6e7a7c0f57b218cef24976b745 + + + 7.0 + Graph RAG is an approach that utilizes community summaries in various capacities to enhance its functionality. Specifically, community summaries are generated within the Graph RAG framework and are employed to generate partial responses. These summaries are also used to compare against source texts, serving as a form of self-memory for the system. By leveraging community summaries, Graph RAG aims to improve the comprehensiveness and diversity of answers. Additionally, Graph RAG incorporates summaries of root-level communities within an entity-based graph index, further enriching its analytical capabilities. + 21e52bc06a82796b1f4bcd73edda1f2a,36db32c37e1987e2c5863898ad882190,6f33a085ff3304e5994f7fbb86c881a4,e4d9b12cf2b4c691c74019eefff4fb39,e8d83e6e7a7c0f57b218cef24976b745,f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is designed to handle global sensemaking questions over large datasets + e8d83e6e7a7c0f57b218cef24976b745 + + + 2.0 + Graph RAG is implemented in Python. + 086021a89900a39bcb62036981737bfa,e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The open-source implementation of Graph RAG will be available at this URL + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Graph RAG uses an entity knowledge graph to index text + e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG is an approach that is evaluated for its comprehensiveness, a target quality used to assess its effectiveness. The method aims to improve the comprehensiveness of generated answers, ensuring that the information provided is thorough and complete. This evaluation metric is crucial in determining the success of Graph RAG in producing detailed and accurate responses. + 21e52bc06a82796b1f4bcd73edda1f2a,e8c8f911135faf3ff35f24107eb3f99c,e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG is an approach in the domain of Natural Language Processing and Information Retrieval that focuses on improving the diversity of generated answers. Diversity, in this context, is a target quality used to evaluate the performance of the Graph RAG approach. By enhancing the diversity of responses, Graph RAG aims to provide a broader range of answers, thereby improving the overall effectiveness and robustness of the system. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4,e8d83e6e7a7c0f57b218cef24976b745 + + + 3.0 + Graph RAG uses a knowledge graph for global summarization + 21e52bc06a82796b1f4bcd73edda1f2a,fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Community detection algorithms are used in the Graph RAG approach to partition graphs + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Podcast transcripts are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + News articles are used as a dataset to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 2.0 + Graph RAG is evaluated using the target quality of Empowerment. Empowerment is specifically utilized to assess Graph RAG's capability in aiding users to achieve an informed understanding. This evaluation metric underscores the importance of user comprehension and the effectiveness of the Graph RAG approach in facilitating informed decision-making processes. + 21e52bc06a82796b1f4bcd73edda1f2a,6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Graph RAG is compared to naive RAG in the evaluation. In this comparison, Graph RAG outperformed naive RAG on comprehensiveness and diversity. + 21e52bc06a82796b1f4bcd73edda1f2a,4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Graph RAG is compared to global map-reduce summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Query-focused summarization is a method used in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Activity-centered sensemaking questions are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 2.0 + The "Graph RAG" approach is evaluated in terms of its performance by considering "Token Costs." Token costs are measured to assess the efficiency of the Graph RAG method, indicating that the computational expense associated with processing tokens is a critical factor in determining the overall effectiveness of this approach. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Data flow describes the high-level process of the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 3.0 + Design parameters are key settings in the Graph RAG approach and significantly influence the Graph RAG approach and pipeline. + 21e52bc06a82796b1f4bcd73edda1f2a,bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Graph RAG uses global summarization to summarize information from a large dataset + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG aims to answer specific queries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG uses a corpus for analysis and summarization + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Activity-centered sensemaking is used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Real-world datasets are used to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Hierarchical level of community summaries is varied to evaluate the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Graph RAG is compared to source text summarization in the evaluation + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Low-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Intermediate-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + High-level community summaries are generated in the Graph RAG approach + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + The Graph RAG approach involves a specific pipeline for processing and summarizing text + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Techniques are specific methods used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Implementation details are specific configurations used in the Graph RAG approach + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Graph RAG is a specific implementation of RAG systems + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + Graph RAG is a system that utilizes root-level community summaries, denoted as C0, to answer user queries. C0 represents these root-level community summaries within the Graph RAG analysis, serving as a foundational element in the system's ability to map out relationships and understand the structural dynamics within specialized communities. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses high-level community summaries (C1) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses intermediate-level community summaries (C2) to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 2.0 + Graph RAG utilizes low-level community summaries, represented by C3, to answer user queries. C3 plays a crucial role in the Graph RAG analysis by providing detailed summaries of community structures, which are essential for effectively addressing user inquiries. + 6f33a085ff3304e5994f7fbb86c881a4,973164fa90bf2b4ee267f4fd795916bf + + + 2.0 + Graph RAG is a key entity in the analysis, serving both as a condition being compared and as a tool for comparing multiple conditions. This dual role highlights its significance in the study, where it functions not only as a subject of comparison but also as a methodological framework for evaluating other conditions. The analysis likely involves a detailed examination of various conditions, with Graph RAG playing a central role in facilitating these comparisons. + 322e02986c8724eedbcf3ebfa20b989c,973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Graph RAG uses different levels of graph communities to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The Graph RAG mechanism uses an LLM evaluator for head-to-head comparison + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Graph RAG is a multi-stage mechanism + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Graph RAG mentions Taylor Swift as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Travis Kelce as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Britney Spears as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG mentions Justin Timberlake as a prominent public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Graph RAG is determined to be the winner based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Answer 1 was generated using the Graph RAG method, which provides a comprehensive and structured overview of public figures across various sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Graph RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Graph RAG is compared with source texts for answer comprehensiveness and diversity + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + TS represents source text summarization in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Root-level summaries are used in the Graph RAG analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Answer comprehensiveness is used to evaluate the performance of Graph RAG + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Win rate is used to measure the success rate of Graph RAG in providing comprehensive and diverse answers + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Element extraction prompts are used in Graph RAG to retain specific details in the index + 6f33a085ff3304e5994f7fbb86c881a4 + + + 2.0 + Graph RAG incorporates the concept of self-memory + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of iterative retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of federated retrieval-generation + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts used in multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts used in multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG uses a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates the concept of a tree of clarifications + f35de4d9fb65f1d5a392064b20545c19 + + + 3.0 + Graph RAG utilizes a self-generated graph index. This self-generated graph index is a crucial component of Graph RAG, enabling it to efficiently manage and retrieve information within its graph-based framework. The use of a self-generated graph index suggests that Graph RAG has an inherent capability to construct and maintain its indexing structure, which likely enhances its performance and adaptability in handling complex data relationships. + e4d9b12cf2b4c691c74019eefff4fb39,f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Gao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Cheng et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Mao et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Shao et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Wang et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Su et al., 2020 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Feng et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Trivedi et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Khattab et al., 2022 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Sarthi et al., 2024 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG incorporates concepts from Kim et al., 2023 + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Graph RAG generates community answers in parallel + f35de4d9fb65f1d5a392064b20545c19 + + + 1.0 + Graph RAG is compared to a graph-free approach for global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG is compared to map-reduce summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses rich text annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses a hierarchical community structure + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can operate using embedding-based matching + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can be part of hybrid RAG schemes + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can employ map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG can extend operations across the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Alonso contributed to the work on Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG includes local graph RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Graph RAG uses an entity-based graph index + e4d9b12cf2b4c691c74019eefff4fb39 + + + 2.0 + NebulaGraph launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Community summaries are used to generate partial responses + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Community summaries are created from graph communities + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Community answers are created from community summaries. These answers are generated by synthesizing information from the summaries provided by the community, ensuring that the responses are comprehensive and reflective of the collective knowledge and insights within the community. + 843fc5421e086120ffa1c75856ecf6cd,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Domain-tailored summarization is used to create community summaries + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Community descriptions are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Partial answers are generated from community summaries + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Community summaries are created for each level in the hierarchical community structure + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Community summaries are useful for understanding the global structure and semantics of the dataset + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Community summaries are used to answer global queries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are generated from root communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are generated from sub-communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are added to the LLM context window until the token limit is reached + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Global answers are generated from community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The level of summary detail affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The scope of information affects the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are used for sensemaking + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Community summaries are divided into chunks of pre-specified token size + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Summary detail and scope affect the content of community summaries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community summaries are divided into chunks + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Community summaries are prepared to answer user queries + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Intermediate answers are generated from community summaries + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Community summaries are part of the graph community hierarchy + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Community summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Summaries of root-level communities are used in Graph RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Global sensemaking questions are evaluated over datasets in the 1 million token range + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + Global sensemaking questions are directed at an entire text corpus + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The Python-based implementation of Graph RAG approaches will be available at this URL + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Query-focused summarization is used to produce the global answer + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Map-reduce is used for query-focused summarization of an entire corpus + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Query-focused summarization is used for answering global queries + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + An entity knowledge graph is derived from source documents + e8d83e6e7a7c0f57b218cef24976b745 + + + 2.0 + In the domain of Natural Language Processing and Information Retrieval, "SOURCE DOCUMENTS" and "TEXT CHUNKS" are closely related entities. Text chunks are extracted from source documents, specifically for processing in the Graph RAG (Retrieval-Augmented Generation) approach. This method involves breaking down the source documents into smaller, manageable pieces, or text chunks, which can then be effectively utilized in various computational processes, enhancing the efficiency and accuracy of information retrieval and generation tasks. + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Intermediate-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Low-level community summaries are derived from source documents + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Document corpus consists of source documents being processed + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Partial responses are summarized to generate a final response + e8d83e6e7a7c0f57b218cef24976b745 + + + 1.0 + The LLM evaluator assesses answers based on the comprehensiveness metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Naive RAG is evaluated for comprehensiveness + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Comprehensiveness is a metric used to determine the decision + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Comprehensiveness is used to evaluate the thoroughness of the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) was universally better for comprehensiveness + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The final evaluation prioritized comprehensiveness in answers + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Global approaches achieved higher comprehensiveness win rates + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The LLM evaluator assesses answers based on the diversity metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Diversity is used to evaluate the variety in the generated answers for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on diversity + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The final evaluation prioritized diversity in answers + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Global approaches achieved higher diversity win rates + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Human endeavors rely on sensemaking to understand and reason about large collections of documents + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Human endeavors rely on analyzing document collections for sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + LLMs are used to automate sensemaking in complex domains + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Microsoft uses LLMs for automating sensemaking in scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Ranade uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Joshi uses LLMs for automating sensemaking in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + LLM prompts are used to tailor the responses of large language models + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Ranade and Joshi discussed the use of LLMs in intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + GPT is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Llama is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Gemini is a type of large language model + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Kuratov et al., 2024, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Liu et al., 2023, discussed the limitations of LLM context windows + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Sensemaking is applied in the domain of scientific discovery + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Sensemaking is applied in the domain of intelligence analysis + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Klein defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Klein et al. defined and discussed the importance of sensemaking + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Element instances are extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Entity references are extracted from text chunks during processing + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Chunk size refers to the length of text chunks used in the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Element instances are converted into element summaries by the LLM (Large Language Model). Element summaries are created from element instances, indicating a transformation process facilitated by the LLM. This process involves the LLM taking detailed element instances and generating concise element summaries, which encapsulate the essential information from the instances. + 2c6ed90897310eea2f28e33fff1c32b0,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Covariates are additional attributes associated with extracted element instances + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Domain-tailored summarization is used to create element summaries + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Element summaries include descriptions of entity nodes + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries include descriptions of relationship edges + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries include descriptions of claim covariates + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Element summaries are used to understand the structure and semantics of graph communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Element summaries include descriptions of nodes + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Element summaries include descriptions of edges + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Element summaries include descriptions of covariates + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Sub-community summaries are used when element summaries exceed the token limit + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Community detection is a technique used to identify graph communities. Graph communities are groups of nodes within a graph that are more densely connected to each other than to the rest of the graph. This process of identifying such communities is crucial for understanding the structural dynamics and relationships within complex networks, particularly in the domain of Natural Language Processing and Information Retrieval. By leveraging community detection algorithms, researchers can uncover hidden patterns and insights within large datasets, facilitating more effective data analysis and interpretation. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Global answer is created from community answers + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Global answers are generated in response to user queries + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Global answer is generated by sorting intermediate answers based on helpfulness scores + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Intermediate answers are combined to form the global answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + The final context window is used to generate the global answer + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Graph RAG pipeline operates at indexing time + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Graph RAG pipeline operates at query time + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Nodes are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Edges are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Covariates are detected in the graph RAG pipeline + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Leiden method is used in the graph RAG pipeline for community detection + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Graph RAG pipeline uses the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + The Podcast dataset graph consists of 8564 nodes + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The News dataset graph consists of 15754 nodes + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The Podcast dataset graph consists of 20691 edges + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + The News dataset graph consists of 19520 edges + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Traag contributed to the development of the Leiden method + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Traag et al. are the authors of the Leiden algorithm and developed the Leiden method. + 7fb7d9ce2da9c940a32afdd87d1d9e56,f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Leiden is a specific type of community detection algorithm used in various analytical pipelines. It is designed to identify and map out the structural dynamics within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. The algorithm is known for its efficiency and accuracy in detecting community structures, making it a valuable tool for researchers and practitioners in the field. + 21e52bc06a82796b1f4bcd73edda1f2a,7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Leiden is known for its ability to recover hierarchical community structures efficiently + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The Leiden algorithm is used to detect graph communities in the MultiHop-RAG + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Figure 3 shows graph communities detected using the Leiden algorithm + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Lewis contributed to the development of the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Lewis et al. developed the RAG approach + f0306814bf64f5c9e79603fc6a52f4ea + + + 2.0 + Lewis et al., 2020, are the authors who established the RAG approach + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Kevin Scott is the CTO of Microsoft + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + Microsoft conducted a study on the impact of large language models on scientific discovery using GPT-4 + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Preprint is available on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Baumel, T. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Elhadad, M. published the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Es, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + James, J. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Espinosa-Anke, L. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Schockaert, S. published the paper "Ragas: Automated evaluation of retrieval augmented generation" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, Z. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, X. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Zhao, D. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Yang, M. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Qin, B. published the paper "Retrieval-generation synergy augmented large language models" on arXiv + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + LangChain is an organization that has published on arXiv + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Wang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, S. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zuccon, G. published the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, R. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Derr, T. published the paper "Knowledge graph prompting for multi-document question answering" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Xu, Y. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lapata, M. published the paper "Text summarization with latent queries" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, J. published the paper "Graph-toolformer: To empower llms with graph reasoning ability via prompt augmented by chatgpt" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Gan, Y. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yao, L. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, C. published the paper "Causal graph discovery with retrieval-augmented generation based large language models" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Chiang, W.-L. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Sheng, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wu, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, Y. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lin, Z. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Li, D. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Xing, E. published the paper "Exploring large language models for knowledge graph completion" on arXiv + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Preprint is classified under cs.CL on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Preprint was submitted on 24 Apr 2024 + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Preprint has the identifier 2404.16130v1 on arXiv + f0306814bf64f5c9e79603fc6a52f4ea + + + 1.0 + Community detection results in the partition of a graph into distinct communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The pipeline includes a step for community detection + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 2.0 + Dang, 2006, is the author who established the QFS approach + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Baumel et al., 2018, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Laskar et al., 2020, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Yao et al., 2017, are the authors who worked on query-focused abstractive summarization + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Goodwin et al., 2020, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Laskar et al., 2022, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Liu and Lapata, 2019, are the authors who worked on the early applications of the transformer architecture + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Achiam et al., 2023, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Brown et al., 2020, are the authors who worked on the GPT series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Touvron et al., 2023, are the authors who worked on the Llama series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 2.0 + Anil et al., 2023, are the authors who worked on the Gemini series of large language models + fb3c48579608fa28be585ceb6cd2f0fe + + + 1.0 + Modularity is an inherent quality of knowledge graphs + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Brown et al. (2020) discuss in-context learning with few-shot examples + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Kuratov et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Liu et al. discussed the potential for information to be lost in longer contexts + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Louvain is a type of community detection algorithm + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Community detection algorithms are used to partition the graph index into communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Fortunato has conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Jin et al. have conducted surveys on community detection algorithms + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + HotPotQA dataset is used to evaluate the entity extraction prompt with gpt-4-turbo + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Yang et al. (2018) introduced the HotPotQA dataset + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 2.0 + Yang et al. are the authors associated with the HotPotQA dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + GPT-4-Turbo was tested with varying context window sizes + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Tech journalist uses podcast transcripts to look for insights and trends in the tech industry + 1d07b4248c2655081c7af0e373bd70c9 + + + 3.0 + Kevin Scott is a participant in the podcast conversations compiled in the Podcast Transcripts dataset. His conversations are included as part of the podcast transcripts, contributing to the overall content and discussions captured within this dataset. + 1d07b4248c2655081c7af0e373bd70c9,922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Technology leaders participate in the podcast conversations + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + RAG systems are used to evaluate the Podcast Transcripts dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + C0 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C1 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C2 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + C3 is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 2.0 + TS is a category used in the analysis of podcast transcripts + 36db32c37e1987e2c5863898ad882190,ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + SS is a category used in the analysis of podcast transcripts + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Units are used to measure the context in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in podcast transcripts + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Both are datasets used in the analysis + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Educator uses news articles to incorporate current affairs into curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + 2.0 + RAG systems are used to evaluate the News Articles dataset for global sensemaking tasks + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + C0 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in news articles + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Map-reduce is the method used in the text summarization condition + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The LLM evaluator assesses answers based on the empowerment metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Empowerment is used to evaluate how empowering the generated answers are for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The smallest context window size (8k) performed comparably with larger context sizes on empowerment + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Empowerment has an average win rate of 51.3% + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Naive RAG mentions Taylor Swift as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Travis Kelce as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Britney Spears as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG mentions Justin Timberlake as a public figure + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Naive RAG is determined to be the loser based on the decision metric + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Global approaches consistently outperformed the naive RAG + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Naive RAG produces the most direct responses + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + SS represents naive RAG in the analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Gao et al., 2023 discusses naive RAG approaches + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + Community partitions enable divide-and-conquer global summarization + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Global summarization can be performed using a graph-free approach + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Source texts are used in global summarization + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Final global answer is generated by combining all relevant partial answers + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Short descriptions are used to generate sensemaking questions + 21e52bc06a82796b1f4bcd73edda1f2a + + + 1.0 + Low-level community summaries are derived from the News dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + The use of rich descriptive text for homogeneous nodes in a graph index aligns with the capabilities of LLMs + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Graph indexes differ from typical knowledge graphs in their use of rich descriptive text instead of concise knowledge triples + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The graph index supports condition C0 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C1 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C2 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index supports condition C3 + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens with 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph index was created using generic prompts for entity and relationship extraction + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Few-shot examples tailored to the domain of the data were used in the graph indexing process + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The graph indexing process used a context window size of 600 tokens + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The decision to build a graph index depends on the expected number of lifetime queries per dataset + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The decision to build a graph index depends on the value obtained from it + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The decision to build a graph index depends on the value obtained from other graph-related RAG approaches + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Recall measures the completeness of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Precision measures the accuracy of entity references extracted from text chunks + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to tailor the default prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to tailor the secondary extraction prompt to the domain + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of science + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of medicine + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Few-shot examples are used to improve LLM performance in the domain of law + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + A single extraction round is part of the extraction process + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Domain refers to the specific area of knowledge of the document corpus + bc9e2c9e369c4108cf4f6dd5f60960f4 + + + 1.0 + Covariate prompts are used to extract claims linked to detected entities + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Source text span is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Start date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + End date is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Description is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Subject is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Object is an attribute of claims + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Communities of entities help manage variations in a noisy graph structure + 2c6ed90897310eea2f28e33fff1c32b0 + + + 1.0 + Common entities are described using rich descriptive text for homogeneous nodes + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + LLMs are used to generate metrics for evaluating natural language generation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Wang et al. (2023) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Zheng et al. (2024) indicated the effectiveness of LLMs in evaluation + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Relationship edges connect homogeneous nodes in a graph + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Edge weights represent the normalized counts of detected relationship instances on relationship edges + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Each level of the hierarchical community structure provides a community partition + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 2.0 + The hierarchical community structure is a framework used to organize and understand the relationships and dynamics within specialized communities. Root communities are an integral part of this structure, serving as the top-level communities. These root communities form the foundational layer in the hierarchical community structure, providing a basis for further subdivision and organization of more specific sub-communities. This hierarchical approach allows for a systematic analysis of complex networks, facilitating a deeper understanding of the interconnections and dependencies within the domain of Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + The hierarchical community structure is a framework that organizes communities into different levels, with sub-communities representing the lower-level communities within this structure. Sub-communities are integral components of the hierarchical community structure, indicating that they are nested within larger communities and contribute to the overall organization and dynamics of the community. This hierarchical arrangement allows for a more detailed and nuanced understanding of the relationships and interactions within the community, facilitating more effective analysis and mapping of complex text data, particularly in specialized domains such as Natural Language Processing and Information Retrieval. + 7fb7d9ce2da9c940a32afdd87d1d9e56,843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Community levels are part of the hierarchical community structure + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + The Leiden algorithm is used to detect communities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + OpenORD is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Force Atlas 2 is used for node layout in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Nodes represent entities in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Edges represent connections between nodes in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Covariates are variables linked to nodes and edges in the MultiHop-RAG dataset + 843fc5421e086120ffa1c75856ecf6cd + + + 2.0 + Tang and Yang are the authors associated with the MultiHop-RAG dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Questions are generated based on the target datasets + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + N represents the number of test questions per dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Table 1 shows example questions for each of the two evaluation datasets + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Root communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Level 0 represents the root-level communities in the hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Reports provide detailed information about specific subtopics within sub-communities + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Sub-communities are identified through hierarchical clustering + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Level 1 represents sub-communities within the root-level communities + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Partitions can be organized into a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Level 0 is the root level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + Level 1 is a sub-level in a hierarchy + 7fb7d9ce2da9c940a32afdd87d1d9e56 + + + 1.0 + The token limit defines the maximum number of tokens in the LLM context window + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Prominence is used to prioritize community edges + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Combined source and target node degree is used to measure prominence + 843fc5421e086120ffa1c75856ecf6cd + + + 1.0 + Chunks are divided based on a pre-specified token size + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Helpfulness scores are assigned to intermediate answers + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in episodes dealing with tech policy and government regulation + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in how guests perceive the impact of privacy laws on technology development + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in discussions about the balance between innovation and ethical considerations + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in suggested changes to current policies + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Tech journalist is interested in discussions about collaborations between tech companies and governments + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in current topics in health that can be integrated into health education curricula + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in how news articles address the concepts of preventive medicine and wellness + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in examples of health articles that contradict each other + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in insights about public health priorities based on news coverage + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Educator is interested in highlighting the importance of health literacy through the dataset + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + The size of the context window and the prompts used for answer generation are the same across all conditions + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + The task is an activity or goal that the user aims to achieve + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Questions are generated based on the user's task + 1d07b4248c2655081c7af0e373bd70c9 + + + 1.0 + Datasets were used in combination with questions for the analysis + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Questions were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + + + 2.0 + Zheng et al. are the authors associated with the MT-Bench dataset + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Zheng, L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Xing, E. is an author of the paper that discusses MT-Bench + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + MT-Bench and Chatbot Arena are both tools used in the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 2.0 + Koesten et al. authored a paper on data sensemaking behaviors + 922778ce1cb2fdd6dbab1746c8795620 + + + 2.0 + Xu and Lapata authored a paper on methods for extracting latent summarization queries from source texts + 922778ce1cb2fdd6dbab1746c8795620 + + + 1.0 + Text summarization method applies a map-reduce approach directly to source texts (TS) + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Text summarization is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Semantic search RAG is a na¨ıve RAG approach where text chunks are retrieved and added to the context window until the token limit is reached (SS) + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + Semantic search RAG is one of the conditions compared in the analysis + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C0 uses root-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C0 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C0 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C0 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 uses high-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C1 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C1 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C1 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C1 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 uses intermediate-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C2 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C2 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C2 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C2 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 uses low-level community summaries to answer user queries + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + C3 is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + C3 showed slight improvements in answer comprehensiveness and diversity over TS + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + C3 is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + C3 is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + TS is a category used in the analysis of the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + TS is a category used in the analysis of the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + SS is a category used in the analysis of news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The graph indexing process used 1 gleaning for the Podcast dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + A graph was created for the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the Podcast dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Intermediate-level summaries are derived from the Podcast dataset for analysis + 6f33a085ff3304e5994f7fbb86c881a4 + + + 1.0 + The graph indexing process used 0 gleanings for the News dataset + 973164fa90bf2b4ee267f4fd795916bf + + + 1.0 + A graph was created for the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Units are used to measure the context in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Tokens are used to measure the word count in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + % Max is used to measure the percentage of maximum token count in the News dataset + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Datasets were evaluated using various metrics + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Wang et al., 2023a discusses the state-of-the-art results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Zheng et al., 2024 discusses the competitive results achieved by Natural Language Generation + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Zheng et al., 2024 discusses the LLM-as-a-judge method + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Embedding-based matching is used to match user queries + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Query-time LLM use was evaluated with different context window sizes + 4c855404ee3d3c94aa2136f1513c666f + + + 2.0 + The **CONTEXT WINDOW SIZE** and **FINAL EVALUATION** are closely related in the given data. A fixed context window size of 8k tokens was used for the final evaluation. This indicates that during the final evaluation phase, the system or model was configured to process and analyze text data within a predefined window of 8,000 tokens, ensuring consistency and standardization in the evaluation process. + 36db32c37e1987e2c5863898ad882190,4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Natural Language Generation achieves state-of-the-art results + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation achieves competitive results + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation is compared against human judgements + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation can generate reference-based metrics + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Natural Language Generation can measure qualities in a reference-free style + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Es et al., 2023 discusses the RAGAS method + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates context relevance + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates faithfulness + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + RAGAS evaluates answer relevance + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator assesses answers based on the directness metric + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Table 2 shows an example of LLM-generated assessment + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses a head-to-head comparison approach + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator assesses answers based on target metrics + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses a control metric for validity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator accounts for stochasticity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + The LLM evaluator uses mean scores from multiple comparisons + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Directness is used to evaluate the straightforwardness of the generated answers for news articles + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The question asks about public figures mentioned in entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Public figures are repeatedly mentioned across various entertainment articles + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Answer 1 covers a wide range of public figures from different sectors of the entertainment industry. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Controversies involve public figures and impact public discourse. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Entertainment articles cover topics related to the entertainment industry + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Taylor Swift is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Travis Kelce is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Britney Spears is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Justin Timberlake is frequently mentioned in entertainment articles + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Taylor Swift is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Travis Kelce is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Britney Spears is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Justin Timberlake is a significant figure in the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Actors and Directors are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Musicians and Executives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Athletes and Coaches are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Influencers and Entrepreneurs are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Public Figures in Controversy are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Film is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Television is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Music is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Sports is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Digital Media is a sector within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Cultural Narratives are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Trends are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Social Discussions are a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Public Discourse is a category within the entertainment industry + e8c8f911135faf3ff35f24107eb3f99c + + + 1.0 + Reference-based metrics require gold standard answers + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Gold standard answers are lacking for sensemaking questions + 322e02986c8724eedbcf3ebfa20b989c + + + 3.0 + End users play a crucial role in the validation process of sensemaking questions and target metrics. Sensemaking questions are specifically validated with end users to ensure their relevance and accuracy. This collaborative approach ensures that the questions and metrics are aligned with the needs and expectations of the end users, thereby enhancing the overall effectiveness and applicability of the sensemaking process. + 92e93fc6449756c0a60200636b297f65,e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Target metrics are validated with end users + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The control metric is used as an indicator of validity + 322e02986c8724eedbcf3ebfa20b989c + + + 1.0 + Taylor Swift is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Taylor Swift is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Travis Kelce is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Travis Kelce is a public figure in the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Britney Spears is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Britney Spears is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Justin Timberlake is one of the specific public figures mentioned in Answer 2. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Justin Timberlake is a public figure in the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the film sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the television sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on public figures primarily from the music sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 focuses on public figures primarily from the sports sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the digital media sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 cites specific data sources from the News article dataset for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 provides insights into controversies involving public figures and their impact on public discourse. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 includes public figures from the gaming sector. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 1 cites specific data sources for each mentioned figure. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Answer 2 was generated using the Naïve RAG method, which directly lists specific public figures who are repeatedly mentioned across various entertainment articles. + 718017a4871c909420f84b85b8ba969d + + + 2.0 + ANSWER 2 is a generated answer for a question in the NEWS ARTICLE DATASET. It relies heavily on a single source from the NEWS ARTICLE DATASET for data. + 718017a4871c909420f84b85b8ba969d,ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Answer 2 relies heavily on a single data source. + 718017a4871c909420f84b85b8ba969d + + + 1.0 + Naïve RAG is used to generate answers for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + The global approach to Graph RAG shows improvements over naïve RAG + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + LLM-generated assessments are used to evaluate the answers produced for questions in the News article dataset + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Example question is part of the News article dataset used for analysis + ebf5249c888e07fedce6572a4c03f88c + + + 1.0 + Head-to-head win rate percentages were used to compare different conditions + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Win rate percentages were used to measure the performance of different conditions + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The overall winner per dataset and metric was determined for each condition + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + Self-win rates were shown as the expected 50% for each condition + 4c855404ee3d3c94aa2136f1513c666f + + + 1.0 + The indexing process resulted in the creation of graphs + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Map-reduce summarization requires the highest number of context tokens + 36db32c37e1987e2c5863898ad882190 + + + 1.0 + Root-level community summaries require dramatically fewer tokens per query + 36db32c37e1987e2c5863898ad882190 + + + 2.0 + Queries are embedded into the same vector space as text chunks to find relevant context + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + Self-memory is related to generation-augmented retrieval + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + CAiRE-COVID is a system for multi-document summarization + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + ITRG is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + IR-CoT is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + DSP is a system for multi-hop question answering + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + RAPTOR is a method for generating a hierarchical index + f35de4d9fb65f1d5a392064b20545c19 + + + 2.0 + The paper by Baek et al. discusses the KAPING method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by He et al. discusses the G-Retriever method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Zhang discusses the Graph-ToolFormer method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Kang et al. discusses the SURGE method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Ranade and Joshi discusses the FABULA method + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Both LangChain and LlamaIndex support a variety of graph databases + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LangChain supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LangChain supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 1.0 + LangChain developed Langchain graphs + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + LlamaIndex supports graph databases in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + LlamaIndex supports graph databases in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + NaLLM is a method that can create and reason over knowledge graphs in Neo4J format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + Neo4J developed Project NaLLM + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + GraphRAG is a method that can create and reason over knowledge graphs in NebulaGraph format + 92e93fc6449756c0a60200636b297f65 + + + 2.0 + The paper by Manakul et al. discusses the SelfCheckGPT method + 92e93fc6449756c0a60200636b297f65 + + + 1.0 + SelfCheckGPT is an approach mentioned in the work by Manakul et al., 2023 + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + SelfCheckGPT is used to compare fabrication rates + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Embedding-based matching is used to match user queries with graph annotations + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Hybrid RAG schemes combine embedding-based matching against community reports + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The roll-up operation can be extended using map-reduce summarization mechanisms + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The drill down mechanism follows the information scent in the community hierarchy + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The global approach to Graph RAG achieves competitive performance at a fraction of the token cost + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The open-source implementation of Graph RAG approaches is Python-based + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + The drill down mechanism follows the information scent + e4d9b12cf2b4c691c74019eefff4fb39 + + + 1.0 + Alonso Guevara Fernández and Amber Hoak both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Sarah Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Alonso Guevara Fernández and Shane Solomon both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Andrés Morales Esquivel both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Ben Cutler both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Billie Rinaldi both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Chris Sanchez both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Chris Trevino both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Christine Caggiano both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and David Tittsworth both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Dayenne de Souza both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Douglas Orbaker both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Ed Clark both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Gabriel Nieves-Ponce both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Gaudy Blanco Meneses both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Kate Lytvynets both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Katy Smith both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Mónica Carvajal both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Nathan Evans both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Richard Ortega both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + Amber Hoak and Rodrigo Racanicci both contributed to the work acknowledged in the document + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Adler co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Achiam and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Agarwal co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Adler and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and L. Ahmad co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Agarwal and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and I. Akkaya co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + L. Ahmad and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and F. L. Aleman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + I. Akkaya and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and D. Almeida co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + F. L. Aleman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and J. Altenschmidt co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + D. Almeida and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Altenschmidt and S. Altman co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Altenschmidt and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Altman and S. Anadkat co-authored the GPT-4 technical report + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and S. Borgeaud co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Anil and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and Y. Wu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + S. Borgeaud and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J.-B. Alayrac co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + Y. Wu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and J. Yu co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J.-B. Alayrac and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and R. Soricut co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Yu and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and J. Schalkwyk co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + R. Soricut and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Schalkwyk and A. M. Dai co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Schalkwyk and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + A. M. Dai and A. Hauth co-authored the Gemini paper + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Baek and A. F. Aji co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + J. Baek and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + A. F. Aji and A. Saffari co-authored the paper on knowledge-augmented language model prompting + 086021a89900a39bcb62036981737bfa + + + 1.0 + T. Ban and L. Chen co-authored the paper on query tools to causal architects + 086021a89900a39bcb62036981737bfa + + + 1.0 + Baumel, T. and Eyal, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Baumel, T. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Baumel, T. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. and Elhadad, M. co-authored the paper "Query focused abstractive summarization: Incorporating query relevance, multi-document coverage, and summary length constraints into seq2seq models" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Eyal, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Elhadad, M. is an author of the paper with arXiv identifier 1801.07704 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Guillaume, J.-L. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Blondel, V. D. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Guillaume, J.-L. and Lambiotte, R. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Guillaume, J.-L. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Lambiotte, R. and Lefebvre, E. co-authored the paper "Fast unfolding of communities in large networks" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Mann, B. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Brown, T. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Ryder, N. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Mann, B. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Subbiah, M. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Ryder, N. and Askell, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Kaplan, J. D. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Dhariwal, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Neelakantan, A. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Shyam, P. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Subbiah, M. and Sastry, G. co-authored the paper "Language models are few-shot learners" + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Zhao, D. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Es, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + James, J. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Espinosa-Anke, L. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Schockaert, S. is an author of the paper with arXiv identifier 2309.15217 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, Z. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Feng, X. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Yang, M. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Qin, B. is an author of the paper with arXiv identifier 2310.05149 + 58ae80c41cfe46db39da26b6a83584e5 + + + 1.0 + Gao, Y. and Xiong, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Gao, X. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Xiong, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Jia, K. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Gao, X. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Pan, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Jia, K. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Bi, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Pan, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Dai, Y. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Bi, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Dai, Y. and Sun, J. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Dai, Y. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Sun, J. and Wang, H. co-authored the paper "Retrieval-augmented generation for large language models: A survey" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Goodwin, T. R. and Savery, M. E. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + 1.0 + Goodwin, T. R. and Demner-Fushman, D. co-authored the paper "Flight of the pegasus? comparing transformers on few-shot and zero-shot multi-document abstractive summarization" + 00e8e4e881bd0862022f4dfc913b900b + + + 2.0 + Khattab, O. and Santhanam, K. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and highlights their collaborative work in the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This work is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive NLP tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Khattab, O. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Khattab, O. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Li, X. L. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This collaboration is mentioned in the text, highlighting their joint contribution to the field of Natural Language Processing and Information Retrieval. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and contributes to the field of Natural Language Processing and Information Retrieval by exploring the integration of retrieval and language models to enhance knowledge-intensive tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Santhanam, K. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Santhanam, K. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Hall, D. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Li, X. L. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Li, X. L. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Liang, P. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Hall, D. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hall, D. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Liang, P. and Potts, C. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + LIANG, P. and ZAHARIA, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Liang, P. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Potts, C. and Zaharia, M. co-authored the paper titled "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive NLP." This paper is mentioned in the text and focuses on integrating retrieval and language models to enhance knowledge-intensive natural language processing tasks. + 00e8e4e881bd0862022f4dfc913b900b,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Potts, C. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Zaharia, M. is an author of the arXiv preprint "Demonstrate-search-predict: Composing retrieval and language models for knowledge-intensive nlp" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Kim, S. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, G. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Jeon, B. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kim, S. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. and Park, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Jeon, B. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Park, J. and Kang, J. co-authored the paper "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Park, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kang, J. is an author of the arXiv preprint "Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. and Moon, B. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Klein, G. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Moon, B. and Hoffman, R. R. co-authored the papers "Making sense of sensemaking 1: Alternative perspectives" and "Making sense of sensemaking 2: A macrocognitive model" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Moon, B. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoffman, R. R. is an author of papers published in IEEE Intelligent Systems + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Gregory, K. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Koesten, L. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. and Groth, P. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Gregory, K. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Groth, P. and Simperl, E. co-authored the paper "Talking datasets–understanding data sensemaking behaviours" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Groth, P. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Simperl, E. is an author of a paper published in International Journal of Human-Computer Studies + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Kuratov, Y. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Bulatov, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Anokhin, P. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Sorokin, D. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Sorokin, A. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Burtsev, M. is an author of the arXiv preprint "In search of needles in a 11m haystack: Recurrent memory finds what llms miss" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 2.0 + Laskar, M. T. R. and Hoque, E. co-authored two significant papers in the field of Natural Language Processing and Information Retrieval. The first paper, titled "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization," explores the use of pre-trained transformer models to enhance the performance of query-focused abstractive summarization through domain adaptation techniques. The second paper, "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models," delves into the integration of query relevance and transfer learning with transformer models to improve the effectiveness of query-focused abstractive summarization. Both works contribute to advancing the application of transformer models in specialized summarization tasks. + 6cd82819982879bd164547d2773ba5c7,71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Laskar, M. T. R. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Laskar, M. T. R. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Hoque, E. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoque, E. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + Hoque, E. and Huang, J. X. co-authored the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Hoque, E. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Huang, J. is an author of a paper presented at Advances in Artificial Intelligence + 71f6daf11e64e5273a3847d46bf228e1 + + + 1.0 + The 33rd Canadian Conference on Artificial Intelligence is also known as Canadian AI 2020 + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Springer published the proceedings of the 33rd Canadian Conference on Artificial Intelligence + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Huang, J. X. published the paper "Domain adaptation with pre-trained transformers for query-focused abstractive text summarization" in Computational Linguistics + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Perez, E. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Lewis, P. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Piktus, A. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Perez, E. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Petroni, F. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Piktus, A. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Karpukhin, V. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Küttler, H. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Lewis, M. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Yih, W.-T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Petroni, F. and Rocktäschel, T. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Karpukhin, V. and Goyal, N. co-authored the paper "Retrieval-augmented generation for knowledge-intensive NLP tasks" + 6cd82819982879bd164547d2773ba5c7 + + + 1.0 + Xu, Y. and Lapata, M. co-authored the paper "Text summarization with latent queries" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Huang, M. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Duan, N. and Chen, W. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 3.0 + Martin, S. and Brown, W. M. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with large graph structures, making it a valuable resource for researchers and practitioners in the domain of graph theory and network analysis. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Martin, S. and Klavans, R. co-authored the paper "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Martin, S. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of graph visualization and analysis. The paper highlights the capabilities and applications of the Openord toolbox, emphasizing its utility in handling extensive graph data efficiently. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Brown, W. M. and Klavans, R. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + Brown, W. M. and Boyack, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of Openord, a toolbox designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 3.0 + KLAVANS, R. and BOYACK, K. co-authored the paper titled "Openord: An open-source toolbox for large graph layout." This work focuses on providing an open-source solution for the layout of large graphs, which is a significant contribution to the field of Natural Language Processing and Information Retrieval. The paper highlights the development and application of the Openord toolbox, which is designed to handle the complexities associated with visualizing large-scale graph data. + 6cd82819982879bd164547d2773ba5c7,833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Newman, M. E. published the paper "Modularity and community structure in networks" in the Proceedings of the National Academy of Sciences + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Levine, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ram, O. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Dalmedigos, I. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Levine, Y. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Muhlgay, D. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Dalmedigos, I. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Shashua, A. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Muhlgay, D. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Shashua, A. and Leyton-Brown, K. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Shashua, A. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Leyton-Brown, K. and Shoham, Y. co-authored the paper "In-context retrieval-augmented language models" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Ranade, P. and Joshi, A. co-authored the paper "Fabula: Intelligence report generation using retrieval-augmented narrative construction" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Abdullah, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Sarthi, P. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Tuli, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Abdullah, S. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Khanna, S. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Goldie, A. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 2.0 + Tuli, A. and Manning, C. D. co-authored the paper "Raptor: Recursive abstractive processing for tree-organized retrieval" + 833e7d67dcd30790b26b71c9b5306f6b + + + 1.0 + Yang, Z. and Manning, C. D. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Huang, M. and Duan, N. co-authored the paper "Enhancing retrieval-augmented large language models with iterative retrieval-generation synergy" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Xu, Y. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Su, D. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Yu, T. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Xu, Y. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Siddique, F. B. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Yu, T. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Siddique, F. B. and Barezi, E. J. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Siddique, F. B. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Barezi, E. J. and Fung, P. co-authored the paper "Caire-covid: A question answering and query-focused multi-document summarization system for covid-19 scholarly information management" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Tang, Y. and Yang, Y. co-authored the paper "MultiHop-RAG: Benchmarking retrieval-augmented generation for multi-hop queries" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Martin, L. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bhargava, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Touvron, H. and Bhosale, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Stone, K. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Albert, P. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Almahairi, A. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Babaei, Y. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Bashlykov, N. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Martin, L. and Batra, S. co-authored the paper "Llama 2: Open foundation and fine-tuned chat models" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Liang, Y. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Wang, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Meng, F. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Liang, Y. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Sun, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Meng, F. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Shi, H. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Sun, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 1.0 + Shi, H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035 + + + 2.0 + Li, Z. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Li, Z. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Li, Z. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Li, Z. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Li, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 2.0 + Xu, J. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Xu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Xu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 2.0 + Qu, J. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + 8d87efac8c50cf20cdf26bf61e5e2035,fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Qu, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + H. and Zhou, J. co-authored the paper "Is chatgpt a good nlg evaluator? a preliminary study" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Khramtsova, E. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. and Zhuang, S. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Khramtsova, E. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhuang, S. and Zuccon, G. co-authored the paper "Feb4rag: Evaluating federated search in the context of retrieval augmented generation" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Zhuang, S. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, S. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wang, Y. and Lipka, N. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Wang, Y. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Rossi, R. A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Lipka, N. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Siu, A. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Rossi, R. A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. and Zhang, R. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Siu, A. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zhang, R. and Derr, T. co-authored the paper "Knowledge graph prompting for multi-document question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Qi, P. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Zhang, S. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Bengio, Y. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Cohen, W. W. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Yang, Z. and Salakhutdinov, R. co-authored the paper "HotpotQA: A dataset for diverse, explainable multi-hop question answering" + fc4b27d64f055b7fc30176ba110dd02e + + + 1.0 + Zheng, L. and Chiang, W.-L. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zheng, L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Sheng, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Chiang, W.-L. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Wu, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Sheng, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Zhuang, Y. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Wu, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Lin, Z. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Zhuang, Y. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Li, D. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Lin, Z. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. and Xing, E. co-authored the paper "Judging llm-as-a-judge with mt-bench and chatbot arena" + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Li, D. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + 1.0 + Xing, E. is an author of the paper that discusses Chatbot Arena + b1bbda43309e8e0e2175ea034aa88e13 + + + \ No newline at end of file diff --git a/graphfleet/output/graphindex/artifacts/top_level_nodes.json b/graphfleet/output/graphindex/artifacts/top_level_nodes.json new file mode 100644 index 000000000..46d499633 --- /dev/null +++ b/graphfleet/output/graphindex/artifacts/top_level_nodes.json @@ -0,0 +1,736 @@ +{"id":"b45241d70f0e43fca764df95b2b81f77","x":null,"y":null} +{"id":"4119fd06010c494caa07f439b333f4c5","x":null,"y":null} +{"id":"d3835bf3dda84ead99deadbeac5d0d7d","x":null,"y":null} +{"id":"077d2820ae1845bcbb1803379a3d1eae","x":null,"y":null} +{"id":"3671ea0dd4e84c1a9b02c5ab2c8f4bac","x":null,"y":null} +{"id":"19a7f254a5d64566ab5cc15472df02de","x":null,"y":null} +{"id":"e7ffaee9d31d4d3c96e04f911d0a8f9e","x":null,"y":null} +{"id":"f7e11b0e297a44a896dc67928368f600","x":null,"y":null} +{"id":"1fd3fa8bb5a2408790042ab9573779ee","x":null,"y":null} +{"id":"27f9fbe6ad8c4a8b9acee0d3596ed57c","x":null,"y":null} +{"id":"e1fd0e904a53409aada44442f23a51cb","x":null,"y":null} +{"id":"de988724cfdf45cebfba3b13c43ceede","x":18.6878509521,"y":7.3225893974} +{"id":"96aad7cb4b7d40e9b7e13b94a67af206","x":9.4982213974,"y":14.8784036636} +{"id":"c9632a35146940c2a86167c7726d35e9","x":5.9654417038,"y":1.3233491182} +{"id":"9646481f66ce4fd2b08c2eddda42fc82","x":8.7502765656,"y":2.2784061432} +{"id":"d91a266f766b4737a06b0fda588ba40b","x":10.0678453445,"y":-3.8332984447} +{"id":"bc0e3f075a4c4ebbb7c7b152b65a5625","x":7.4744772911,"y":5.2438702583} +{"id":"254770028d7a4fa9877da4ba0ad5ad21","x":7.3455500603,"y":5.5255217552} +{"id":"4a67211867e5464ba45126315a122a8a","x":7.5212817192,"y":2.3584136963} +{"id":"04dbbb2283b845baaeac0eaf0c34c9da","x":6.331038475,"y":4.1710038185} +{"id":"1943f245ee4243bdbfbd2fd619ae824a","x":10.3827180862,"y":-1.8795469999} +{"id":"273daeec8cad41e6b3e450447db58ee7","x":16.7139911652,"y":8.9250364304} +{"id":"e69dc259edb944ea9ea41264b9fcfe59","x":7.568649292,"y":5.2975921631} +{"id":"e2f5735c7d714423a2c4f61ca2644626","x":8.8236865997,"y":1.3624777794} +{"id":"deece7e64b2a4628850d4bb6e394a9c3","x":16.2440032959,"y":-0.2270402759} +{"id":"e657b5121ff8456b9a610cfaead8e0cb","x":10.3789052963,"y":-4.7525854111} +{"id":"bf4e255cdac94ccc83a56435a5e4b075","x":10.5739068985,"y":-4.5526881218} +{"id":"3b040bcc19f14e04880ae52881a89c1c","x":8.4412469864,"y":6.6898360252} +{"id":"3d6b216c14354332b1bf1927ba168986","x":8.3578100204,"y":7.2702813148} +{"id":"1c109cfdc370463eb6d537e5b7b382fb","x":12.4210853577,"y":-19.2660865784} +{"id":"3d0dcbc8971b415ea18065edc4d8c8ef","x":6.2686991692,"y":11.8130121231} +{"id":"68105770b523412388424d984e711917","x":12.9253435135,"y":-19.7715568542} +{"id":"85c79fd84f5e4f918471c386852204c5","x":12.9709300995,"y":-19.8169384003} +{"id":"eae4259b19a741ab9f9f6af18c4a0470","x":12.6514644623,"y":-19.495885849} +{"id":"3138f39f2bcd43a69e0697cd3b05bc4d","x":15.6555585861,"y":-0.2303035855} +{"id":"dde131ab575d44dbb55289a6972be18f","x":15.1102590561,"y":-0.6151524186} +{"id":"de9e343f2e334d88a8ac7f8813a915e5","x":8.021531105,"y":-9.5909557343} +{"id":"e2bf260115514fb3b252fd879fb3e7be","x":8.4682445526,"y":-7.6289596558} +{"id":"b462b94ce47a4b8c8fffa33f7242acec","x":10.1844482422,"y":-3.3716790676} +{"id":"17ed1d92075643579a712cc6c29e8ddb","x":9.2791452408,"y":-2.1916351318} +{"id":"3ce7c210a21b4deebad7cc9308148d86","x":13.9476470947,"y":-1.0576649904} +{"id":"d64ed762ea924caa95c8d06f072a9a96","x":14.5025415421,"y":-1.229126811} +{"id":"adf4ee3fbe9b4d0381044838c4f889c8","x":13.5139389038,"y":-0.608899951} +{"id":"32ee140946e5461f9275db664dc541a5","x":12.9846420288,"y":-0.1456168294} +{"id":"c160b9cb27d6408ba6ab20214a2f3f81","x":13.1480007172,"y":0.0469199866} +{"id":"23527cd679ff4d5a988d52e7cd056078","x":14.6860551834,"y":-0.9547269344} +{"id":"f1c6eed066f24cbdb376b910fce29ed4","x":20.2195167542,"y":-0.6977658868} +{"id":"83a6cb03df6b41d8ad6ee5f6fef5f024","x":13.5842924118,"y":-1.4765719175} +{"id":"147c038aef3e4422acbbc5f7938c4ab8","x":6.5686979294,"y":10.9189186096} +{"id":"b7702b90c7f24190b864e8c6e64612a5","x":7.0128946304,"y":10.5659894943} +{"id":"de6fa24480894518ab3cbcb66f739266","x":7.1633963585,"y":10.4044294357} +{"id":"6fae5ee1a831468aa585a1ea09095998","x":12.8228502274,"y":-19.6692085266} +{"id":"ef32c4b208d041cc856f6837915dc1b0","x":13.945763588,"y":-1.7620911598} +{"id":"07b2425216bd4f0aa4e079827cb48ef5","x":20.632188797,"y":-0.9460879564} +{"id":"2670deebfa3f4d69bb82c28ab250a209","x":-5.9716773033,"y":2.8975236416} +{"id":"404309e89a5241d6bff42c05a45df206","x":-7.8244924545,"y":4.4370045662} +{"id":"b785a9025069417f94950ad231bb1441","x":-7.7471175194,"y":4.4961700439} +{"id":"3b6cd96a27304614850709aba1c9598b","x":-7.3820567131,"y":4.2585458755} +{"id":"d54956b79dd147f894b67a8b97dcbef0","x":-7.6798944473,"y":4.6415610313} +{"id":"958beecdb5bb4060948415ffd75d2b03","x":12.4357852936,"y":-19.281414032} +{"id":"b999ed77e19e4f85b7f1ae79af5c002a","x":7.2093577385,"y":10.2916297913} +{"id":"48c0c4d72da74ff5bb926fa0c856d1a7","x":8.8621921539,"y":-6.9418330193} +{"id":"4f3c97517f794ebfb49c4c6315f9cf23","x":8.2203025818,"y":-8.2812328339} +{"id":"1745a2485a9443bab76587ad650e9be0","x":12.6375141144,"y":-19.4837779999} +{"id":"32e6ccab20d94029811127dbbe424c64","x":7.2607922554,"y":10.8026103973} +{"id":"94a964c6992945ebb3833dfdfdc8d655","x":13.8970508575,"y":-1.653788209} +{"id":"1eb829d0ace042089f0746f78729696c","x":20.145822525,"y":-0.8447364569} +{"id":"015e7b58d1a14b44beab3bbc9f912c18","x":null,"y":null} +{"id":"26f88ab3e2e04c33a459ad6270ade565","x":null,"y":null} +{"id":"babe97e1d9784cffa1c85abc1e588126","x":null,"y":null} +{"id":"1033a18c45aa4584b2aef6ab96890351","x":7.1134057045,"y":12.1753330231} +{"id":"c9b8ce91fc2945b4907fe35519339cac","x":6.0801525116,"y":12.1497211456} +{"id":"fa3c4204421c48609e52c8de2da4c654","x":5.737970829,"y":11.7709150314} +{"id":"53af055f068244d0ac861b2e89376495","x":6.1288747787,"y":1.9010978937} +{"id":"c03ab3ce8cb74ad2a03b94723bfab3c7","x":13.7226495743,"y":-1.9063518047} +{"id":"ed6d2eee9d7b4f5db466b1f6404d31cc","x":null,"y":null} +{"id":"fc01e9baa80e417c9206f941bb279407","x":null,"y":null} +{"id":"56d0e5ebe79e4814bd1463cf6ca21394","x":null,"y":null} +{"id":"7c49f2710e8b4d3b8dc9310834406ea5","x":null,"y":null} +{"id":"c6d1e4f56c2843e89cf0b91c10bb6de2","x":null,"y":null} +{"id":"0adb2d9941f34ef7b2f7743cc6225844","x":null,"y":null} +{"id":"6b02373137fd438ba96af28f735cdbdb","x":null,"y":null} +{"id":"36a4fcd8efc144e6b8af9a1c7ab8b2ce","x":7.5496196747,"y":11.9959564209} +{"id":"fbeef791d19b413a9c93c6608286ab63","x":7.6088643074,"y":12.6738739014} +{"id":"d2b629c0396f4180a03e16ddf3818589","x":6.124127388,"y":11.6245079041} +{"id":"6102fc6619ed422ebc42588bfa97355d","x":5.9122271538,"y":12.0285387039} +{"id":"8d141c0b80f74b79a05eed7fe161fe49","x":7.7193078995,"y":10.4320163727} +{"id":"e22d1d1cd8d14f12b81828d940f40d70","x":7.6487746239,"y":10.0138368607} +{"id":"9ab48505fb1b487babd0d1f6d3a3f980","x":20.6168746948,"y":-0.1938226223} +{"id":"148fffeb994541b2b4b6dcefda7001a8","x":20.9376544952,"y":-0.2675468028} +{"id":"89c08e793298442686292454a1abff31","x":7.3022489548,"y":8.4898433685} +{"id":"0467928aa65e4a4fba62bdb1467e3a54","x":7.0959320068,"y":8.2336854935} +{"id":"43c3390303c6476cb65f584e37c3e81c","x":11.7299795151,"y":2.5142509937} +{"id":"fa14b16c17e3417dba5a4b473ea5b18d","x":12.5013046265,"y":1.8384599686} +{"id":"7cc3356d38de4328a51a5cbcb187dac3","x":10.372338295,"y":-1.1013549566} +{"id":"bef16fb5fd7344cca5e295b13ef3e0cd","x":8.6710567474,"y":7.2337183952} +{"id":"bb9e01bc171d4326a29afda59ece8d17","x":9.1079540253,"y":6.1263818741} +{"id":"3c063eea52e94164b70c99431ea30bae","x":8.1956634521,"y":2.7720737457} +{"id":"252cc8452bfc4c2aa58cab68d8b61879","x":9.4469690323,"y":3.6355266571} +{"id":"7e2c84548fb94ee395ba8588d8f2a006","x":8.874083519,"y":4.2797689438} +{"id":"f034618dde7948beb6dab30176d0fc87","x":7.6778812408,"y":3.2551436424} +{"id":"5c41f96be13e49dba649454297834546","x":9.0166501999,"y":3.2908613682} +{"id":"7ea4afbf8a264f29af29950ce98105ba","x":6.3879308701,"y":1.3283433914} +{"id":"91ff849d12b24574b0691dbddf44968b","x":6.9506235123,"y":-0.1251165718} +{"id":"d73c1f2fb3094d8dace42ad2a76e9a52","x":6.7397451401,"y":3.0836231709} +{"id":"cdc8901e668749889bd49bebdc4ff1f6","x":10.6651353836,"y":-4.0287017822} +{"id":"36084a9fab53433493f079e97e68bf65","x":10.1729316711,"y":3.7292821407} +{"id":"eebcc7ec8e3e4df7aea83659bbdc2199","x":8.7341938019,"y":4.5519070625} +{"id":"ceadf262ef834e9ab146b20650912cae","x":9.6401777267,"y":-4.7224192619} +{"id":"7f65feab75424b53b24470d305ba331a","x":9.6941070557,"y":-4.7777142525} +{"id":"fd9cb733b28d420cb5cef01e545a132c","x":7.7190909386,"y":4.4298920631} +{"id":"0fbcca3f17c649a08aea64b5a7d9ef36","x":7.2919359207,"y":4.8076348305} +{"id":"482027a59f32484c9c44fd700615c1b6","x":8.5062437057,"y":3.7735004425} +{"id":"de837ff3d626451282ff6ac77a82216d","x":9.8931121826,"y":4.0732960701} +{"id":"460295fed3ae4cd39f9f274cec9c2506","x":9.0467538834,"y":2.9402649403} +{"id":"553b285bba60460ab1ed8341ae61282b","x":9.4724845886,"y":0.9986838698} +{"id":"cec95bf17e7e4c939b56c9c6f402a29f","x":9.1924104691,"y":1.2911343575} +{"id":"599164aead034bc19446efacc77554d2","x":8.1298570633,"y":3.1723437309} +{"id":"bbf148ae4d48422f8fdef754cfa2b9e4","x":11.4409227371,"y":-4.5197863579} +{"id":"de61b2670999433f807a6a1dc2b81e43","x":12.3119592667,"y":4.6016364098} +{"id":"3e95dacfe57b4d57b5da4310ef2e157f","x":15.769947052,"y":0.3212119341} +{"id":"1f1545308e9347af91fd03b94aadc21f","x":15.8499650955,"y":0.1924337149} +{"id":"6ea81acaf232485e94fff638e03336e1","x":15.4041070938,"y":0.4885048866} +{"id":"d136b08d586d488f9e4188b524c85a29","x":8.5173873901,"y":13.4037303925} +{"id":"cccfa151fedc4b218a8d96adc7dceabe","x":8.7835493088,"y":16.5811271667} +{"id":"ce54725672a74ebcabe6127577dacb2b","x":6.8737163544,"y":8.4750709534} +{"id":"ea2b28ca1a974ffab4517811dc1d1e5c","x":9.6368637085,"y":2.6183474064} +{"id":"aff21f1da1654e7babdcf3fb0e4a75fc","x":9.2778816223,"y":4.4500360489} +{"id":"dc2cc9016e3f49dbac7232f05cce794d","x":8.1217718124,"y":15.4246225357} +{"id":"6ea0cef05f694dcea455478f40674e45","x":15.3389492035,"y":0.0953150019} +{"id":"7ab5d53a872f4dfc98f3d386879f3c75","x":8.6676435471,"y":16.4780158997} +{"id":"af1d0fec22114a3398b8016f5225f9ed","x":8.4686231613,"y":14.9649868011} +{"id":"b07a7f088364459098cd8511ff27a4c8","x":15.8907680511,"y":0.4980384111} +{"id":"8870cf2b5df64d2cab5820f67e29b9f1","x":16.1501293182,"y":0.2773395777} +{"id":"cd130938a2844050be991af70baf5ee0","x":8.5292339325,"y":13.5534315109} +{"id":"43544b99c3b04b059546198a0ae6366d","x":8.6486320496,"y":13.7581682205} +{"id":"a671bf7fea2f4514b6e96ba99127fafd","x":-3.4649548531,"y":-4.3357319832} +{"id":"525f41ea20274a05af4e52b625b473f3","x":-3.7228152752,"y":-4.1975355148} +{"id":"071a416efbec4f0886c19ac68f6d43cb","x":8.4629621506,"y":15.7238321304} +{"id":"6d8473ef3b1042bf87178a611e3dbcc6","x":8.426612854,"y":16.0815620422} +{"id":"30c9641543c24773938bd8ec57ea98ab","x":8.2019062042,"y":-8.6085538864} +{"id":"18b839da898e4026b81727d759d95c6a","x":8.3969593048,"y":-8.6477441788} +{"id":"eeef6ae5c464400c8755900b4f1ac37a","x":8.1723451614,"y":-9.1721544266} +{"id":"422433aa45804c7ebb973b2fafce5da6","x":8.6411552429,"y":15.4889497757} +{"id":"86505bca739d4bccaaa1a8e0f3baffdc","x":8.2488765717,"y":15.8081693649} +{"id":"1af9faf341e14a5bbf4ddc9080e8dc0b","x":7.917757988,"y":14.1830978394} +{"id":"353d91abc68648639d65a549e59b5cf3","x":7.973361969,"y":13.6083335876} +{"id":"7ce637e4f35b42e3a9f8272cab69cd22","x":8.0209789276,"y":13.5623989105} +{"id":"4d999d7744b04a998475f8f8531589f0","x":-3.3145289421,"y":-3.6454744339} +{"id":"9a6f414210e14841a5b0e661aedc898d","x":-3.8749907017,"y":-3.9412593842} +{"id":"db541b7260974db8bac94e953009f60e","x":-3.9556856155,"y":-3.8994352818} +{"id":"f2ff8044718648e18acef16dd9a65436","x":-3.7088928223,"y":-3.9251067638} +{"id":"00d785e7d76b47ec81b508e768d40584","x":-3.8289749622,"y":-3.5053477287} +{"id":"87915637da3e474c9349bd0ae604bd95","x":-4.2378511429,"y":-3.7274646759} +{"id":"8f1eba29f39e411188200bf0d14628ec","x":null,"y":null} +{"id":"7282c73622b8408e97289d959faff483","x":15.2195539474,"y":4.7894220352} +{"id":"3deb220d31f74103aa44870a36a63220","x":11.8790330887,"y":3.8985233307} +{"id":"af7a1584dd15492cb9a4940e285f57fc","x":null,"y":null} +{"id":"6e8d9029ce4e4ea182367173ab2c7bbf","x":null,"y":null} +{"id":"cbf232211e7d4eb6abdbe182f71c2cf0","x":null,"y":null} +{"id":"bb0cff774a4440b289cc6f3b929fe13c","x":7.6129412651,"y":-5.1423573494} +{"id":"ce55841ebfdd47008bab8c258f10372e","x":7.490228653,"y":-4.6051912308} +{"id":"6090e736374d45fd84f0e4610a314f8f","x":6.2765789032,"y":-11.3484268188} +{"id":"0e8d921ccd8d4a8594b65b7fd19f7120","x":21.3102836609,"y":-0.4946330488} +{"id":"59c726a8792d443e84ab052cb7942b4a","x":21.1255569458,"y":-0.5498127341} +{"id":"4f2c665decf242b0bfcaf7350b0e02ed","x":13.0853738785,"y":7.6657042503} +{"id":"66cdf168f36d4a57a505028c97dc06e0","x":10.3675880432,"y":-2.6679039001} +{"id":"38f51478f41f48db9bee570859b6f43e","x":7.472237587,"y":-4.8778424263} +{"id":"896d2a51e8de47de85ba8ced108c3d53","x":7.4398751259,"y":-5.4202651978} +{"id":"14555b518e954637b83aa762dc03164e","x":7.5513091087,"y":-5.2735309601} +{"id":"b1f6164116d44fe8b8f135d7f65b9e58","x":8.5102558136,"y":-6.6555190086} +{"id":"c8b2408617804483b620e1a6691ac90d","x":8.3684368134,"y":-6.3442425728} +{"id":"a5e0d1644eb547ba9a5c3211aac4631a","x":8.0968885422,"y":-6.078877449} +{"id":"5a28b94bc63b44edb30c54748fd14f15","x":7.9879498482,"y":-5.9166789055} +{"id":"f97011b2a99d44648e18d517e1eae15c","x":20.6092243195,"y":-0.8742604256} +{"id":"35489ca6a63b47d6a8913cf333818bc1","x":6.6857504845,"y":-11.4462814331} +{"id":"5d3344f45e654d2c808481672f2f08dd","x":6.6227312088,"y":-11.1960849762} +{"id":"6fb57f83baec45c9b30490ee991f433f","x":6.8509364128,"y":-10.9554796219} +{"id":"68762e6f0d1c41cd857c6b964a8e76c3","x":7.5318922997,"y":-9.7246751785} +{"id":"70634e10a5e845aa8c6a32fe7e8eb2b2","x":7.5641183853,"y":-10.0437803268} +{"id":"04085f7cf46544b79597fc49286ff84d","x":7.1315488815,"y":-10.463558197} +{"id":"d203efdbfb2f4b2a899abfb31cf72e82","x":10.1866779327,"y":-4.9914851189} +{"id":"6731a665561840c2898ce8c9788e4c88","x":7.4429445267,"y":-5.335588932} +{"id":"4026806fa92f4e849a59a7f5c9a45c79","x":10.2374830246,"y":-5.0434689522} +{"id":"68e0c60d2e8845d89d9d0ad397833648","x":11.3818302155,"y":-4.1388726234} +{"id":"101572f552b54e529fe7765c05168981","x":10.828918457,"y":-3.838596344} +{"id":"60c58026b2764b40adffca6eaa31d6d9","x":9.85435009,"y":-2.1369416714} +{"id":"ad1595a78935472999444c9330e7730e","x":10.4193029404,"y":-4.2753653526} +{"id":"735d19aea0744b2295556841c5c4c3fd","x":null,"y":null} +{"id":"c725babdb14a485582f8fbdf95429030","x":null,"y":null} +{"id":"a0047221896d418d849847d422fa4bb8","x":null,"y":null} +{"id":"98fc2ee593184c5a839454db4eec7013","x":7.99893713,"y":-8.9205131531} +{"id":"80020a1da63042459e00266b2a605452","x":11.0422639847,"y":-4.4528822899} +{"id":"31a7e680c4d54101afe4c8d52d246913","x":7.9817881584,"y":-4.8735117912} +{"id":"351abba16e5c448994c6daf48121b14d","x":10.9834680557,"y":-3.2228424549} +{"id":"50ea7d3b69614bcdbfbff7ddbfbf3d34","x":10.5759019852,"y":-2.9143710136} +{"id":"004f40a5aeca48a1879db728eb12bcba","x":-12.0172700882,"y":11.2135028839} +{"id":"4465efb7f6ed4dedad72a658184addd2","x":10.9046640396,"y":27.386138916} +{"id":"b0dd60e11dad4ff782623acf039b3948","x":-12.0538930893,"y":11.1783771515} +{"id":"db8c43fa4df947b09e5754d3b1393ead","x":-11.5990009308,"y":11.6333341599} +{"id":"5dabc4cd05da425cb194a04482bf0c29","x":-12.0954084396,"y":11.1368894577} +{"id":"9d08f285a7be4c79b8f359c51d51db37","x":-11.6815767288,"y":11.5506706238} +{"id":"adffed660d154b519c1817e514e83096","x":-12.1827030182,"y":11.0490589142} +{"id":"b7e9c9ef572c445a9574ca571e41fb96","x":11.5482501984,"y":26.7427101135} +{"id":"dcb9f281cd6248c699e0ebb285a42a5e","x":10.8825950623,"y":27.408367157} +{"id":"072cdee531b74513984f49d99a8d64a0","x":11.1809387207,"y":27.1101665497} +{"id":"5ae335d9210a45fda3f92a9a028d6d9b","x":11.2498083115,"y":27.0411720276} +{"id":"5ac60a941a5b4934bdc43d2f87de601c","x":11.2533082962,"y":27.0377502441} +{"id":"d405c3154d0e48ce96fad4c28fe20590","x":9.9092855453,"y":-2.4518845081} +{"id":"7923d8521c744bd9aab131c1aea91ffd","x":11.3254203796,"y":-3.6214504242} +{"id":"5bd156c87ec44e19ae6f8f62e6e50b9d","x":9.4222230911,"y":-2.1544210911} +{"id":"c1a146d7fb16429ea6d0aa2a55ee597f","x":10.96281147,"y":3.1404595375} +{"id":"ede9350632084da5b0b577ff799ab14b","x":10.8953580856,"y":2.2660200596} +{"id":"ed559fb4ebde45518849ec803b350fa3","x":13.4801588058,"y":7.374712944} +{"id":"f422035f8b78417f98e4d116971cf9f3","x":13.5035228729,"y":7.4014849663} +{"id":"c79d686eba044c5586c706cdc096817d","x":13.0235424042,"y":7.6441159248} +{"id":"0f70db1e598d463fbbcdd1e288bd9490","x":-11.1498603821,"y":1.3169480562} +{"id":"b35c3d1a7daa4924b6bdb58bc69c354d","x":null,"y":null} +{"id":"a97e2ecd870944cfbe71c79bc0fcc752","x":11.8509540558,"y":1.7040245533} +{"id":"3e1b063bbfa9423d84e50311296d2f3c","x":null,"y":null} +{"id":"9a8ce816ee954bdabd01ea2081538009","x":null,"y":null} +{"id":"09f18f81442d4d6d93a90f0fac683f9b","x":6.5547347069,"y":-11.5148382187} +{"id":"e02be3e37ca0454883a4c1fd859c24bb","x":7.0462579727,"y":8.2149295807} +{"id":"6e0c81bef5364c988b21bf9b709d9861","x":-11.2842254639,"y":0.9239032269} +{"id":"1dbc51475cb04dafa4a8833a8378635e","x":null,"y":null} +{"id":"c12b9ebd8b4e42b7896822a32e3fa6eb","x":null,"y":null} +{"id":"27505f6ade4b4e5f9316ffe9c34821f7","x":null,"y":null} +{"id":"0ee7db2c6bea4630ba9f0c25e8a967ad","x":null,"y":null} +{"id":"5a6c1d15424149f69052cd8d91fbff75","x":null,"y":null} +{"id":"d005bf75c31d4848ad7041f39651e59c","x":null,"y":null} +{"id":"9b3eef8f3a3a45e6873838db95295b8a","x":12.5564374924,"y":8.1180391312} +{"id":"fdc954b454744820804d7798f3e0b5de","x":10.4003171921,"y":-0.0966932327} +{"id":"49c1383836934ec495c3b35769100a73","x":10.4374866486,"y":0.9379460216} +{"id":"859dedcc3736439a8a563419f16cb3d8","x":12.3475551605,"y":1.7517055273} +{"id":"6078b9980a6c4dcd9198d151b833ead7","x":12.3501777649,"y":2.2055008411} +{"id":"f93cd6b8213e46dda67af7e5382e1bd2","x":12.1904201508,"y":2.5731811523} +{"id":"496f17c2f74244c681db1b23c7a39c0c","x":12.0817241669,"y":2.1693849564} +{"id":"da1684437ab04f23adac28ff70bd8429","x":11.8501968384,"y":1.9139729738} +{"id":"4517768fc4e24bd2a790be0e08a7856e","x":10.888756752,"y":1.4466644526} +{"id":"545edff337344e518f68d1301d745455","x":9.5926237106,"y":-2.4531729221} +{"id":"9376ce8940e647a99e5e087514b88fa4","x":12.8688344955,"y":0.4367204309} +{"id":"b38a636e86984600bb4b57c2e2df9747","x":12.9344234467,"y":0.751052916} +{"id":"4bc7440b8f4b4e4cae65a5c49defa923","x":13.340965271,"y":7.1130313873} +{"id":"5d1b038ce8be4533b54dd79d6496de9b","x":15.5612277985,"y":4.5652532578} +{"id":"ac6e5a44e0c04a4fa93589376fde4c34","x":15.5780467987,"y":4.5048666} +{"id":"40e4ef7dbc98473ba311bd837859a62a","x":12.6663885117,"y":8.0187654495} +{"id":"222f0ea8a5684123a7045986640ec844","x":10.1186971664,"y":0.3710429668} +{"id":"668cf1fdfd644d39acc6350b86117ea2","x":12.4757804871,"y":2.4581239223} +{"id":"478e4c72d8fb46dd8cc9f0691c9878fd","x":12.1899585724,"y":4.4049186707} +{"id":"82b0446e7c9d4fc793f7b97f890e9049","x":7.9968628883,"y":8.4220905304} +{"id":"8169efeea3ce473d9fd2f1c688126a1c","x":12.3670730591,"y":0.5278597474} +{"id":"c2d48b75af6a4d7989ccf9eceabd934e","x":16.4724082947,"y":4.1250844002} +{"id":"5f1fc373a8f34050a5f7dbd8ac852c1b","x":15.7823009491,"y":4.4424171448} +{"id":"0c010fa3aeac4b28b2fbb8c2339c2521","x":null,"y":null} +{"id":"c2999bdca08a478b84b10219875b285e","x":null,"y":null} +{"id":"263d07354a1b4336b462024288f9bcd3","x":14.829451561,"y":-13.7789487839} +{"id":"f9005e5c01b44bb489f7112322fd1162","x":8.4622936249,"y":6.4865961075} +{"id":"d9ef017549724f4fbc4ff4ba6701dac0","x":14.4984455109,"y":-14.0333385468} +{"id":"33b9e826af3f43838c07c847b6349497","x":9.3639860153,"y":16.0032749176} +{"id":"dbe9063124d047dc8d6fcaeadcda038f","x":18.8840579987,"y":11.2171592712} +{"id":"c885166d0c454a748376b56279f96408","x":19.3415679932,"y":10.5153055191} +{"id":"586bccefb1e344289c1ee984e165de9c","x":18.8092727661,"y":11.0580530167} +{"id":"a2201b8753ba4847ab0b22054e27d2c0","x":16.4479980469,"y":4.7523179054} +{"id":"b5ecd0553dd742f5813c9b855d548a41","x":16.8326950073,"y":4.4692921638} +{"id":"89b2003e97804961805ea1886d078ebd","x":16.6025733948,"y":4.5357551575} +{"id":"6dd7f5f6b4544271a97f6a136f82fc3d","x":17.0270385742,"y":4.7359013557} +{"id":"eb01db8435554f2cbafe39a50f62f20a","x":16.3862113953,"y":4.6762862206} +{"id":"3d175ad1f0014cd4871eff4e86db9f88","x":null,"y":null} +{"id":"c8e706fbdc90420d952deed03c4f04b4","x":null,"y":null} +{"id":"cf6115e69d6649cc99ef2bd11854ccfb","x":null,"y":null} +{"id":"9ed7e3d187b94ab0a90830b17d66615e","x":9.5399494171,"y":3.5461647511} +{"id":"b4c7432f712849d7aba9dccbb77471ef","x":16.9714012146,"y":5.5094804764} +{"id":"434e752b992c4e6a812557529315c5b9","x":17.0225868225,"y":5.8992295265} +{"id":"df79a27b9a4f42fd839c90bb8a79ad91","x":14.4201602936,"y":-14.1847038269} +{"id":"8f140fd7126f47b6b00307b0181509f9","x":16.8164863586,"y":6.7105836868} +{"id":"40450f2c91944a81944621b94f190b49","x":14.8968734741,"y":-13.9669647217} +{"id":"5b9fa6a959294dc29c8420b2d7d3096f","x":14.792974472,"y":-13.7484903336} +{"id":"b84d71ed9c3b45819eb3205fd28e13a0","x":14.399266243,"y":-14.4081230164} +{"id":"b0b464bc92a541e48547fe9738378dab","x":14.3889293671,"y":-14.0776290894} +{"id":"44c65dda6fb7472dae36f6eea720ab47","x":20.0700702667,"y":10.0296630859} +{"id":"5d97ff82691c4482973d73d1860e4757","x":20.1038837433,"y":9.5098457336} +{"id":"2567445079794d1e84f17abc48776002","x":19.8126678467,"y":10.0125865936} +{"id":"392be891f8b649fabdc20e7bf549f669","x":19.798707962,"y":9.6621608734} +{"id":"0111777c4e9e4260ab2e5ddea7cbcf58","x":18.1092853546,"y":13.8273820877} +{"id":"785f7f32471c439e89601ab81c828d1d","x":17.6408576965,"y":12.9609260559} +{"id":"6768339b54084020aec27adcef8994ff","x":17.832780838,"y":13.6822004318} +{"id":"f09f381c319f4251847d1a4bb8cdcac1","x":18.3368186951,"y":14.1290502548} +{"id":"eec11f567e7f4943b157c3a657eb9a46","x":17.9658279419,"y":13.2721042633} +{"id":"efef117839b64ce9adf614a461d41ba6","x":8.6846418381,"y":5.737988472} +{"id":"2171091ada0942d8ae7944df11659f6e","x":8.9258985519,"y":16.030544281} +{"id":"bcfdc48e5f044e1d84c5d217c1992d4b","x":17.917016983,"y":11.9641542435} +{"id":"b232fb0f2ac14790b931d1e7fcddd8ad","x":17.6870975494,"y":11.6105337143} +{"id":"1c16b22e18d3483b8d41b284754274e2","x":19.3580608368,"y":10.6510066986} +{"id":"0080f96708cd4054a5f0986ca86889f4","x":18.5859279633,"y":10.8156290054} +{"id":"e683130322ac47708a852a5e51abb7c5","x":18.272687912,"y":11.5834732056} +{"id":"71a0a8c1beb64da08124205e9a803d98","x":18.0239524841,"y":12.6254234314} +{"id":"f84314943bee4c859c9a62f268c9c216","x":17.5952720642,"y":13.0616769791} +{"id":"ba481175ee1d4329bf07757a30abd3a1","x":18.2132453918,"y":13.9680376053} +{"id":"8d8da35190bf43c5878fa38f3eb4f3d2","x":17.800151825,"y":12.3543014526} +{"id":"2fb7e14a3f124526bd7b24867fc18e81","x":18.5599079132,"y":11.2735147476} +{"id":"5c13c7d61e6c4bfe839f21e7ad3530a7","x":18.6896209717,"y":10.5633087158} +{"id":"a621663edba64d99b7e50f1e53f32ee7","x":17.8283081055,"y":9.9390106201} +{"id":"42be4e140061482ea509dd3e26189480","x":11.2955770493,"y":1.7941998243} +{"id":"4da4ef951ff340f1a3dd679de4be3341","x":18.2015914917,"y":11.400967598} +{"id":"2f05fcce857e4a499ca4e89a3cefbcb3","x":17.6205272675,"y":11.2590885162} +{"id":"b3aeb7ae009a4f52ae3ae4586e32fe11","x":17.7774162292,"y":10.7299156189} +{"id":"089b9b9841714b8da043777e2cda3767","x":10.2635793686,"y":2.0356020927} +{"id":"38f1e44579d0437dac1203c34678d3c3","x":10.5088071823,"y":1.5419698954} +{"id":"1ca24718a96b47f3a8855550506c4b41","x":13.1635742188,"y":7.5203948021} +{"id":"9c980dfe3cab44b7a83408405edab0b6","x":-15.8562936783,"y":8.7750167847} +{"id":"f23484b1b45d44c3b7847e1906dddd37","x":-15.9518795013,"y":8.8674688339} +{"id":"929f30875e1744b49e7b416eaf5a790c","x":-15.6512432098,"y":8.5662336349} +{"id":"4920fda031804ce8a1073ace8e061ed6","x":-15.6382474899,"y":8.5542888641} +{"id":"4b8aa4587c7344adac2cbfa69d5e40fa","x":-15.5927658081,"y":8.5215501785} +{"id":"52701d941dfb45359693baae8f267056","x":7.9470171928,"y":7.7608885765} +{"id":"31499ee6277a4d71b19cb5b6be554c69","x":8.2998952866,"y":7.838136673} +{"id":"d99eabad5dfd47278692569d2a9395b1","x":11.7973594666,"y":0.2062926441} +{"id":"d53f15cb7f7845de91cc44ad44ff9f6e","x":11.9850711823,"y":0.1631195545} +{"id":"23becf8c6fca4f47a53ec4883d4bf63f","x":8.5405006409,"y":6.590965271} +{"id":"d0ffa3bcd1234258953ff4956d19f561","x":12.8176622391,"y":1.0516904593} +{"id":"ac41b77ba33c4c84877eb425aba03aa1","x":11.325881958,"y":2.8742239475} +{"id":"5d3184dabfd647a5a7e565f72c60ff24","x":12.9636697769,"y":1.3083833456} +{"id":"0ec262c2cfef4dd581f3655e5e496e31","x":10.0309734344,"y":2.9207353592} +{"id":"100c2fccd7f74d9281707082f062ba72","x":11.295290947,"y":3.5417804718} +{"id":"378fc7636eeb4aabbfd40995a6960c64","x":6.7602396011,"y":0.4677597582} +{"id":"80a04aa18cd649d584292f23b10c0727","x":16.8261032104,"y":8.5943689346} +{"id":"4e9ca18ccc1d4527a3bc035d07f5e162","x":9.5091667175,"y":6.0825176239} +{"id":"5564257e89f1428486a64fcf52f49490","x":11.4330348969,"y":1.0503296852} +{"id":"83c76fbd2a004d90a5b0a6736ffed61d","x":8.5658340454,"y":2.9363348484} +{"id":"d9779c41e3c74fe0b26e23822a4b995b","x":9.3409328461,"y":4.5817985535} +{"id":"9d7a563b3b2d405092c31f1fe08cff77","x":8.3611249924,"y":16.6468238831} +{"id":"bd43f3d439a54781bd4b721a9a269b92","x":9.2976675034,"y":3.7600119114} +{"id":"adc0f95733e74351a891c4dadf650a52","x":null,"y":null} +{"id":"225105a7be14447cb03186bd40756059","x":null,"y":null} +{"id":"efce8a9d61254447a26aee99e53f0398","x":17.146572113,"y":7.6836538315} +{"id":"4a75a9f0b18a48bea9c0601c0fc395c4","x":5.7041563988,"y":2.6790053844} +{"id":"e19287afe00a431f9a593a4827d1b448","x":5.900416851,"y":3.1417694092} +{"id":"f2c06f3a0c704296bf3353b91ee8af47","x":7.7464866638,"y":2.1043095589} +{"id":"f512103ed4624accac6cbbf90d7d250a","x":6.1949729919,"y":1.9623711109} +{"id":"2325dafe50d1435cbee8ebcaa69688df","x":5.6093215942,"y":2.0680634975} +{"id":"469aeef98cd1421fa123277b93d7b83a","x":19.8753490448,"y":-4.219019413} +{"id":"2fb66f9a0de6406d83b61742a3b52cd6","x":5.6556725502,"y":1.6665394306} +{"id":"b0e6cfd979ea48b997019b059999d3c2","x":5.523478508,"y":1.4942002296} +{"id":"ef00ec3a324f4f5986141401002af3f6","x":7.8821263313,"y":2.6966621876} +{"id":"a542fd7aed7341468028928937ea2983","x":9.0619478226,"y":14.6377372742} +{"id":"1c5e296a5ac541c1b5cac4357537c22d","x":9.5715303421,"y":15.0845317841} +{"id":"5ecf534a9ffe46e0b1c2144110c691c0","x":9.2646389008,"y":15.1881198883} +{"id":"4d183e7007624fcd98af96b9d752c16d","x":8.7806882858,"y":2.7121636868} +{"id":"718c507cb8ac49e6a35c251ac951b5ca","x":7.5333352089,"y":2.4474527836} +{"id":"b45ef27279c043269b23b894461d7d8c","x":7.3321523666,"y":2.943775177} +{"id":"10983a248cc448c59c94df4d1d0898f0","x":6.6826682091,"y":1.5992435217} +{"id":"e2ec7d3cdbeb4dd086ae6eb399332363","x":6.6238098145,"y":2.6979117393} +{"id":"67f10971666240ea930f3b875aabdc1a","x":5.94201231,"y":2.3290472031} +{"id":"8b95083939ad4771b57a97c2d5805f36","x":6.7749390602,"y":2.3810799122} +{"id":"3c4062de44d64870a3cc5913d5769244","x":6.4403324127,"y":1.7907252312} +{"id":"24652fab20d84381b112b8491de2887e","x":7.0881075859,"y":2.0311043262} +{"id":"d4602d4a27b34358baa86814a3836d68","x":7.4207558632,"y":2.735124588} +{"id":"36be44627ece444284f9e759b8cd25c6","x":9.9741191864,"y":14.7055273056} +{"id":"a64b4b17b07a44e4b1ac33580d811936","x":9.6824121475,"y":14.6451234818} +{"id":"423b72bbd56f4caa98f3328202c1c3c9","x":9.4494400024,"y":15.1425065994} +{"id":"5c7ef01f46a94641bf1ae5cd25f8a538","x":5.6141166687,"y":2.3556201458} +{"id":"aefde1f7617f4c0e9aed31db77f6d862","x":19.590133667,"y":-3.9390347004} +{"id":"ad52ba79a84748a49067e53b1d5095f9","x":20.0881175995,"y":-4.4340319633} +{"id":"289616058bf4495887292003b27ba216","x":19.7248840332,"y":-4.0869522095} +{"id":"7ffa3a064bce468082739c5a164df5a3","x":null,"y":null} +{"id":"ce36d1d637cf4a4e93f5e37ffbc6bd76","x":null,"y":null} +{"id":"eeb9c02c0efa4131b9e95d33c31019fc","x":null,"y":null} +{"id":"7b2472c5dd9949c58828413387b94659","x":null,"y":null} +{"id":"bdddcb17ba6c408599dd395ce64f960a","x":null,"y":null} +{"id":"bc70fee2061541148833d19e86f225b3","x":18.3320064545,"y":7.4575948715} +{"id":"0fc15cc3b44c4142a770feb4c037a6f7","x":17.3951377869,"y":8.3325119019} +{"id":"a24e9df02e1b4b43bf6324b039e28285","x":18.0267448425,"y":7.6458668709} +{"id":"ab3a5a6713244fd595a1ace978c3d960","x":17.5528488159,"y":7.8847327232} +{"id":"02a88c0d128e4586b2f1f64329786d3c","x":17.6920032501,"y":6.4137382507} +{"id":"1ca41537c47c4752a17a44d1d7086d96","x":17.5065040588,"y":6.3203043938} +{"id":"7e0d14ca308b4796bdc675a64bd3a36e","x":17.7184944153,"y":7.7127637863} +{"id":"8323efc8e539419e9ca3c98e758f6609","x":18.3199768066,"y":7.7212734222} +{"id":"a80c7c98c0b647f8b9f6f8cc09168e44","x":17.2645797729,"y":6.9621787071} +{"id":"2d66a15939294d21b83b3e277f0a4e46","x":17.1422786713,"y":7.3766274452} +{"id":"47f6d6573cf34e1096c95e36251dd60c","x":18.0669403076,"y":6.5817828178} +{"id":"2fbd74d5ccca4be99c5257b3ac95cfba","x":18.1342144012,"y":7.0624470711} +{"id":"a2b1621a3e424ae29a6a73f00edbeca3","x":16.885055542,"y":7.7954030037} +{"id":"ec45e1c400654c4f875046926486ded7","x":-5.4275560379,"y":21.2119579315} +{"id":"047cd93e9d704c7d8dadb6e79f9458df","x":-4.5389995575,"y":21.3071041107} +{"id":"5b71ee73a5b6484495b2a0a75219426c","x":-4.949657917,"y":21.6474552155} +{"id":"e1f524d4b9754ce2b64a0a4c8f73b854","x":-4.9982962608,"y":21.2816867828} +{"id":"ae1fe1c014c54ec4bcdf10dbdaed5068","x":-4.4568037987,"y":21.3807640076} +{"id":"92646910ee624bd7909fac2b5c0232e3","x":-4.3771481514,"y":21.5225963593} +{"id":"05913bee89a94bca88449249e35ba74d","x":null,"y":null} +{"id":"57b8930790c34dcba4a32c6be703ed78","x":null,"y":null} +{"id":"838c4498bc3c437f8d65428b580766a2","x":16.6706752777,"y":6.2300262451} +{"id":"1b893f24eb98477aad6ce49c0f26737e","x":17.1266651154,"y":7.2002148628} +{"id":"6573bc2af4f94596a3f4452a602d6fc4","x":17.6128311157,"y":7.196750164} +{"id":"0dddcca0e5df4b16bc03a51a2d2d8e16","x":17.7456989288,"y":7.0899457932} +{"id":"df40ad480a3c47299a6c8fad05349304","x":17.3777217865,"y":8.274851799} +{"id":"fe98fb197d294b0b837aee8d5a98dfb1","x":17.9317893982,"y":6.877781868} +{"id":"feb9ddd0ac2949178f26a36949aa5422","x":18.0033397675,"y":7.9679112434} +{"id":"b4e4fa2e3dfc46e68d532d659b18d17d","x":18.0467624664,"y":8.0781259537} +{"id":"f58813d090b947a48c1b4614b92c3ec3","x":null,"y":null} +{"id":"30a251bc3d04430d82b5a1a98c7b8c75","x":6.8284087181,"y":0.8472428322} +{"id":"93e1d19f9bfa4c6b8962d56d10ea9483","x":null,"y":null} +{"id":"8046335ba70b434aa3188392a746fd78","x":null,"y":null} +{"id":"5c02b1ab32064c64a0f8b27b219e358a","x":8.6036672592,"y":3.8665347099} +{"id":"c5f77ba0c261408780db3d50346f16b7","x":10.0777873993,"y":2.2445056438} +{"id":"453ecf5476f64f4a8d5020b95baf1314","x":7.2367253304,"y":4.0829496384} +{"id":"6a1d83c9ce2b483dbd7de5ab3ae2487d","x":7.1147413254,"y":4.4984660149} +{"id":"66c3dffb7d7a4fa8bb6b48a22ca917a6","x":6.0575084686,"y":3.6949422359} +{"id":"6f3dd1fd6d7f4df4af0656ed0525c92e","x":17.7945346832,"y":9.5636339188} +{"id":"711eb39432794b0a91110358dd536517","x":17.4612560272,"y":9.7223653793} +{"id":"0e00585b08044954a254116665400463","x":6.515668869,"y":4.8572230339} +{"id":"db0147eff2204a20b5e5e6bec7a8bae5","x":9.5290527344,"y":3.2889456749} +{"id":"67bb4f4678284819add02ba04f3b1103","x":null,"y":null} +{"id":"2033ec0487f04240abb3bdbe77b39087","x":12.0219087601,"y":4.233361721} +{"id":"f026fab8fec948ae9e7baa2ad715e6ef","x":11.7886981964,"y":4.3207683563} +{"id":"d0d7ed36d6f54b5d986dfd854096b728","x":12.1996231079,"y":4.1498856544} +{"id":"bf6a4c18f44042799eb7456a6b85b54a","x":9.2177696228,"y":3.71022439} +{"id":"fac4a59c2278498d83f9f1b4231ad62e","x":10.1304759979,"y":2.954236269} +{"id":"d6d2b5862ddc4c4d87deee3423506817","x":7.6431255341,"y":4.6975774765} +{"id":"47d588d26e2b4cccb68fe2af4c147c8f","x":6.9845056534,"y":4.7900419235} +{"id":"c0f2dc03d8df400db4997c1a0babd6ad","x":6.4602637291,"y":4.0095019341} +{"id":"0211d61aae834229a3a1e004ff5cc658","x":6.4358196259,"y":4.2524733543} +{"id":"ccbbbcc055c34709abcf103208c2c299","x":11.6083316803,"y":-3.8404254913} +{"id":"989add81cf874018a569239b68d17ff2","x":8.8454046249,"y":3.5835118294} +{"id":"fd7d94fbab084bc380480abeef6bfade","x":6.4138445854,"y":4.817322731} +{"id":"cfb915c95caf41c6a25e99a9f37f03a2","x":null,"y":null} +{"id":"8815ed80f9b741dbb458d902024f34a4","x":null,"y":null} +{"id":"dddb831546354e088d29aebd154e3a31","x":null,"y":null} +{"id":"005d2154da754b21adcd90ac921bd5f7","x":null,"y":null} +{"id":"711ba818354546cea69f1532b92a2f26","x":null,"y":null} +{"id":"5c4d8a8f9c104176b87d2bfdf04ae0bd","x":null,"y":null} +{"id":"5a781604f1fb4719b730f43f534627f6","x":null,"y":null} +{"id":"ecdc1020b10e49ca869d399825e16fa3","x":null,"y":null} +{"id":"0d8fde01d7234726a00d7e73e2e01d66","x":null,"y":null} +{"id":"9c4bd60958fd4e09a6d5b9e2ab163b5a","x":null,"y":null} +{"id":"39d31f770cf740e78d526a2e1101a1db","x":null,"y":null} +{"id":"9d282b2250f7408888504f1f93c202a8","x":null,"y":null} +{"id":"c063484895794a0eaae1b0ff070ad4c9","x":null,"y":null} +{"id":"e8868920e21b4431aad16e86db977ecb","x":null,"y":null} +{"id":"aea3378bfff842e5b3f4b7a4b55b3879","x":null,"y":null} +{"id":"d562223c17d948bf98e34b4d97dde932","x":null,"y":null} +{"id":"cde2d75c51d245879265b79d14b8699b","x":null,"y":null} +{"id":"44594467054849d4a1fadb46ddd51641","x":null,"y":null} +{"id":"2918130221f94f4387da049b647bfe6a","x":null,"y":null} +{"id":"fd139ac75b0e4777ab67b7423eaaa37f","x":null,"y":null} +{"id":"a701c349eb7142d48ba7efad89caf9d2","x":null,"y":null} +{"id":"e5d40a1b17f74b1db5d18279caedb04a","x":null,"y":null} +{"id":"de25d06733d04385825ee082792f5e52","x":null,"y":null} +{"id":"32f6f11a7845416b8c6eb9fb0b382140","x":null,"y":null} +{"id":"91407be8c3e54e23918d3a7183d962db","x":null,"y":null} +{"id":"3831134696584d83bbf676a6b3bfa8f9","x":null,"y":null} +{"id":"50e512a5dbe941f5af68bfdf74b1c3c0","x":null,"y":null} +{"id":"edc717747e904728b57185f5013461f9","x":null,"y":null} +{"id":"8fba1fea719d49d380ac2d9c310d68b3","x":null,"y":null} +{"id":"532da08f04f645708e747c57e9c4ee05","x":null,"y":null} +{"id":"3cf0ab4cf14e47ddabd49d500a3dc488","x":null,"y":null} +{"id":"a39b72f8921f43ef8ef295c7cc8f7294","x":null,"y":null} +{"id":"9f5adbeb6cf04f089abe78d86cfa6aba","x":null,"y":null} +{"id":"efb6350e65964659bc20396c0166b296","x":null,"y":null} +{"id":"e095cc36da784300b27c6f8c60a96440","x":null,"y":null} +{"id":"c68893ca39d74ba08c6eb138f24441e1","x":null,"y":null} +{"id":"472b23bb92834173b4118d101040c726","x":null,"y":null} +{"id":"81869985b45a4fefbbbb23ea118a3de4","x":null,"y":null} +{"id":"42b8584c5a874eb08fbd61f0c18f3ca0","x":null,"y":null} +{"id":"824d93d9840a4b7c8b1f31bc6816b497","x":null,"y":null} +{"id":"f209a808f1f04a5699601e672f4abd06","x":null,"y":null} +{"id":"ccb335166f6c4564ac1c61549d8ded50","x":null,"y":null} +{"id":"cbe1a41a82aa4f268e8264568b25938f","x":null,"y":null} +{"id":"28e7639f55ce464c8a080cbb2c745fa2","x":null,"y":null} +{"id":"3f3a2d7aa1294116814f0b4d89baa23d","x":null,"y":null} +{"id":"3073b33926bd4f33807ffa3befacefaf","x":null,"y":null} +{"id":"2b916117691c4872a9c4e4888d4fe4ab","x":null,"y":null} +{"id":"1f7b02bf486e4f42b23e9cb1a63207f3","x":null,"y":null} +{"id":"e744c118ae7f4638a01d060bbaedd6e9","x":null,"y":null} +{"id":"e1c1080c717d437996def1a41772d179","x":null,"y":null} +{"id":"63fba9a7c47a4f14ac0bee6bc90d0fea","x":null,"y":null} +{"id":"6bfc2395b4f54a528a1ebac94a43acb8","x":null,"y":null} +{"id":"1cce5cebf437428eb1a60dffbdfa603f","x":null,"y":null} +{"id":"dc94039d6643460ca3c66150b9087129","x":null,"y":null} +{"id":"f197d75f159943f8a3ff441199790bc7","x":null,"y":null} +{"id":"4d8890c699684c9381105b03b0b41b03","x":null,"y":null} +{"id":"b1658adfa43847eabad1437db235e858","x":null,"y":null} +{"id":"a1773cac7d4c4939aec965660e5015fe","x":null,"y":null} +{"id":"6a054cb59fb44cf494b93988b5f88833","x":null,"y":null} +{"id":"e7b103a52e384e3e8bf14105223e7e82","x":1.8886551857,"y":14.8363647461} +{"id":"3f1042452c254cecaf7189e89162adc8","x":2.1692409515,"y":14.5427055359} +{"id":"fd31d549420744d1bd1a6b1112a9a6ba","x":2.1614716053,"y":14.8653726578} +{"id":"f7ab348030714072a277682b51f7c588","x":null,"y":null} +{"id":"2139b0906dc541e094138a978d070416","x":null,"y":null} +{"id":"ff5466607e5d4453b1d833629292f664","x":null,"y":null} +{"id":"71f95003936e46a98d90757ffd845d40","x":null,"y":null} +{"id":"bada987ea7da4c939393ee1c3d08ccd4","x":null,"y":null} +{"id":"d0a274e7934d446fb91847bb53a961a6","x":null,"y":null} +{"id":"0a799eab61bc4e6b884db6689f9c2c4a","x":null,"y":null} +{"id":"8c34cd494a63438dac219c1dc0f73100","x":null,"y":null} +{"id":"c6f428af0c5e4f629902fd5455bf19ac","x":null,"y":null} +{"id":"d1fd271d16c348019c2fcced762b35a2","x":null,"y":null} +{"id":"ffa128c9c0c84d39bad1bba8cfa4adc5","x":null,"y":null} +{"id":"058f66cc356b43cc9433bd3c8d57fa46","x":null,"y":null} +{"id":"ff74091eaba246698fcae59c21eec828","x":null,"y":null} +{"id":"f6cbbf1b8f4b48a28a16e4dd8976b9bb","x":null,"y":null} +{"id":"757ca40654d5476aa949a26b733be8d4","x":null,"y":null} +{"id":"539d55e7c42e44b59d98f59fae3e0ee1","x":null,"y":null} +{"id":"3785eeadea9042bfb2e50f16c0397a12","x":null,"y":null} +{"id":"48cd97f2297143e09d61ff2a8542c0c5","x":null,"y":null} +{"id":"ff95eb0d5f7f49b782027d5c7ae3c3fe","x":null,"y":null} +{"id":"086da554db5b4ad5806aedeb0024197c","x":null,"y":null} +{"id":"216ee8a907a0466a88b27f8ada19ffa0","x":-15.1354284286,"y":5.1517944336} +{"id":"6fefb317687d4ac98efe39a52f3e190f","x":null,"y":null} +{"id":"320d9d91238948a8be67972ccceab878","x":null,"y":null} +{"id":"bdcbcccadd474b3bbe9a8f56c811bab4","x":null,"y":null} +{"id":"f127fc4d87f94794be89134406ba0694","x":-1.0490934849,"y":-13.361166954} +{"id":"c27966a4e3be434686454204ac7b3ab4","x":-0.6869778037,"y":-12.9910898209} +{"id":"dab39f92d0ed468c80699f28c05c45fa","x":-0.7393257022,"y":-13.0500297546} +{"id":"3076f330d121489aa50964ce54a3b1ac","x":-1.0465801954,"y":-13.3642187119} +{"id":"c8e5d3afdcb54c8589e280f0c4a87417","x":-15.564991951,"y":5.5821127892} +{"id":"f3d30627e19245649e497ab49bf0fa30","x":-15.3349027634,"y":5.3504748344} +{"id":"e3f1098c3d984bc7b5f30b9c0101f7a6","x":-15.4290647507,"y":5.4432139397} +{"id":"24b4a5f4db67418cbfa08c5316f0ab51","x":-15.7432994843,"y":5.758934021} +{"id":"e4b707e3e6964197855b82fc66ef59e7","x":null,"y":null} +{"id":"109b8be5a8ee4180a1465cd23f019d7b","x":null,"y":null} +{"id":"49f771e31a0c4b35bc39e389f3623509","x":null,"y":null} +{"id":"aa946d4379694a74ba0da37e69d2810a","x":null,"y":null} +{"id":"268446fc52a54fd2837f73aeb3e0b74f","x":null,"y":null} +{"id":"f6ddfa8491ff40d2839bb5b2e105df22","x":null,"y":null} +{"id":"db1295504da645b69d9786d54f233fed","x":null,"y":null} +{"id":"6ff4ed0dda4f4158af37be99f505565f","x":null,"y":null} +{"id":"5d398b88ee4242a59c32feb188683ec3","x":null,"y":null} +{"id":"0a784e00c9464bd3aeb830b908f73170","x":null,"y":null} +{"id":"b0966a0f455e44229e6c9705d57bfca9","x":null,"y":null} +{"id":"99761e9b89cc4060be3ed6b34532e7ff","x":1.9553822279,"y":14.5677165985} +{"id":"8130a1a82bde46048952cf147690e630","x":-1.117939353,"y":-13.4356231689} +{"id":"79c99026b7ef4946b9b8e0be841fd4c5","x":-14.9940023422,"y":5.0192642212} +{"id":"fdcb1673254842f1935f53d0c38c467e","x":null,"y":null} +{"id":"dcb3f4cc8abc46faabc193d9885e91d0","x":null,"y":null} +{"id":"3295be59128d451bb720c6688adc1e0b","x":null,"y":null} +{"id":"aca3eb8924ac494486fe0bfe892f7f2e","x":null,"y":null} +{"id":"66689accdd974295b7eb779e43578748","x":null,"y":null} +{"id":"6b49c78aa1524609ab7aa74aeaa3e01d","x":null,"y":null} +{"id":"7ff31ce54f424f0bbb297b0b3ba7c757","x":null,"y":null} +{"id":"bac51e00d486420c8e91e824d8e17411","x":null,"y":null} +{"id":"4adee3aad6524a4aa4c4711c1ee05e64","x":null,"y":null} +{"id":"d034e4fd8ac849278e658daad1a1f033","x":null,"y":null} +{"id":"091e998370dd42d1b05ab0fcf6595a7e","x":null,"y":null} +{"id":"1e6cabc18fab4c048281fd29d3044438","x":null,"y":null} +{"id":"dc08f6d7398b4b798a3bdccf508a2ad4","x":null,"y":null} +{"id":"1c7fd5af8d8041e186eae2431fc627cd","x":null,"y":null} +{"id":"b16eda56dcec40f2b3e109fb9246bee3","x":null,"y":null} +{"id":"43c68f9a86654a32a2215e23957ed184","x":null,"y":null} +{"id":"1ba06fe2e86140a59bbc4f4e969d0f71","x":null,"y":null} +{"id":"36caa0a230c8422c8acb4dc62e35bb32","x":null,"y":null} +{"id":"09940fed9d154504948bba2df1789a50","x":null,"y":null} +{"id":"4d6608557eed49368a6d09c7c5c664c5","x":null,"y":null} +{"id":"eb7c93eeb9dc41aab57d29e97ebb4951","x":null,"y":null} +{"id":"3b6e2ac584b64847b53828c9d779fed3","x":null,"y":null} +{"id":"e9b68002e035447baae848208cea5503","x":null,"y":null} +{"id":"fe18353546824ca98294ce4be7b96e02","x":null,"y":null} +{"id":"0e9740e25f5a460c81318336e00ac880","x":null,"y":null} +{"id":"b7cd9a62710849778fdadced0d754687","x":null,"y":null} +{"id":"432a6b4962544200949421a96a405142","x":null,"y":null} +{"id":"d6700b360ac141d282cdb567414bf4ce","x":null,"y":null} +{"id":"c1b40a4039b44061a358e098867f7412","x":null,"y":null} +{"id":"4643a7a319674adfb732b6f6122c7c64","x":null,"y":null} +{"id":"46e8056fb2ec4811ab33cb34a0dc9fb3","x":null,"y":null} +{"id":"8b57a9f43a1942a49b58cf881835f974","x":null,"y":null} +{"id":"f78b01b0d93948c283644ec58f7be74a","x":null,"y":null} +{"id":"8dbe8f9867e4448f998416c18923eac4","x":null,"y":null} +{"id":"fe8ea8bf1395434393e04e8f7a33025f","x":null,"y":null} +{"id":"7d58b089bfc549e8951e91ad62541119","x":null,"y":null} +{"id":"1fa6d3118bd846c8837b5fa9fb78f262","x":null,"y":null} +{"id":"62c65bbae33c4ee9a21b61f6f454c4b4","x":null,"y":null} +{"id":"30b7034c4468473f98ee18d00ee73b33","x":null,"y":null} +{"id":"00f78b85e5b84999a810e311e540037b","x":null,"y":null} +{"id":"3e460d9f011d4b0b9ccaae7b6a5202de","x":null,"y":null} +{"id":"9d98dece22eb401aa1a5ce9c88c603f0","x":null,"y":null} +{"id":"81446ea789b24eaf9eab02dc07c3d984","x":null,"y":null} +{"id":"79f4b1c1b2be4cf7aa828846e20a4eb6","x":null,"y":null} +{"id":"de04830d6e414fd5b39a9e90769d9452","x":null,"y":null} +{"id":"69db426b97714835bf4937180774787a","x":null,"y":null} +{"id":"9c7bc862339d4a5bb21ee5154d9b33bb","x":null,"y":null} +{"id":"17bad53a0ebe4569839e5e151ff78593","x":null,"y":null} +{"id":"53d98f08e7c74158b7318357b6c660b3","x":null,"y":null} +{"id":"cd601f77419c403889aadeee591915b5","x":null,"y":null} +{"id":"0f564ebd53e940fba9d16674ac7bc038","x":null,"y":null} +{"id":"7deb75816e4f473480e0c79ae99b5bf4","x":null,"y":null} +{"id":"7f85b181f1184f77aeb3ea2155cf4027","x":null,"y":null} +{"id":"d148b2b2033048618f1a090a492a40a5","x":null,"y":null} +{"id":"4d839a10353e4144a26563b0966721d5","x":null,"y":null} +{"id":"521a862bb196488389f17c0b0f4b6f4d","x":null,"y":null} +{"id":"22ea3328fb6343f4ad2862495ea27640","x":null,"y":null} +{"id":"3f9a2a2c1c0a424e8b4980ea9d48bdbe","x":-4.7582893372,"y":20.9018955231} +{"id":"aa2ec452728a4703ae1bdabe85b6c079","x":null,"y":null} +{"id":"c5ddb31e0a9c4b2683e4631283dd505b","x":null,"y":null} +{"id":"07d8eeb549044ac88d2e788c146a0ef1","x":null,"y":null} +{"id":"47df2815030c4f1c99facd5cf2482526","x":null,"y":null} +{"id":"ae521508bdc244f99c4fce4ab5214c79","x":null,"y":null} +{"id":"6315b4bf135c40358823ed7e4e4060e2","x":null,"y":null} +{"id":"33905debec1a45ecae1c65daac1d854c","x":null,"y":null} +{"id":"bfbe904780fe47daad1a04126b12923c","x":null,"y":null} +{"id":"0614f00e932c4cd0b53928053811ebc1","x":null,"y":null} +{"id":"9ef487dd0b574b108c60a56d6a2f146c","x":null,"y":null} +{"id":"4067269e7f6943cdbc299ce02b7eadbd","x":null,"y":null} +{"id":"094a736ba43c4da48c556437f47f88d1","x":null,"y":null} +{"id":"563c2af32bb3476299e9b24a646097ab","x":null,"y":null} +{"id":"d59b49eb94ce442d89907e90c5d3a44e","x":null,"y":null} +{"id":"8ea7cef407df48098046551e303e1c64","x":null,"y":null} +{"id":"186e60d2176547bf84e5bf87bd16bb40","x":null,"y":null} +{"id":"e65017091c8d4c7daa45b6c8414e0465","x":null,"y":null} +{"id":"a0f326b9597b49dda6563e9208316117","x":null,"y":null} +{"id":"bff3db70f9af4f2c87a93df48ecbb6bc","x":null,"y":null} +{"id":"bf91f36307cb43e1ab1e967cb3ba8274","x":null,"y":null} +{"id":"cd58a8740ba54d86a77db9bb9544ef0d","x":null,"y":null} +{"id":"e96d3475d43b42a781b297ae7e650afe","x":null,"y":null} +{"id":"1ce76a5547854d458878bd445f0ccbd6","x":null,"y":null} +{"id":"11e4325f59394ff1bc89892f79288702","x":null,"y":null} +{"id":"71743537a07c440ea1710a269da8b538","x":null,"y":null} +{"id":"1389192ce5464be6b3b5749bc9536709","x":null,"y":null} +{"id":"b349041c0be64c62b964ab1234e055e6","x":-4.1061320305,"y":3.1022937298} +{"id":"969e1ea0b1e443a68e9a65dfef91d161","x":null,"y":null} +{"id":"8e09e7cfea7d405db8b22ae2f836ccb1","x":null,"y":null} +{"id":"490583524d394bf79289c5fe34f7dcf1","x":null,"y":null} +{"id":"d7db38bb599c42cab7066f3fdd282282","x":null,"y":null} +{"id":"efd87a59d01e47c8adc02f63ef2c5c3e","x":null,"y":null} +{"id":"80e3ce3de41e4601823a333e22b7bb3f","x":null,"y":null} +{"id":"50eabc166e8944a49197e79c32f27597","x":null,"y":null} +{"id":"5197a3fb02ef4677abd1900aa87e4efa","x":null,"y":null} +{"id":"887f444240bb474da23cdfb6abf7a998","x":null,"y":null} +{"id":"5d29053f2ce74442aa1855b327ef3bb7","x":null,"y":null} +{"id":"7e40cd12839a4577a95e33d785147a31","x":null,"y":null} +{"id":"8fe58de8a04f4f8f807c77fb41829a3a","x":null,"y":null} +{"id":"a9f50861273c4bb697d868a9d049d392","x":null,"y":null} +{"id":"be4820f29fd942b282049fa49697b4ed","x":null,"y":null} +{"id":"6deaefe707f84b3dbda979dea0d095ac","x":null,"y":null} +{"id":"d053ea9432a24fb192e8d6aa993b0caa","x":null,"y":null} +{"id":"a3e683d294ed42a28d60d09a36cbeb54","x":6.8404831886,"y":10.4973592758} +{"id":"39887ca8567141d5b857b87a2bca4086","x":-5.0493412018,"y":21.2849559784} +{"id":"8df8563ab0394ee9a91b89dea7d59404","x":null,"y":null} +{"id":"12398f70065143839d812fd42ac4b2e7","x":null,"y":null} +{"id":"74d43d20f251441baf8e3db64fedca43","x":null,"y":null} +{"id":"1b7a22f76f7741e8b140bdc3d8856d76","x":null,"y":null} +{"id":"b823ba1bfe944fa9887edd8faf8a5f17","x":null,"y":null} +{"id":"d0bfb473fdc64643954cdb4675e2f389","x":null,"y":null} +{"id":"a4db1b2a9c3e4d2d838725f8166c36b4","x":null,"y":null} +{"id":"8dae140578c841ae9373cbc607c4a6e6","x":null,"y":null} +{"id":"b215cc33cf40434f87f284ff8f3506a4","x":null,"y":null} +{"id":"c1ff9d8e1b8745d6860c34ce26122d79","x":null,"y":null} +{"id":"9d1e6ca9ae8e4e068fb74631a633b20b","x":null,"y":null} +{"id":"1d7b0deca7674777bf76c163ac065845","x":null,"y":null} +{"id":"03afe9988f864c9fa501bfbf043f74c0","x":null,"y":null} +{"id":"4084f614af494fa8ab73095fb5b6b07b","x":null,"y":null} +{"id":"3ce25564af6e47f390a0b16b6f9433a1","x":null,"y":null} +{"id":"78213664d0eb45d1a9239ba4b85b10f7","x":null,"y":null} +{"id":"1226e4a4077b4b3a970db4d2509b590c","x":null,"y":null} +{"id":"b4c7de7a824a4a71b9f52193d2f1a10d","x":null,"y":null} +{"id":"b609f1939dae4c7383c7d199bb3c7dc3","x":null,"y":null} +{"id":"aeee2f443dfb4e3ea80af6ae1d9197ce","x":null,"y":null} +{"id":"8c46d37bc26e4d4dbd37d6ee26867bc6","x":null,"y":null} +{"id":"58a8fa7f29e347bdb9689b70b065a779","x":null,"y":null} +{"id":"fae3fe31deb141ab93143ac411f1eaaa","x":null,"y":null} +{"id":"a2cb46c226b94831853a5d28c5d94b0a","x":-2.6890439987,"y":2.7342894077} +{"id":"d3511ecd27cd4166bdb39e757e275300","x":-3.3049843311,"y":2.9782123566} +{"id":"de3b561f5cce4c83bccb39180e362c97","x":-2.5151634216,"y":3.2286095619} +{"id":"5bfefaa0fce04002851733337bed714c","x":-2.0915775299,"y":2.8603157997} +{"id":"b5fed5609f154df58c6a9f74e55fc0ba","x":-2.6874377728,"y":3.0045831203} +{"id":"91ae5251eaab4c08afe6cd4cbefcaa6b","x":-2.2914175987,"y":2.9969851971} +{"id":"bbdd53a15e99452a9deff05d1de2d965","x":null,"y":null} +{"id":"532bf54d5a924ff48aee254970efb914","x":null,"y":null} +{"id":"2489232bd2bb492babe00617e7290282","x":null,"y":null} +{"id":"d2ed972353af4d1db74702638bfdbb58","x":null,"y":null} +{"id":"575befc8d64c47eb95af8b1096e02963","x":null,"y":null} +{"id":"d6e6366617e04b0ba6732fd1d2d76429","x":null,"y":null} +{"id":"b4c4354c8edb40db984942799fe0c8b1","x":null,"y":null} +{"id":"170507a64973429f818067b80506d428","x":null,"y":null} +{"id":"fd9b298e6aea4685bbb2064b05fcda79","x":null,"y":null} +{"id":"eeecb159cc8a4c8989f8da0f3df09f2a","x":null,"y":null} +{"id":"70f22b1d7336492dbade94b8edefe457","x":null,"y":null} +{"id":"66e098dc431146e19fc4bc2ea37efbd9","x":null,"y":null} +{"id":"932e213c57134098a07073febd51dcc2","x":null,"y":null} +{"id":"9593428ad36746ae8af6d8ce639834ef","x":null,"y":null} +{"id":"1bcaeb58479d42a6963a073c09f3f397","x":null,"y":null} +{"id":"1ef0c1c59ce946668ccf1a6a4f5ab7cc","x":null,"y":null} +{"id":"d734746e3d6146f780af91827e578dfd","x":null,"y":null} +{"id":"21ed913271614cbeb1b754cdbbef13af","x":null,"y":null} +{"id":"1505dfebbfb04652b0ba57de1a251d67","x":null,"y":null} +{"id":"907ec65076e5494a8631efffb81b3178","x":null,"y":null} +{"id":"2dc7f6b230db452190a09643ca3d5ec0","x":null,"y":null} +{"id":"c20ecfc93b3a4875ade5c92cfe4b94a1","x":null,"y":null} +{"id":"4bc7dc91ede345dfb63d7d4f7ac3554f","x":null,"y":null} +{"id":"0b2b815c9f834aaaac0c341097def9ba","x":-2.7582564354,"y":-8.6630039215} +{"id":"424ae71c56024094a02e6fd9bfcfbb04","x":-2.9365997314,"y":-8.8427972794} +{"id":"400d10f2ee1d49be9a66efa34dada0e6","x":-2.5346698761,"y":-8.4429206848} +{"id":"91deb9f152264e958d106d481ff2e1ee","x":-2.0110886097,"y":-7.9206728935} +{"id":"586cf02da9494088aed9b3419725638f","x":-2.5314006805,"y":-8.4376468658} +{"id":"229d85a2783e4a2991f17d2ab5750af7","x":-2.0421271324,"y":-7.9088606834} +{"id":"b7f97d1909a3433abef8ca8e9334fafa","x":-2.373016119,"y":-8.2822990417} +{"id":"b7fdfffc38b94bf7872eabe9b022c8fd","x":-1.9533445835,"y":-7.8647270203} +{"id":"6242e0c237a348908d0256ea790a0211","x":-2.0474627018,"y":-7.9563288689} +{"id":"7cc9f26737e1442595e53253e98015ef","x":-6.9051356316,"y":2.7752327919} +{"id":"1868fec1493643208dbdcad7bc97dfa0","x":null,"y":null} +{"id":"a87aa935dccf49cd98b40fb5afe7ad5c","x":-2.5083763599,"y":-8.4104852676} +{"id":"36870a3393f6413e9bf647168eb6977a","x":-6.8342895508,"y":2.5542294979} +{"id":"4fe3ff52700c491f8cc650aadb4d7cb0","x":-9.4894418716,"y":1.6655651331} +{"id":"f1f6f6435a444e388d67e16e847afca6","x":-6.7472057343,"y":2.5103824139} +{"id":"0af2ca1c090843ea92679fd14c1fbc9a","x":-6.6448841095,"y":-0.0083856769} +{"id":"1b06d3e53ffd4771952fbef04d1e666c","x":-6.8072009087,"y":0.1406142265} +{"id":"b8e966b34cba4b11b9995106767212ba","x":-6.4710240364,"y":-0.191854015} +{"id":"f6de923de6474d2cab6a9c2f0d81fa59","x":-6.6300406456,"y":-0.0426665545} +{"id":"6915637e8d124fdc8473111d501e3703","x":-6.8250803947,"y":0.1514172256} +{"id":"2233f31929194eac89333ce8731a5584","x":-7.0507774353,"y":0.3937795758} +{"id":"61f1dc4267314470ac820b6a46c61f7b","x":null,"y":null} +{"id":"f0c578614b224345974c3e4c110878af","x":null,"y":null} +{"id":"7ffb88ebc729492c897ccfb569d7f6d0","x":null,"y":null} +{"id":"60dce7d8bc1b4729a038178a400b9a59","x":null,"y":null} +{"id":"4cbb4e238c5b4656803fb9b4b6c3512e","x":null,"y":null} +{"id":"652873bcd6d5432187e5deafc4fc5211","x":null,"y":null} +{"id":"78f9b30c08134ac5abb4f4e0bff0f7f2","x":null,"y":null} +{"id":"f33e4e897b1e422bb516e8a2c941d9dc","x":null,"y":null} +{"id":"fac4e1553a9840e990bbfff46e64ff27","x":null,"y":null} +{"id":"029a55d327ee4fb3a8314b36d52bdf34","x":null,"y":null} +{"id":"5a636c894c384532bff66212cf9d5824","x":null,"y":null} +{"id":"a9c468ef78704e9aabfc0317a5b1b42d","x":-4.6287069321,"y":3.7560479641} +{"id":"5df80c25d33a4d148a14aa614343cc6b","x":null,"y":null} +{"id":"6a87f06ed55a46f29b24f77e548a3f1d","x":null,"y":null} +{"id":"0daf88ac4ec94cbb868e27e956c6d7f1","x":null,"y":null} +{"id":"9ed120043e6247be9965e4904920991b","x":-5.2956299782,"y":3.5257122517} +{"id":"94d81d7de9254ae4b3b16fcc69aa22ea","x":-5.0318026543,"y":3.1229879856} +{"id":"60c9212246f84ae5b6ab254127a39262","x":-4.907330513,"y":3.3409392834} +{"id":"0f8d0c36a4274526a9eddedae5e63881","x":-4.706278801,"y":3.3143031597} +{"id":"6aedd377efbe4f07ae42e546996e7bfa","x":-10.4078292847,"y":1.8026002645} +{"id":"1aa8484562784f378851c33843c89687","x":-10.4976444244,"y":0.8868256807} +{"id":"f1a65d05dd5d456b889217020475ef80","x":-10.1209726334,"y":1.4226669073} +{"id":"c077d92b48b6477db91e1a0460600f52","x":-10.2925043106,"y":1.284881711} +{"id":"5ca888df9b884e54accdd2ff29d125c1","x":-10.7187824249,"y":0.6120534539} +{"id":"8290a6212d6c4430ae0056c7e8eccd5f","x":-10.6491918564,"y":0.8072271943} +{"id":"14f8ac195fdb4e06a0b9ebc6ef391180","x":-9.9971961975,"y":0.9321267009} +{"id":"667ee58a79194316ae2b82eadd3fc575","x":-10.5873994827,"y":1.5912019014} +{"id":"b0e3ee2324054c88adacdf80db13278f","x":-10.8272399902,"y":1.1503276825} diff --git a/graphfleet/output/graphindex/reports/logs.json b/graphfleet/output/graphindex/reports/logs.json new file mode 100644 index 000000000..e3cd7b82e --- /dev/null +++ b/graphfleet/output/graphindex/reports/logs.json @@ -0,0 +1,49 @@ +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 41 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 41 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\n-Goal-\nGiven a text document that is potentially relevant to this activity, first identify all entities needed from the text in order to capture the information and ideas in the text.\nNext, report all relationships among the identified entities.\n\n-Steps-\n1. Identify all entities. For each identified entity, extract the following information:\n- entity_name: Name of the entity, capitalized\n- entity_type: Suggest several labels or categories for the entity. The categories should not be specific, but should be as general as possible.\n- entity_description: Comprehensive description of the entity's attributes and activities\nFormat each entity as (\"entity\"<|><|><|>)\n\n2. From the entities identified in step 1, identify all pairs of (source_entity, target_entity) that are *clearly related* to each other.\nFor each pair of related entities, extract the following information:\n- source_entity: name of the source entity, as identified in step 1\n- target_entity: name of the target entity, as identified in step 1\n- relationship_description: explanation as to why you think the source entity and the target entity are related to each other\n- relationship_strength: a numeric score indicating strength of the relationship between the source entity and target entity\nFormat each relationship as (\"relationship\"<|><|><|><|>)\n\n3. Return output in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing. as a single list of all the entities and relationships identified in steps 1 and 2. Use **##** as the list delimiter.\n\n4. If you have to translate into The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing., just translate the descriptions, nothing else!\n\n5. When finished, output <|COMPLETE|>.\n\n-Examples-\n######################\n\nExample 1:\n\ntext:\n results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most\n------------------------\noutput:\n(\"entity\"<|>RAG<|>METHOD<|>RAG (Retrieval-Augmented Generation) is a method that produces direct responses in text generation tasks)\n##\n(\"entity\"<|>PODCAST TRANSCRIPTS<|>DATASET<|>A dataset consisting of transcripts from podcasts used for analysis)\n##\n(\"entity\"<|>NEWS ARTICLES<|>DATASET<|>A dataset consisting of news articles used for analysis)\n##\n(\"entity\"<|>C0<|>CATEGORY<|>A category or cluster used in the analysis, representing a specific subset of the data)\n##\n(\"entity\"<|>C1<|>CATEGORY<|>A category or cluster used in the analysis, representing a specific subset of the data)\n##\n(\"entity\"<|>C2<|>CATEGORY<|>A category or cluster used in the analysis, representing a specific subset of the data)\n##\n(\"entity\"<|>C3<|>CATEGORY<|>A category or cluster used in the analysis, representing a specific subset of the data)\n##\n(\"entity\"<|>TS<|>CATEGORY<|>A category or cluster used in the analysis, representing a specific subset of the data)\n##\n(\"entity\"<|>UNITS<|>METRIC<|>The number of context units, such as community summaries or text chunks, used in the analysis)\n##\n(\"entity\"<|>TOKENS<|>METRIC<|>The number of tokens, or individual words, used in the analysis)\n##\n(\"entity\"<|>% MAX<|>METRIC<|>The percentage of the maximum token count used in the analysis)\n##\n(\"relationship\"<|>RAG<|>PODCAST TRANSCRIPTS<|>RAG is used to produce direct responses from podcast transcripts<|>7)\n##\n(\"relationship\"<|>RAG<|>NEWS ARTICLES<|>RAG is used to produce direct responses from news articles<|>7)\n##\n(\"relationship\"<|>C0<|>PODCAST TRANSCRIPTS<|>C0 is a category used in the analysis of podcast transcripts<|>5)\n##\n(\"relationship\"<|>C1<|>PODCAST TRANSCRIPTS<|>C1 is a category used in the analysis of podcast transcripts<|>5)\n##\n(\"relationship\"<|>C2<|>PODCAST TRANSCRIPTS<|>C2 is a category used in the analysis of podcast transcripts<|>5)\n##\n(\"relationship\"<|>C3<|>PODCAST TRANSCRIPTS<|>C3 is a category used in the analysis of podcast transcripts<|>5)\n##\n(\"relationship\"<|>TS<|>PODCAST TRANSCRIPTS<|>TS is a category used in the analysis of podcast transcripts<|>5)\n##\n(\"relationship\"<|>C0<|>NEWS ARTICLES<|>C0 is a category used in the analysis of news articles<|>5)\n##\n(\"relationship\"<|>C1<|>NEWS ARTICLES<|>C1 is a category used in the analysis of news articles<|>5)\n##\n(\"relationship\"<|>C2<|>NEWS ARTICLES<|>C2 is a category used in the analysis of news articles<|>5)\n##\n(\"relationship\"<|>C3<|>NEWS ARTICLES<|>C3 is a category used in the analysis of news articles<|>5)\n##\n(\"relationship\"<|>TS<|>NEWS ARTICLES<|>TS is a category used in the analysis of news articles<|>5)\n##\n(\"relationship\"<|>UNITS<|>PODCAST TRANSCRIPTS<|>Units are used to measure the context in podcast transcripts<|>6)\n##\n(\"relationship\"<|>UNITS<|>NEWS ARTICLES<|>Units are used to measure the context in news articles<|>6)\n##\n(\"relationship\"<|>TOKENS<|>PODCAST TRANSCRIPTS<|>Tokens are used to measure the word count in podcast transcripts<|>6)\n##\n(\"relationship\"<|>TOKENS<|>NEWS ARTICLES<|>Tokens are used to measure the word count in news articles<|>6)\n##\n(\"relationship\"<|>% MAX<|>PODCAST TRANSCRIPTS<|>% Max is used to measure the percentage of maximum token count in podcast transcripts<|>6)\n##\n(\"relationship\"<|>% MAX<|>NEWS ARTICLES<|>% Max is used to measure the percentage of maximum token count in news articles<|>6)\n<|COMPLETE|>\n#############################\n\n\nExample 2:\n\ntext:\n , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: \n------------------------\noutput:\n(\"entity\"<|>KOESTEN, L.<|>PERSON<|>Koesten, L. is an author of the paper \"Talking datasetsā€“understanding data sensemaking behaviours\")\n##\n(\"entity\"<|>GREGORY, K.<|>PERSON<|>Gregory, K. is an author of the paper \"Talking datasetsā€“understanding data sensemaking behaviours\")\n##\n(\"entity\"<|>GROTH, P.<|>PERSON<|>Groth, P. is an author of the paper \"Talking datasetsā€“understanding data sensemaking behaviours\")\n##\n(\"entity\"<|>SIMPERL, E.<|>PERSON<|>Simperl, E. is an author of the paper \"Talking datasetsā€“understanding data sensemaking behaviours\")\n##\n(\"entity\"<|>INTERNATIONAL JOURNAL OF HUMAN-COMPUTER STUDIES<|>PUBLICATION<|>The journal where the paper \"Talking datasetsā€“understanding data sensemaking behaviours\" was published)\n##\n(\"entity\"<|>KURATOV, Y.<|>PERSON<|>Kuratov, Y. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>BULATOV, A.<|>PERSON<|>Bulatov, A. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>ANOKHIN, P.<|>PERSON<|>Anokhin, P. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>SOROKIN, D.<|>PERSON<|>Sorokin, D. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>SOROKIN, A.<|>PERSON<|>Sorokin, A. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>BURTSEV, M.<|>PERSON<|>Burtsev, M. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>LANGCHAIN<|>ORGANIZATION<|>LangChain is an organization that developed Langchain graphs)\n##\n(\"entity\"<|>LANGCHAIN GRAPHS<|>TECHNOLOGY<|>Langchain graphs is a technology developed by LangChain)\n##\n(\"entity\"<|>LASKAR, M. T. R.<|>PERSON<|>Laskar, M. T. R. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\")\n##\n(\"entity\"<|>HOQUE, E.<|>PERSON<|>Hoque, E. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\")\n##\n(\"entity\"<|>HUANG, J.<|>PERSON<|>Huang, J. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\")\n##\n(\"entity\"<|>ADVANCES IN ARTIFICIAL INTELLIGENCE<|>PUBLICATION<|>The conference where the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\" was presented)\n##\n(\"relationship\"<|>KOESTEN, L.<|>GREGORY, K.<|>Koesten, L. and Gregory, K. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>KOESTEN, L.<|>GROTH, P.<|>Koesten, L. and Groth, P. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>KOESTEN, L.<|>SIMPERL, E.<|>Koesten, L. and Simperl, E. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>GREGORY, K.<|>GROTH, P.<|>Gregory, K. and Groth, P. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>GREGORY, K.<|>SIMPERL, E.<|>Gregory, K. and Simperl, E. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>GROTH, P.<|>SIMPERL, E.<|>Groth, P. and Simperl, E. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>KURATOV, Y.<|>BULATOV, A.<|>Kuratov, Y. and Bulatov, A. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>KURATOV, Y.<|>ANOKHIN, P.<|>Kuratov, Y. and Anokhin, P. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>KURATOV, Y.<|>SOROKIN, D.<|>Kuratov, Y. and Sorokin, D. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>KURATOV, Y.<|>SOROKIN, A.<|>Kuratov, Y. and Sorokin, A. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>KURATOV, Y.<|>BURTSEV, M.<|>Kuratov, Y. and Burtsev, M. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>BULATOV, A.<|>ANOKHIN, P.<|>Bulatov, A. and Anokhin, P. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>BULATOV, A.<|>SOROKIN, D.<|>Bulatov, A. and Sorokin, D. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>BULATOV, A.<|>SOROKIN, A.<|>Bulatov, A. and Sorokin, A. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>BULATOV, A.<|>BURTSEV, M.<|>Bulatov, A. and Burtsev, M. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>ANOKHIN, P.<|>SOROKIN, D.<|>Anokhin, P. and Sorokin, D. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>ANOKHIN, P.<|>SOROKIN, A.<|>Anokhin, P. and Sorokin, A. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>ANOKHIN, P.<|>BURTSEV, M.<|>Anokhin, P. and Burtsev, M. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>SOROKIN, D.<|>SOROKIN, A.<|>Sorokin, D. and Sorokin, A. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>SOROKIN, D.<|>BURTSEV, M.<|>Sorokin, D. and Burtsev, M. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>SOROKIN, A.<|>BURTSEV, M.<|>Sorokin, A. and Burtsev, M. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>LANGCHAIN<|>LANGCHAIN GRAPHS<|>LangChain developed Langchain graphs<|>9)\n##\n(\"relationship\"<|>LASKAR, M. T. R.<|>HOQUE, E.<|>Laskar, M. T. R. and Hoque, E. co-authored the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<|>8)\n##\n(\"relationship\"<|>LASKAR, M. T. R.<|>HUANG, J.<|>Laskar, M. T. R. and Huang, J. co-authored the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<|>8)\n##\n(\"relationship\"<|>HOQUE, E.<|>HUANG, J.<|>Hoque, E. and Huang, J. co-authored the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<|>8)\n<|COMPLETE|>\n#############################\n\n\n\n-Real Data-\n######################\ntext: H., Li, Z., Xu, J., Qu, J., and Zhou, J. (2023a). Is chatgpt\na good nlg evaluator? a preliminary study. arXiv preprint arXiv:2303.04048 .\nWang, S., Khramtsova, E., Zhuang, S., and Zuccon, G. (2024). Feb4rag: Evaluating federated search\nin the context of retrieval augmented generation. arXiv preprint arXiv:2402.11891 .\nWang, Y ., Lipka, N., Rossi, R. A., Siu, A., Zhang, R., and Derr, T. (2023b). Knowledge graph\nprompting for multi-document question answering.\nXu, Y . and Lapata, M. (2021). Text summarization with latent queries. arXiv preprint\narXiv:2106.00104 .\nYang, Z., Qi, P., Zhang, S., Bengio, Y ., Cohen, W. W., Salakhutdinov, R., and Manning, C. D. (2018).\nHotpotQA: A dataset for diverse, explainable multi-hop question answering. In Conference on\nEmpirical Methods in Natural Language Processing (EMNLP) .\nYao, J.-g., Wan, X., and Xiao, J. (2017). Recent advances in document summarization. Knowledge\nand Information Systems , 53:297ā€“336.\n14Yao, L., Peng, J., Mao, C., and Luo, Y . (2023). Exploring large language models for knowledge\ngraph completion.\nZhang, J. (2023). Graph-toolformer: To empower llms with graph reasoning ability via prompt\naugmented by chatgpt. arXiv preprint arXiv:2304.11116 .\nZhang, Y ., Zhang, Y ., Gan, Y ., Yao, L., and Wang, C. (2024). Causal graph discovery with retrieval-\naugmented generation based large language models. arXiv preprint arXiv:2402.15301 .\nZheng, L., Chiang, W.-L., Sheng, Y ., Zhuang, S., Wu, Z., Zhuang, Y ., Lin, Z., Li, Z., Li, D., Xing,\nE., et al. (202\n######################\noutput:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 41 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 41 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\n-Goal-\nGiven a text document that is potentially relevant to this activity, first identify all entities needed from the text in order to capture the information and ideas in the text.\nNext, report all relationships among the identified entities.\n\n-Steps-\n1. Identify all entities. For each identified entity, extract the following information:\n- entity_name: Name of the entity, capitalized\n- entity_type: Suggest several labels or categories for the entity. The categories should not be specific, but should be as general as possible.\n- entity_description: Comprehensive description of the entity's attributes and activities\nFormat each entity as (\"entity\"<|><|><|>)\n\n2. From the entities identified in step 1, identify all pairs of (source_entity, target_entity) that are *clearly related* to each other.\nFor each pair of related entities, extract the following information:\n- source_entity: name of the source entity, as identified in step 1\n- target_entity: name of the target entity, as identified in step 1\n- relationship_description: explanation as to why you think the source entity and the target entity are related to each other\n- relationship_strength: a numeric score indicating strength of the relationship between the source entity and target entity\nFormat each relationship as (\"relationship\"<|><|><|><|>)\n\n3. Return output in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing. as a single list of all the entities and relationships identified in steps 1 and 2. Use **##** as the list delimiter.\n\n4. If you have to translate into The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing., just translate the descriptions, nothing else!\n\n5. When finished, output <|COMPLETE|>.\n\n-Examples-\n######################\n\nExample 1:\n\ntext:\n results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most\n------------------------\noutput:\n(\"entity\"<|>RAG<|>METHOD<|>RAG (Retrieval-Augmented Generation) is a method that produces direct responses in text generation tasks)\n##\n(\"entity\"<|>PODCAST TRANSCRIPTS<|>DATASET<|>A dataset consisting of transcripts from podcasts used for analysis)\n##\n(\"entity\"<|>NEWS ARTICLES<|>DATASET<|>A dataset consisting of news articles used for analysis)\n##\n(\"entity\"<|>C0<|>CATEGORY<|>A category or cluster used in the analysis, representing a specific subset of the data)\n##\n(\"entity\"<|>C1<|>CATEGORY<|>A category or cluster used in the analysis, representing a specific subset of the data)\n##\n(\"entity\"<|>C2<|>CATEGORY<|>A category or cluster used in the analysis, representing a specific subset of the data)\n##\n(\"entity\"<|>C3<|>CATEGORY<|>A category or cluster used in the analysis, representing a specific subset of the data)\n##\n(\"entity\"<|>TS<|>CATEGORY<|>A category or cluster used in the analysis, representing a specific subset of the data)\n##\n(\"entity\"<|>UNITS<|>METRIC<|>The number of context units, such as community summaries or text chunks, used in the analysis)\n##\n(\"entity\"<|>TOKENS<|>METRIC<|>The number of tokens, or individual words, used in the analysis)\n##\n(\"entity\"<|>% MAX<|>METRIC<|>The percentage of the maximum token count used in the analysis)\n##\n(\"relationship\"<|>RAG<|>PODCAST TRANSCRIPTS<|>RAG is used to produce direct responses from podcast transcripts<|>7)\n##\n(\"relationship\"<|>RAG<|>NEWS ARTICLES<|>RAG is used to produce direct responses from news articles<|>7)\n##\n(\"relationship\"<|>C0<|>PODCAST TRANSCRIPTS<|>C0 is a category used in the analysis of podcast transcripts<|>5)\n##\n(\"relationship\"<|>C1<|>PODCAST TRANSCRIPTS<|>C1 is a category used in the analysis of podcast transcripts<|>5)\n##\n(\"relationship\"<|>C2<|>PODCAST TRANSCRIPTS<|>C2 is a category used in the analysis of podcast transcripts<|>5)\n##\n(\"relationship\"<|>C3<|>PODCAST TRANSCRIPTS<|>C3 is a category used in the analysis of podcast transcripts<|>5)\n##\n(\"relationship\"<|>TS<|>PODCAST TRANSCRIPTS<|>TS is a category used in the analysis of podcast transcripts<|>5)\n##\n(\"relationship\"<|>C0<|>NEWS ARTICLES<|>C0 is a category used in the analysis of news articles<|>5)\n##\n(\"relationship\"<|>C1<|>NEWS ARTICLES<|>C1 is a category used in the analysis of news articles<|>5)\n##\n(\"relationship\"<|>C2<|>NEWS ARTICLES<|>C2 is a category used in the analysis of news articles<|>5)\n##\n(\"relationship\"<|>C3<|>NEWS ARTICLES<|>C3 is a category used in the analysis of news articles<|>5)\n##\n(\"relationship\"<|>TS<|>NEWS ARTICLES<|>TS is a category used in the analysis of news articles<|>5)\n##\n(\"relationship\"<|>UNITS<|>PODCAST TRANSCRIPTS<|>Units are used to measure the context in podcast transcripts<|>6)\n##\n(\"relationship\"<|>UNITS<|>NEWS ARTICLES<|>Units are used to measure the context in news articles<|>6)\n##\n(\"relationship\"<|>TOKENS<|>PODCAST TRANSCRIPTS<|>Tokens are used to measure the word count in podcast transcripts<|>6)\n##\n(\"relationship\"<|>TOKENS<|>NEWS ARTICLES<|>Tokens are used to measure the word count in news articles<|>6)\n##\n(\"relationship\"<|>% MAX<|>PODCAST TRANSCRIPTS<|>% Max is used to measure the percentage of maximum token count in podcast transcripts<|>6)\n##\n(\"relationship\"<|>% MAX<|>NEWS ARTICLES<|>% Max is used to measure the percentage of maximum token count in news articles<|>6)\n<|COMPLETE|>\n#############################\n\n\nExample 2:\n\ntext:\n , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: \n------------------------\noutput:\n(\"entity\"<|>KOESTEN, L.<|>PERSON<|>Koesten, L. is an author of the paper \"Talking datasetsā€“understanding data sensemaking behaviours\")\n##\n(\"entity\"<|>GREGORY, K.<|>PERSON<|>Gregory, K. is an author of the paper \"Talking datasetsā€“understanding data sensemaking behaviours\")\n##\n(\"entity\"<|>GROTH, P.<|>PERSON<|>Groth, P. is an author of the paper \"Talking datasetsā€“understanding data sensemaking behaviours\")\n##\n(\"entity\"<|>SIMPERL, E.<|>PERSON<|>Simperl, E. is an author of the paper \"Talking datasetsā€“understanding data sensemaking behaviours\")\n##\n(\"entity\"<|>INTERNATIONAL JOURNAL OF HUMAN-COMPUTER STUDIES<|>PUBLICATION<|>The journal where the paper \"Talking datasetsā€“understanding data sensemaking behaviours\" was published)\n##\n(\"entity\"<|>KURATOV, Y.<|>PERSON<|>Kuratov, Y. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>BULATOV, A.<|>PERSON<|>Bulatov, A. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>ANOKHIN, P.<|>PERSON<|>Anokhin, P. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>SOROKIN, D.<|>PERSON<|>Sorokin, D. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>SOROKIN, A.<|>PERSON<|>Sorokin, A. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>BURTSEV, M.<|>PERSON<|>Burtsev, M. is an author of the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\")\n##\n(\"entity\"<|>LANGCHAIN<|>ORGANIZATION<|>LangChain is an organization that developed Langchain graphs)\n##\n(\"entity\"<|>LANGCHAIN GRAPHS<|>TECHNOLOGY<|>Langchain graphs is a technology developed by LangChain)\n##\n(\"entity\"<|>LASKAR, M. T. R.<|>PERSON<|>Laskar, M. T. R. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\")\n##\n(\"entity\"<|>HOQUE, E.<|>PERSON<|>Hoque, E. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\")\n##\n(\"entity\"<|>HUANG, J.<|>PERSON<|>Huang, J. is an author of the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\")\n##\n(\"entity\"<|>ADVANCES IN ARTIFICIAL INTELLIGENCE<|>PUBLICATION<|>The conference where the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\" was presented)\n##\n(\"relationship\"<|>KOESTEN, L.<|>GREGORY, K.<|>Koesten, L. and Gregory, K. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>KOESTEN, L.<|>GROTH, P.<|>Koesten, L. and Groth, P. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>KOESTEN, L.<|>SIMPERL, E.<|>Koesten, L. and Simperl, E. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>GREGORY, K.<|>GROTH, P.<|>Gregory, K. and Groth, P. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>GREGORY, K.<|>SIMPERL, E.<|>Gregory, K. and Simperl, E. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>GROTH, P.<|>SIMPERL, E.<|>Groth, P. and Simperl, E. co-authored the paper \"Talking datasetsā€“understanding data sensemaking behaviours\"<|>8)\n##\n(\"relationship\"<|>KURATOV, Y.<|>BULATOV, A.<|>Kuratov, Y. and Bulatov, A. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>KURATOV, Y.<|>ANOKHIN, P.<|>Kuratov, Y. and Anokhin, P. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>KURATOV, Y.<|>SOROKIN, D.<|>Kuratov, Y. and Sorokin, D. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>KURATOV, Y.<|>SOROKIN, A.<|>Kuratov, Y. and Sorokin, A. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>KURATOV, Y.<|>BURTSEV, M.<|>Kuratov, Y. and Burtsev, M. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>BULATOV, A.<|>ANOKHIN, P.<|>Bulatov, A. and Anokhin, P. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>BULATOV, A.<|>SOROKIN, D.<|>Bulatov, A. and Sorokin, D. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>BULATOV, A.<|>SOROKIN, A.<|>Bulatov, A. and Sorokin, A. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>BULATOV, A.<|>BURTSEV, M.<|>Bulatov, A. and Burtsev, M. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>ANOKHIN, P.<|>SOROKIN, D.<|>Anokhin, P. and Sorokin, D. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>ANOKHIN, P.<|>SOROKIN, A.<|>Anokhin, P. and Sorokin, A. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>ANOKHIN, P.<|>BURTSEV, M.<|>Anokhin, P. and Burtsev, M. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>SOROKIN, D.<|>SOROKIN, A.<|>Sorokin, D. and Sorokin, A. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>SOROKIN, D.<|>BURTSEV, M.<|>Sorokin, D. and Burtsev, M. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>SOROKIN, A.<|>BURTSEV, M.<|>Sorokin, A. and Burtsev, M. co-authored the paper \"In search of needles in a 11m haystack: Recurrent memory finds what llms miss\"<|>8)\n##\n(\"relationship\"<|>LANGCHAIN<|>LANGCHAIN GRAPHS<|>LangChain developed Langchain graphs<|>9)\n##\n(\"relationship\"<|>LASKAR, M. T. R.<|>HOQUE, E.<|>Laskar, M. T. R. and Hoque, E. co-authored the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<|>8)\n##\n(\"relationship\"<|>LASKAR, M. T. R.<|>HUANG, J.<|>Laskar, M. T. R. and Huang, J. co-authored the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<|>8)\n##\n(\"relationship\"<|>HOQUE, E.<|>HUANG, J.<|>Hoque, E. and Huang, J. co-authored the paper \"Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models\"<|>8)\n<|COMPLETE|>\n#############################\n\n\n\n-Real Data-\n######################\ntext: Xiv:2402.15301 .\nZheng, L., Chiang, W.-L., Sheng, Y ., Zhuang, S., Wu, Z., Zhuang, Y ., Lin, Z., Li, Z., Li, D., Xing,\nE., et al. (2024). Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in Neural\nInformation Processing Systems , 36.\n15\n######################\noutput:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 40 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 40 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 40 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 40 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 40 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 40 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 37 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 37 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 34 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 34 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 32 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 32 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 32 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 32 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 30 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 30 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 28 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 28 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 25 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 25 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 21 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 21 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 20 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 20 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 11 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 11 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 9 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 9 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 1 second. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 1 second. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 32 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 32 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 10 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 10 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 9 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 9 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "MANY entities and relationships were missed in the last extraction. Remember to ONLY emit entities that match any of the previously extracted types. Add them below using the same format:\n"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 17 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 17 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n280,TRAVIS KELCE,\"Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry and his high-profile personal life.\",6\n293,SPORTS,\"The entity \"\"SPORTS\"\" refers to a sector within the entertainment industry that encompasses athletic events and competitions. This sector includes public figures involved in sports, such as athletes, coaches, and sports commentators.\",4\n\n\n-----Claims-----\nhuman_readable_id,subject_id,type,status,description\n34,TRAVIS KELCE,PUBLIC INTEREST,TRUE,Travis Kelce is frequently mentioned in entertainment articles due to his high-profile status and the publicā€™s interest in his career and personal life.\n39,TRAVIS KELCE,FREQUENT MENTIONS,TRUE,Travis Kelce is repeatedly mentioned across various entertainment articles.\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n142,GRAPH RAG,TRAVIS KELCE,Graph RAG mentions Travis Kelce as a prominent public figure,96\n574,ENTERTAINMENT INDUSTRY,TRAVIS KELCE,Travis Kelce is a significant figure in the entertainment industry,25\n585,ENTERTAINMENT INDUSTRY,SPORTS,Sports is a sector within the entertainment industry,23\n394,NAIVE RAG,TRAVIS KELCE,Naive RAG mentions Travis Kelce as a public figure,19\n598,TRAVIS KELCE,ANSWER 2,Travis Kelce is one of the specific public figures mentioned in Answer 2.,16\n608,SPORTS,ANSWER 1,Answer 1 includes public figures from the sports sector.,15\n609,SPORTS,ANSWER 2,Answer 2 focuses on public figures primarily from the sports sector.,14\n570,ENTERTAINMENT ARTICLES,TRAVIS KELCE,Travis Kelce is frequently mentioned in entertainment articles,12\n599,TRAVIS KELCE,SPORTS,Travis Kelce is a public figure in the sports sector.,10\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 17 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 17 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n216,MT-BENCH,\"MT-BENCH is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical tool in evaluating the performance of models in retrieving accurate and relevant information from a broad range of topics. MT-Bench is prominently featured in the academic paper titled \"\"Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena,\"\" where it is utilized to assess the capabilities of large language models in a structured and rigorous manner. This dataset plays a significant role in advancing the field of Natural Language Processing and Information Retrieval by providing a standardized metric for model comparison and evaluation.\",12\n223,ZHENG ET AL.,Authors associated with the MT-Bench dataset,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n495,MT-BENCH,\"LI, Z.\",\"Li, Z. is an author of the paper that discusses MT-Bench\",32\n491,MT-BENCH,\"ZHUANG, S.\",\"Zhuang, S. is an author of the paper that discusses MT-Bench\",27\n488,MT-BENCH,\"ZHENG, L.\",\"Zheng, L. is an author of the paper that discusses MT-Bench\",24\n489,MT-BENCH,\"CHIANG, W.-L.\",\"Chiang, W.-L. is an author of the paper that discusses MT-Bench\",24\n490,MT-BENCH,\"SHENG, Y.\",\"Sheng, Y. is an author of the paper that discusses MT-Bench\",24\n492,MT-BENCH,\"WU, Z.\",\"Wu, Z. is an author of the paper that discusses MT-Bench\",24\n493,MT-BENCH,\"ZHUANG, Y.\",\"Zhuang, Y. is an author of the paper that discusses MT-Bench\",24\n494,MT-BENCH,\"LIN, Z.\",\"Lin, Z. is an author of the paper that discusses MT-Bench\",24\n496,MT-BENCH,\"LI, D.\",\"Li, D. is an author of the paper that discusses MT-Bench\",24\n497,MT-BENCH,\"XING, E.\",\"Xing, E. is an author of the paper that discusses MT-Bench\",24\n498,MT-BENCH,CHATBOT ARENA,\"MT-Bench and Chatbot Arena are both tools used in the paper \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\"\",23\n487,MT-BENCH,ZHENG ET AL.,Zheng et al. are the authors associated with the MT-Bench dataset,13\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n688,\"LIANG, Y.\",\"Liang, Y. is an author of the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",8\n689,\"MENG, F.\",\"Meng, F. is an author of the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",8\n691,\"SHI, H.\",\"Shi, H. is an author of the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",8\n690,\"SUN, Z.\",\"Sun, Z. is an author of the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",8\n687,\"WANG, J.\",\"Wang, J. is an author of the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",8\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n1069,\"LIANG, Y.\",\"LI, Z.\",\"Liang, Y. and Li, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",28\n1075,\"MENG, F.\",\"LI, Z.\",\"Meng, F. and Li, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",28\n1084,\"SHI, H.\",\"LI, Z.\",\"Shi, H. and Li, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",28\n1080,\"SUN, Z.\",\"LI, Z.\",\"Sun, Z. and Li, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",28\n1062,\"WANG, J.\",\"LI, Z.\",\"Wang, J. and Li, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",28\n1070,\"LIANG, Y.\",\"XU, J.\",\"Liang, Y. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1071,\"LIANG, Y.\",\"QU, J.\",\"Liang, Y. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1072,\"LIANG, Y.\",\"ZHOU, J.\",\"Liang, Y. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1076,\"MENG, F.\",\"XU, J.\",\"Meng, F. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1077,\"MENG, F.\",\"QU, J.\",\"Meng, F. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1078,\"MENG, F.\",\"ZHOU, J.\",\"Meng, F. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1085,\"SHI, H.\",\"XU, J.\",\"Shi, H. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1086,\"SHI, H.\",\"QU, J.\",\"Shi, H. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1087,\"SHI, H.\",\"ZHOU, J.\",\"Shi, H. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1081,\"SUN, Z.\",\"XU, J.\",\"Sun, Z. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1082,\"SUN, Z.\",\"QU, J.\",\"Sun, Z. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1083,\"SUN, Z.\",\"ZHOU, J.\",\"Sun, Z. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1063,\"WANG, J.\",\"XU, J.\",\"Wang, J. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1064,\"WANG, J.\",\"QU, J.\",\"Wang, J. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1065,\"WANG, J.\",\"ZHOU, J.\",\"Wang, J. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1066,\"LIANG, Y.\",\"MENG, F.\",\"Liang, Y. and Meng, F. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",16\n1067,\"LIANG, Y.\",\"SUN, Z.\",\"Liang, Y. and Sun, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",16\n1068,\"LIANG, Y.\",\"SHI, H.\",\"Liang, Y. and Shi, H. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",16\n1058,\"WANG, J.\",\"LIANG, Y.\",\"Wang, J. and Liang, Y. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",16\n1073,\"MENG, F.\",\"SUN, Z.\",\"Meng, F. and Sun, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",16\n1074,\"MENG, F.\",\"SHI, H.\",\"Meng, F. and Shi, H. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",16\n1059,\"WANG, J.\",\"MENG, F.\",\"Wang, J. and Meng, F. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",16\n1061,\"WANG, J.\",\"SHI, H.\",\"Wang, J. and Shi, H. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",16\n1079,\"SUN, Z.\",\"SHI, H.\",\"Sun, Z. and Shi, H. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",16\n1060,\"WANG, J.\",\"SUN, Z.\",\"Wang, J. and Sun, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",16\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n692,\"LI, Z.\",\"Li, Z. is an author who has contributed to the field of Natural Language Processing and Information Retrieval through their work on evaluating language models. Specifically, Li, Z. has co-authored the paper titled \"\"Is ChatGPT a Good NLG Evaluator? A Preliminary Study,\"\" which explores the effectiveness of ChatGPT as a natural language generation evaluator. Additionally, Li, Z. has co-authored another paper, \"\"Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena,\"\" which examines the performance of large language models in evaluative roles using specific benchmarking tools. These contributions highlight Li, Z.'s active involvement in advancing the understanding and assessment of language models within the academic community.\",20\n693,\"XU, J.\",\"Xu, J. is an author of the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",9\n694,\"QU, J.\",\"Qu, J. is an author of the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",9\n695,\"ZHOU, J.\",\"Zhou, J. is an author of the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",9\n698,H.,\"H. is an author of the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",4\n\n\n-----Claims-----\nhuman_readable_id,subject_id,type,status,description\n163,\"LI, Z.\",PUBLICATION,TRUE,\"Li, Z. is an author of the publication titled \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\" published in 2023.\"\n199,\"LI, Z.\",AUTHORSHIP,TRUE,\"Li, Z. is listed as an author of the paper titled \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\" published in Advances in Neural Information Processing Systems, 36.\"\n164,\"XU, J.\",PUBLICATION,TRUE,\"Xu, J. is an author of the publication titled \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\" published in 2023.\"\n165,\"QU, J.\",PUBLICATION,TRUE,\"Qu, J. is an author of the publication titled \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\" published in 2023.\"\n166,\"ZHOU, J.\",PUBLICATION,TRUE,\"Zhou, J. is an author of the publication titled \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\" published in 2023.\"\n162,H.,PUBLICATION,TRUE,\"H. is an author of the publication titled \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\" published in 2023.\"\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n1095,\"LI, Z.\",\"ZHUANG, S.\",\"Zhuang, S. and Li, Z. co-authored the paper \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\"\",35\n1092,\"LI, Z.\",\"ZHENG, L.\",\"Zheng, L. and Li, Z. co-authored the paper \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\"\",32\n1093,\"LI, Z.\",\"CHIANG, W.-L.\",\"Chiang, W.-L. and Li, Z. co-authored the paper \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\"\",32\n1094,\"LI, Z.\",\"SHENG, Y.\",\"Sheng, Y. and Li, Z. co-authored the paper \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\"\",32\n1096,\"LI, Z.\",\"WU, Z.\",\"Wu, Z. and Li, Z. co-authored the paper \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\"\",32\n1097,\"LI, Z.\",\"ZHUANG, Y.\",\"Zhuang, Y. and Li, Z. co-authored the paper \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\"\",32\n1098,\"LI, Z.\",\"LIN, Z.\",\"Lin, Z. and Li, Z. co-authored the paper \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\"\",32\n1099,\"LI, Z.\",\"LI, D.\",\"Li, Z. and Li, D. co-authored the paper \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\"\",32\n1100,\"LI, Z.\",\"XING, E.\",\"Li, Z. and Xing, E. co-authored the paper \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\"\",32\n495,MT-BENCH,\"LI, Z.\",\"Li, Z. is an author of the paper that discusses MT-Bench\",32\n1101,\"LI, Z.\",CHATBOT ARENA,\"Li, Z. is an author of the paper that discusses Chatbot Arena\",31\n1088,\"LI, Z.\",\"XU, J.\",\"Li, Z. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",29\n1089,\"LI, Z.\",\"QU, J.\",\"Li, Z. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",29\n1090,\"LI, Z.\",\"ZHOU, J.\",\"Li, Z. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",29\n1062,\"WANG, J.\",\"LI, Z.\",\"Wang, J. and Li, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",28\n1069,\"LIANG, Y.\",\"LI, Z.\",\"Liang, Y. and Li, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",28\n1075,\"MENG, F.\",\"LI, Z.\",\"Meng, F. and Li, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",28\n1080,\"SUN, Z.\",\"LI, Z.\",\"Sun, Z. and Li, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",28\n1084,\"SHI, H.\",\"LI, Z.\",\"Shi, H. and Li, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",28\n1091,\"LI, Z.\",H.,\"H. and Li, Z. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",24\n1105,\"QU, J.\",\"ZHOU, J.\",\"Qu, J. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",18\n1102,\"XU, J.\",\"QU, J.\",\"Xu, J. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",18\n1103,\"XU, J.\",\"ZHOU, J.\",\"Xu, J. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",18\n1064,\"WANG, J.\",\"QU, J.\",\"Wang, J. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1071,\"LIANG, Y.\",\"QU, J.\",\"Liang, Y. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1077,\"MENG, F.\",\"QU, J.\",\"Meng, F. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1082,\"SUN, Z.\",\"QU, J.\",\"Sun, Z. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1086,\"SHI, H.\",\"QU, J.\",\"Shi, H. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1063,\"WANG, J.\",\"XU, J.\",\"Wang, J. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1070,\"LIANG, Y.\",\"XU, J.\",\"Liang, Y. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1076,\"MENG, F.\",\"XU, J.\",\"Meng, F. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1081,\"SUN, Z.\",\"XU, J.\",\"Sun, Z. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1085,\"SHI, H.\",\"XU, J.\",\"Shi, H. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1065,\"WANG, J.\",\"ZHOU, J.\",\"Wang, J. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1072,\"LIANG, Y.\",\"ZHOU, J.\",\"Liang, Y. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1078,\"MENG, F.\",\"ZHOU, J.\",\"Meng, F. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1083,\"SUN, Z.\",\"ZHOU, J.\",\"Sun, Z. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1087,\"SHI, H.\",\"ZHOU, J.\",\"Shi, H. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",17\n1104,\"XU, J.\",H.,\"H. and Xu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",13\n1106,\"QU, J.\",H.,\"H. and Qu, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",13\n1107,\"ZHOU, J.\",H.,\"H. and Zhou, J. co-authored the paper \"\"Is chatgpt a good nlg evaluator? a preliminary study\"\"\",13\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n108,PARTIAL ANSWERS,Intermediate answers generated from community summaries before being combined into a final global answer,2\n109,FINAL GLOBAL ANSWER,The comprehensive answer generated by combining all relevant partial answers,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n191,COMMUNITY SUMMARIES,PARTIAL ANSWERS,Partial answers are generated from community summaries,28\n405,PARTIAL ANSWERS,FINAL GLOBAL ANSWER,Final global answer is generated by combining all relevant partial answers,3\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n96,NAIVE RAG,\"Naive RAG is a basic retrieval-augmented generation (RAG) method used as a baseline for comparison in text generation tasks. It converts documents to text, splits them into chunks, and embeds these chunks into a vector space for query matching. While it produces the most direct responses, it is outperformed by global approaches in terms of comprehensiveness and diversity. Naive RAG is also noted for listing public figures mentioned in entertainment articles, focusing on their professional achievements and personal lives.\",13\n27,COMPREHENSIVENESS,\"COMPREHENSIVENESS is a metric used to evaluate the quality of generated responses by measuring how much detail an answer provides to cover all aspects and details of a question. It assesses the completeness and thoroughness of answers, ensuring that they encompass all relevant information. This metric is particularly important in evaluating the summarization approach, focusing on the completeness of the summary. In practical applications, such as evaluating Podcast transcripts and News articles, comprehensiveness has shown win rates between 72-83% and 72-80%, respectively.\",9\n288,DECISION,A metric used to determine the winner in the comparison of generated responses,3\n326,\"GAO ET AL., 2023\",\"\"\"GAO ET AL., 2023\"\" is a paper published in 2023 by Gao et al. that delves into advanced Retrieval-Augmented Generation (RAG) techniques, specifically where the index is a knowledge graph. The publication also touches upon naive RAG approaches, providing a comprehensive examination of both advanced and basic methodologies within the domain of Natural Language Processing and Information Retrieval.\",3\n318,GLOBAL APPROACHES,Approaches that consistently outperformed the naive RAG in comprehensiveness and diversity metrics,3\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n111,GRAPH RAG,NAIVE RAG,\"Graph RAG is compared to naive RAG in the evaluation. In this comparison, Graph RAG outperformed naive RAG on comprehensiveness and diversity.\",103\n104,GRAPH RAG,COMPREHENSIVENESS,\"Graph RAG is an approach that is evaluated for its comprehensiveness, a target quality used to assess its effectiveness. The method aims to improve the comprehensiveness of generated answers, ensuring that the information provided is thorough and complete. This evaluation metric is crucial in determining the success of Graph RAG in producing detailed and accurate responses.\",99\n145,GRAPH RAG,DECISION,Graph RAG is determined to be the winner based on the decision metric,93\n162,GRAPH RAG,\"GAO ET AL., 2023\",\"Graph RAG incorporates concepts from Gao et al., 2023\",93\n88,LLM,NAIVE RAG,LLM uses Naive RAG to list public figures mentioned in entertainment articles,45\n40,RAG,NAIVE RAG,Naive RAG is a specific implementation of RAG,44\n49,RAG,\"GAO ET AL., 2023\",The paper by Gao et al. discusses advanced RAG where the index is a knowledge graph,34\n227,COMPREHENSIVENESS,PODCAST TRANSCRIPTS,Comprehensiveness is used to evaluate the thoroughness of the generated answers for podcast transcripts,28\n226,COMPREHENSIVENESS,NEWS ARTICLE DATASET,Comprehensiveness is used to evaluate the thoroughness of the generated answers for news articles,25\n224,COMPREHENSIVENESS,NAIVE RAG,Naive RAG is evaluated for comprehensiveness,22\n223,COMPREHENSIVENESS,LLM EVALUATOR,The LLM evaluator assesses answers based on the comprehensiveness metric,20\n393,NAIVE RAG,TAYLOR SWIFT,Naive RAG mentions Taylor Swift as a public figure,19\n394,NAIVE RAG,TRAVIS KELCE,Naive RAG mentions Travis Kelce as a public figure,19\n395,NAIVE RAG,BRITNEY SPEARS,Naive RAG mentions Britney Spears as a public figure,19\n396,NAIVE RAG,JUSTIN TIMBERLAKE,Naive RAG mentions Justin Timberlake as a public figure,19\n228,COMPREHENSIVENESS,CONTEXT WINDOW SIZE,The smallest context window size (8k) was universally better for comprehensiveness,18\n400,NAIVE RAG,SS,SS represents naive RAG in the analysis,17\n397,NAIVE RAG,DECISION,Naive RAG is determined to be the loser based on the decision metric,16\n401,NAIVE RAG,\"GAO ET AL., 2023\",\"Gao et al., 2023 discusses naive RAG approaches\",16\n398,NAIVE RAG,GLOBAL APPROACHES,Global approaches consistently outperformed the naive RAG,16\n399,NAIVE RAG,DIRECTNESS,Naive RAG produces the most direct responses,16\n225,COMPREHENSIVENESS,DECISION,Comprehensiveness is a metric used to determine the decision,12\n229,COMPREHENSIVENESS,FINAL EVALUATION,The final evaluation prioritized comprehensiveness in answers,12\n230,COMPREHENSIVENESS,GLOBAL APPROACHES,Global approaches achieved higher comprehensiveness win rates,12\n236,DIVERSITY,GLOBAL APPROACHES,Global approaches achieved higher diversity win rates,10\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n261,ENTERTAINMENT INDUSTRY,\"The **ENTERTAINMENT INDUSTRY** is a multifaceted sector that encompasses various forms of entertainment, including movies, music, television, sports, and digital media. This industry is characterized by its diverse range of content and mediums, which collectively contribute to its broad appeal and significant cultural impact. The entertainment industry plays a crucial role in shaping public opinion, trends, and cultural norms through its extensive reach and influence across different platforms and genres.\",19\n283,ACTORS AND DIRECTORS,\"A category of public figures in the entertainment industry, including those involved in film and television\",1\n285,ATHLETES AND COACHES,\"A category of public figures in the entertainment industry, including those involved in sports\",1\n295,CULTURAL NARRATIVES,A category within the entertainment industry that includes stories and themes that shape culture,1\n284,MUSICIANS AND EXECUTIVES,\"A category of public figures in the entertainment industry, including those involved in music\",1\n286,INFLUENCERS AND ENTREPRENEURS,\"A category of public figures in the entertainment industry, including those involved in digital media and business\",1\n287,PUBLIC FIGURES IN CONTROVERSY,A category of public figures in the entertainment industry who are involved in controversies,1\n296,TRENDS,A category within the entertainment industry that includes popular movements and styles,1\n297,SOCIAL DISCUSSIONS,A category within the entertainment industry that includes public conversations and debates,1\n298,PUBLIC DISCOURSE,A category within the entertainment industry that includes formal discussions and communications,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n573,ENTERTAINMENT INDUSTRY,TAYLOR SWIFT,Taylor Swift is a significant figure in the entertainment industry,25\n574,ENTERTAINMENT INDUSTRY,TRAVIS KELCE,Travis Kelce is a significant figure in the entertainment industry,25\n575,ENTERTAINMENT INDUSTRY,BRITNEY SPEARS,Britney Spears is a significant figure in the entertainment industry,25\n576,ENTERTAINMENT INDUSTRY,JUSTIN TIMBERLAKE,Justin Timberlake is a significant figure in the entertainment industry,25\n584,ENTERTAINMENT INDUSTRY,MUSIC,Music is a sector within the entertainment industry,25\n568,ENTERTAINMENT ARTICLES,ENTERTAINMENT INDUSTRY,Entertainment articles cover topics related to the entertainment industry,25\n585,ENTERTAINMENT INDUSTRY,SPORTS,Sports is a sector within the entertainment industry,23\n582,ENTERTAINMENT INDUSTRY,FILM,Film is a sector within the entertainment industry,21\n583,ENTERTAINMENT INDUSTRY,TELEVISION,Television is a sector within the entertainment industry,21\n586,ENTERTAINMENT INDUSTRY,DIGITAL MEDIA,Digital Media is a sector within the entertainment industry,21\n577,ENTERTAINMENT INDUSTRY,ACTORS AND DIRECTORS,Actors and Directors are a category within the entertainment industry,20\n579,ENTERTAINMENT INDUSTRY,ATHLETES AND COACHES,Athletes and Coaches are a category within the entertainment industry,20\n587,ENTERTAINMENT INDUSTRY,CULTURAL NARRATIVES,Cultural Narratives are a category within the entertainment industry,20\n578,ENTERTAINMENT INDUSTRY,MUSICIANS AND EXECUTIVES,Musicians and Executives are a category within the entertainment industry,20\n580,ENTERTAINMENT INDUSTRY,INFLUENCERS AND ENTREPRENEURS,Influencers and Entrepreneurs are a category within the entertainment industry,20\n581,ENTERTAINMENT INDUSTRY,PUBLIC FIGURES IN CONTROVERSY,Public Figures in Controversy are a category within the entertainment industry,20\n588,ENTERTAINMENT INDUSTRY,TRENDS,Trends are a category within the entertainment industry,20\n589,ENTERTAINMENT INDUSTRY,SOCIAL DISCUSSIONS,Social Discussions are a category within the entertainment industry,20\n590,ENTERTAINMENT INDUSTRY,PUBLIC DISCOURSE,Public Discourse is a category within the entertainment industry,20\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n281,BRITNEY SPEARS,\"Britney Spears is a public figure frequently mentioned in entertainment articles, known for her significant contributions to the music industry and her high-profile personal life.\",6\n282,JUSTIN TIMBERLAKE,\"Justin Timberlake is a public figure frequently mentioned in entertainment articles, known for his significant contributions to the music industry and his high-profile personal life.\",6\n279,TAYLOR SWIFT,\"Taylor Swift is a public figure frequently mentioned in entertainment articles, known for her contributions to the music industry and her high-profile personal life.\",6\n300,ANSWER 2,\"\"\"ANSWER 2\"\" is a generated answer for the example question in the News article dataset. It focuses on a smaller group of public figures, primarily from the music industry and sports, and relies heavily on a single source for data. \"\"ANSWER 2\"\" provides concise explanations for the frequent mentions of specific public figures such as Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake.\",10\n260,ENTERTAINMENT ARTICLES,\"ENTERTAINMENT ARTICLES is a collection of articles focused on the entertainment industry. This dataset consists of articles related to various aspects of the entertainment sector, providing a comprehensive resource for understanding trends, developments, and key topics within this field.\",6\n292,MUSIC,\"MUSIC is a sector within the entertainment industry that encompasses musical performances and recordings. This sector includes public figures involved in the music industry, such as singers, musicians, and producers.\",6\n\n\n-----Claims-----\nhuman_readable_id,subject_id,type,status,description\n33,TAYLOR SWIFT,PUBLIC INTEREST,TRUE,Taylor Swift is frequently mentioned in entertainment articles due to her high-profile status and the publicā€™s interest in her career and personal life.\n38,TAYLOR SWIFT,FREQUENT MENTIONS,TRUE,Taylor Swift is repeatedly mentioned across various entertainment articles.\n35,BRITNEY SPEARS,PUBLIC INTEREST,TRUE,Britney Spears is frequently mentioned in entertainment articles due to her high-profile status and the publicā€™s interest in her career and personal life.\n40,BRITNEY SPEARS,FREQUENT MENTIONS,TRUE,Britney Spears is repeatedly mentioned across various entertainment articles.\n36,JUSTIN TIMBERLAKE,PUBLIC INTEREST,TRUE,Justin Timberlake is frequently mentioned in entertainment articles due to his high-profile status and the publicā€™s interest in his career and personal life.\n41,JUSTIN TIMBERLAKE,FREQUENT MENTIONS,TRUE,Justin Timberlake is repeatedly mentioned across various entertainment articles.\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n143,GRAPH RAG,BRITNEY SPEARS,Graph RAG mentions Britney Spears as a prominent public figure,96\n144,GRAPH RAG,JUSTIN TIMBERLAKE,Graph RAG mentions Justin Timberlake as a prominent public figure,96\n141,GRAPH RAG,TAYLOR SWIFT,Graph RAG mentions Taylor Swift as a prominent public figure,96\n616,ANSWER 2,NEWS ARTICLE DATASET,ANSWER 2 is a generated answer for a question in the NEWS ARTICLE DATASET. It relies heavily on a single source from the NEWS ARTICLE DATASET for data.,26\n575,ENTERTAINMENT INDUSTRY,BRITNEY SPEARS,Britney Spears is a significant figure in the entertainment industry,25\n568,ENTERTAINMENT ARTICLES,ENTERTAINMENT INDUSTRY,Entertainment articles cover topics related to the entertainment industry,25\n576,ENTERTAINMENT INDUSTRY,JUSTIN TIMBERLAKE,Justin Timberlake is a significant figure in the entertainment industry,25\n584,ENTERTAINMENT INDUSTRY,MUSIC,Music is a sector within the entertainment industry,25\n573,ENTERTAINMENT INDUSTRY,TAYLOR SWIFT,Taylor Swift is a significant figure in the entertainment industry,25\n395,NAIVE RAG,BRITNEY SPEARS,Naive RAG mentions Britney Spears as a public figure,19\n396,NAIVE RAG,JUSTIN TIMBERLAKE,Naive RAG mentions Justin Timberlake as a public figure,19\n393,NAIVE RAG,TAYLOR SWIFT,Naive RAG mentions Taylor Swift as a public figure,19\n606,MUSIC,ANSWER 1,Answer 1 includes public figures from the music sector.,17\n596,TAYLOR SWIFT,ANSWER 2,Taylor Swift is one of the specific public figures mentioned in Answer 2.,16\n598,TRAVIS KELCE,ANSWER 2,Travis Kelce is one of the specific public figures mentioned in Answer 2.,16\n600,BRITNEY SPEARS,ANSWER 2,Britney Spears is one of the specific public figures mentioned in Answer 2.,16\n602,JUSTIN TIMBERLAKE,ANSWER 2,Justin Timberlake is one of the specific public figures mentioned in Answer 2.,16\n607,MUSIC,ANSWER 2,Answer 2 focuses on public figures primarily from the music sector.,16\n566,PUBLIC FIGURES,ANSWER 2,\"Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports.\",15\n615,ANSWER 2,NAƏVE RAG,\"Answer 2 was generated using the NaĆÆve RAG method, which directly lists specific public figures who are repeatedly mentioned across various entertainment articles.\",14\n609,SPORTS,ANSWER 2,Answer 2 focuses on public figures primarily from the sports sector.,14\n617,ANSWER 2,DATA SOURCES,Answer 2 relies heavily on a single data source.,12\n601,BRITNEY SPEARS,MUSIC,Britney Spears is a public figure in the music sector.,12\n571,ENTERTAINMENT ARTICLES,BRITNEY SPEARS,Britney Spears is frequently mentioned in entertainment articles,12\n569,ENTERTAINMENT ARTICLES,TAYLOR SWIFT,Taylor Swift is frequently mentioned in entertainment articles,12\n570,ENTERTAINMENT ARTICLES,TRAVIS KELCE,Travis Kelce is frequently mentioned in entertainment articles,12\n572,ENTERTAINMENT ARTICLES,JUSTIN TIMBERLAKE,Justin Timberlake is frequently mentioned in entertainment articles,12\n603,JUSTIN TIMBERLAKE,MUSIC,Justin Timberlake is a public figure in the music sector.,12\n597,TAYLOR SWIFT,MUSIC,Taylor Swift is a public figure in the music sector.,12\n564,PUBLIC FIGURES,ENTERTAINMENT ARTICLES,Public figures are repeatedly mentioned across various entertainment articles,11\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n299,ANSWER 1,\"Answer 1 provides a varied and rich response by covering a wide range of public figures from different sectors of the entertainment industry, including film, television, music, sports, gaming, and digital media. It offers insights into the contributions and influence of these figures, as well as controversies and their impact on public discourse. The answer also cites specific data sources for each mentioned figure, indicating a diverse range of evidence to support the claims.\",11\n294,DIGITAL MEDIA,\"DIGITAL MEDIA is a sector within the entertainment industry that encompasses online content and social media. This sector includes public figures involved in online platforms, such as influencers, content creators, and digital marketers. These individuals play a significant role in shaping digital landscapes through their engagement with audiences and their ability to leverage various online tools and platforms for content dissemination and marketing purposes.\",2\n290,FILM,\"The entity \"\"FILM\"\" refers to a sector within the entertainment industry that encompasses movies and cinema. This sector includes public figures involved in the movie industry, such as actors, directors, and producers.\",2\n291,TELEVISION,\"The entity \"\"TELEVISION\"\" refers to a sector within the entertainment industry that encompasses TV shows and series. This sector includes public figures involved in TV shows, such as actors, hosts, and producers.\",2\n305,DATA SOURCES,Data sources are references or reports used to support claims about public figures and their influence.,2\n304,GAMING,\"The gaming sector includes public figures involved in the gaming industry, including gamers, developers, and streamers.\",1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n146,GRAPH RAG,ANSWER 1,\"Answer 1 was generated using the Graph RAG method, which provides a comprehensive and structured overview of public figures across various sectors of the entertainment industry.\",101\n611,ANSWER 1,NEWS ARTICLE DATASET,Answer 1 cites specific data sources from the News article dataset for each mentioned figure.,27\n586,ENTERTAINMENT INDUSTRY,DIGITAL MEDIA,Digital Media is a sector within the entertainment industry,21\n582,ENTERTAINMENT INDUSTRY,FILM,Film is a sector within the entertainment industry,21\n583,ENTERTAINMENT INDUSTRY,TELEVISION,Television is a sector within the entertainment industry,21\n606,MUSIC,ANSWER 1,Answer 1 includes public figures from the music sector.,17\n565,PUBLIC FIGURES,ANSWER 1,Answer 1 covers a wide range of public figures from different sectors of the entertainment industry.,16\n608,SPORTS,ANSWER 1,Answer 1 includes public figures from the sports sector.,15\n612,ANSWER 1,CONTROVERSIES,Answer 1 provides insights into controversies involving public figures and their impact on public discourse.,13\n614,ANSWER 1,DATA SOURCES,Answer 1 cites specific data sources for each mentioned figure.,13\n604,FILM,ANSWER 1,Answer 1 includes public figures from the film sector.,13\n605,TELEVISION,ANSWER 1,Answer 1 includes public figures from the television sector.,13\n610,DIGITAL MEDIA,ANSWER 1,Answer 1 includes public figures from the digital media sector.,13\n613,ANSWER 1,GAMING,Answer 1 includes public figures from the gaming sector.,12\n617,ANSWER 2,DATA SOURCES,Answer 2 relies heavily on a single data source.,12\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n13,GRAPH RAG,\"Graph RAG (Retrieval-Augmented Generation) is a sophisticated method that leverages the natural modularity of graphs to partition data for global summarization tasks. This approach integrates multiple stages and concepts, including knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS), to support human sensemaking over text corpora. It is particularly effective in providing comprehensive and structured overviews of public figures across various sectors of the entertainment industry, as well as generating answers for questions in the News article dataset.\n\nGraph RAG employs a high-level data flow and pipeline for processing and summarizing text, combining both global and local approaches to optimize token usage in text generation tasks. It uses community summaries to improve answer comprehensiveness and diversity while requiring fewer tokens compared to traditional source text summarization methods. This method has been shown to outperform naive RAG in terms of comprehensiveness and diversity in text generation tasks.\n\nA specific implementation of Graph RAG involves using four levels of graph communities, incorporating concepts from other systems such as self-memory and parallel generation of community answers. This multi-stage mechanism allows for the comparison of multiple conditions, enhancing the overall effectiveness of the summarization process.\n\nGraph RAG, launched by NebulaGraph, is a retrieval-augmented generation technology based on knowledge graphs. It combines global summarization with graph-based text indexing to answer questions over private text corpora, making it a versatile tool for various text analysis and summarization applications.\",90\n28,DIVERSITY,\"DIVERSITY is a metric used to evaluate the variety and richness of answers generated in response to a question. It measures how varied and rich an answer is in providing different perspectives and insights. This metric is particularly important in assessing the quality of summarization approaches, focusing on the variety of information included in the summary. DIVERSITY is applied to various types of content, including Podcast transcripts, where win rates range from 75-82%, and News articles, with win rates ranging from 62-71%. It is a crucial target quality for evaluating the effectiveness of different methods in generating diverse and informative responses.\",7\n246,CONDITIONS,\"The entity \"\"CONDITIONS\"\" refers to the different scenarios or variables that are compared in an experiment. Specifically, in the context of the analysis, these conditions include Graph RAG, text summarization, and semantic search RAG. These conditions are used to evaluate and compare various aspects of performance and effectiveness within the domain of Natural Language Processing and Information Retrieval.\",3\n98,ACTIVITY-CENTERED SENSEMAKING QUESTIONS,\"Questions generated to evaluate the summarization approach, focusing on understanding activities\",1\n410,ALONSO,A person who contributed to the work mentioned in the acknowledgements,1\n329,ANSWER COMPREHENSIVENESS,Answer comprehensiveness is a measure used to evaluate the completeness of answers provided by different methods,1\n347,\"CHENG ET AL., 2024\",A reference to a publication by Cheng et al. in 2024,1\n107,CORPUS,A large collection of texts or documents used for analysis and summarization,1\n101,DATA FLOW,The high-level process of the Graph RAG approach and pipeline,1\n102,DESIGN PARAMETERS,\"Design parameters are key settings and configurations in the Graph RAG approach. These parameters are crucial as they influence the design of the Graph RAG approach and pipeline, determining the effectiveness and efficiency of the overall system.\",1\n331,ELEMENT EXTRACTION PROMPTS,Element extraction prompts are used to extract specific details in the Graph RAG index,1\n422,ENTITY-BASED GRAPH INDEX,A graph index organized around entities,1\n338,FEDERATED RETRIEVAL-GENERATION (FEB4RAG),A federated strategy for retrieval and generation,1\n352,\"FENG ET AL., 2023\",A reference to a publication by Feng et al. in 2023,1\n97,GLOBAL MAP-REDUCE SUMMARIZATION,A method for summarizing source texts using a map-reduce approach,1\n14,QFS,QFS (Query-Focused Summarization) is a method used to generate summaries based on specific user queries,1\n18,PYTHON,\"Python is a programming language used for implementing both global and local Graph RAG approaches. Additionally, Python is utilized to implement the open-source version of the Graph RAG approach.\",1\n99,HIERARCHICAL LEVEL,The level of detail in community summaries used to answer queries,1\n100,TOKEN COSTS,\"TOKEN COSTS refer to the computational cost measured in tokens used in the summarization process. Specifically, in the context of the Graph RAG (Retrieval-Augmented Generation) approach, token costs denote the number of tokens required for processing text. This metric is crucial for evaluating the efficiency and scalability of text processing methods within the Natural Language Processing and Information Retrieval domain.\",1\n106,QUERY,A specific question or request for information that the summarization methods aim to answer,1\n112,REAL-WORLD DATASETS,\"Datasets that represent real-world information, such as podcast transcripts and news articles\",1\n113,HIERARCHICAL LEVEL OF COMMUNITY SUMMARIES,The level of detail in community summaries used to answer queries,1\n114,SOURCE TEXT SUMMARIZATION,A method that summarizes the original source texts directly,1\n117,HIGH-LEVEL COMMUNITY SUMMARIES,Summaries generated from higher hierarchical levels of the community in the knowledge graph,1\n126,TECHNIQUES,Techniques refer to the specific methods used in the Graph RAG approach,1\n127,IMPLEMENTATION DETAILS,Implementation details are specific configurations and settings used in the Graph RAG approach,1\n270,MULTI-STAGE,A method involving multiple stages or steps,1\n328,ROOT-LEVEL SUMMARIES,Root-level summaries are a type of community summary used in the analysis,1\n337,ITERATIVE RETRIEVAL-GENERATION (ITER-RETGEN),A strategy for iterative retrieval and generation,1\n343,TREE OF CLARIFICATIONS,A method for answering multiple interpretations of ambiguous questions by generating a hierarchical structure,1\n348,\"MAO ET AL., 2020\",A reference to a publication by Mao et al. in 2020,1\n349,\"SHAO ET AL., 2023\",A reference to a publication by Shao et al. in 2023,1\n350,\"WANG ET AL., 2024\",A reference to a publication by Wang et al. in 2024,1\n351,\"SU ET AL., 2020\",A reference to a publication by Su et al. in 2020,1\n353,\"TRIVEDI ET AL., 2022\",A reference to a publication by Trivedi et al. in 2022,1\n354,\"KHATTAB ET AL., 2022\",A reference to a publication by Khattab et al. in 2022,1\n355,\"SARTHI ET AL., 2024\",A reference to a publication by Sarthi et al. in 2024,1\n356,\"KIM ET AL., 2023\",A reference to a publication by Kim et al. in 2023,1\n402,RICH TEXT ANNOTATIONS,Annotations that provide detailed information about the text,1\n415,LOCAL GRAPH RAG APPROACHES,Graph RAG approaches that operate in a more localized manner,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n67,LLM,GRAPH RAG,\"Graph RAG utilizes Large Language Models (LLMs) to construct a graph-based text index, enabling the generation of summaries and the answering of queries. In this approach, LLMs play a crucial role in analyzing and generating text based on the information retrieved through the graph structure. Additionally, LLMs leverage the Graph RAG framework to provide comprehensive overviews of public figures in the entertainment industry. This integration of LLMs within Graph RAG enhances the system's ability to process and synthesize complex text data, making it a powerful tool for information retrieval and natural language processing tasks.\",122\n37,RAG,GRAPH RAG,\"Graph RAG is a specific implementation of RAG that combines the strengths of RAG with graph-based text indexing. This method leverages the natural modularity of graphs to partition data, facilitating global summarization. As a specialized approach within the RAG framework, Graph RAG enhances the capabilities of RAG by integrating graph structures to improve the efficiency and effectiveness of text data processing and summarization.\",121\n99,GRAPH RAG,COMMUNITY SUMMARIES,\"Graph RAG is an approach that utilizes community summaries in various capacities to enhance its functionality. Specifically, community summaries are generated within the Graph RAG framework and are employed to generate partial responses. These summaries are also used to compare against source texts, serving as a form of self-memory for the system. By leveraging community summaries, Graph RAG aims to improve the comprehensiveness and diversity of answers. Additionally, Graph RAG incorporates summaries of root-level communities within an entity-based graph index, further enriching its analytical capabilities.\",116\n108,GRAPH RAG,PODCAST TRANSCRIPTS,Podcast transcripts are used as a dataset to evaluate the Graph RAG approach,109\n161,GRAPH RAG,GRAPH INDEX,\"Graph RAG utilizes a self-generated graph index. This self-generated graph index is a crucial component of Graph RAG, enabling it to efficiently manage and retrieve information within its graph-based framework. The use of a self-generated graph index suggests that Graph RAG has an inherent capability to construct and maintain its indexing structure, which likely enhances its performance and adaptability in handling complex data relationships.\",108\n147,GRAPH RAG,NEWS ARTICLE DATASET,Graph RAG is used to generate answers for questions in the News article dataset,106\n109,GRAPH RAG,NEWS ARTICLES,News articles are used as a dataset to evaluate the Graph RAG approach,103\n111,GRAPH RAG,NAIVE RAG,\"Graph RAG is compared to naive RAG in the evaluation. In this comparison, Graph RAG outperformed naive RAG on comprehensiveness and diversity.\",103\n139,GRAPH RAG,LLM EVALUATOR,The Graph RAG mechanism uses an LLM evaluator for head-to-head comparison,101\n146,GRAPH RAG,ANSWER 1,\"Answer 1 was generated using the Graph RAG method, which provides a comprehensive and structured overview of public figures across various sectors of the entertainment industry.\",101\n149,GRAPH RAG,TS,TS represents source text summarization in the Graph RAG analysis,100\n104,GRAPH RAG,COMPREHENSIVENESS,\"Graph RAG is an approach that is evaluated for its comprehensiveness, a target quality used to assess its effectiveness. The method aims to improve the comprehensiveness of generated answers, ensuring that the information provided is thorough and complete. This evaluation metric is crucial in determining the success of Graph RAG in producing detailed and accurate responses.\",99\n134,GRAPH RAG,C1,Graph RAG uses high-level community summaries (C1) to answer user queries,99\n135,GRAPH RAG,C2,Graph RAG uses intermediate-level community summaries (C2) to answer user queries,99\n136,GRAPH RAG,C3,\"Graph RAG utilizes low-level community summaries, represented by C3, to answer user queries. C3 plays a crucial role in the Graph RAG analysis by providing detailed summaries of community structures, which are essential for effectively addressing user inquiries.\",99\n133,GRAPH RAG,C0,\"Graph RAG is a system that utilizes root-level community summaries, denoted as C0, to answer user queries. C0 represents these root-level community summaries within the Graph RAG analysis, serving as a foundational element in the system's ability to map out relationships and understand the structural dynamics within specialized communities.\",98\n105,GRAPH RAG,DIVERSITY,\"Graph RAG is an approach in the domain of Natural Language Processing and Information Retrieval that focuses on improving the diversity of generated answers. Diversity, in this context, is a target quality used to evaluate the performance of the Graph RAG approach. By enhancing the diversity of responses, Graph RAG aims to provide a broader range of answers, thereby improving the overall effectiveness and robustness of the system.\",97\n177,GRAPH RAG,HIERARCHICAL COMMUNITY STRUCTURE,Graph RAG uses a hierarchical community structure,97\n107,GRAPH RAG,COMMUNITY DETECTION ALGORITHMS,Community detection algorithms are used in the Graph RAG approach to partition graphs,96\n110,GRAPH RAG,EMPOWERMENT,Graph RAG is evaluated using the target quality of Empowerment. Empowerment is specifically utilized to assess Graph RAG's capability in aiding users to achieve an informed understanding. This evaluation metric underscores the importance of user comprehension and the effectiveness of the Graph RAG approach in facilitating informed decision-making processes.,96\n138,GRAPH RAG,USER QUERIES,Graph RAG uses different levels of graph communities to answer user queries,96\n141,GRAPH RAG,TAYLOR SWIFT,Graph RAG mentions Taylor Swift as a prominent public figure,96\n142,GRAPH RAG,TRAVIS KELCE,Graph RAG mentions Travis Kelce as a prominent public figure,96\n143,GRAPH RAG,BRITNEY SPEARS,Graph RAG mentions Britney Spears as a prominent public figure,96\n144,GRAPH RAG,JUSTIN TIMBERLAKE,Graph RAG mentions Justin Timberlake as a prominent public figure,96\n113,GRAPH RAG,QUERY-FOCUSED SUMMARIZATION,Query-focused summarization is a method used in the Graph RAG approach,95\n119,GRAPH RAG,GLOBAL SUMMARIZATION,Graph RAG uses global summarization to summarize information from a large dataset,95\n158,GRAPH RAG,MULTI-HOP QUESTION ANSWERING,Graph RAG incorporates concepts used in multi-hop question answering,94\n185,GRAPH RAG,NEBULAGRAPH,NebulaGraph launched the industry-first graph RAG: Retrieval-augmented generation with LLM based on knowledge graphs,94\n137,GRAPH RAG,CONDITIONS,\"Graph RAG is a key entity in the analysis, serving both as a condition being compared and as a tool for comparing multiple conditions. This dual role highlights its significance in the study, where it functions not only as a subject of comparison but also as a methodological framework for evaluating other conditions. The analysis likely involves a detailed examination of various conditions, with Graph RAG playing a central role in facilitating these comparisons.\",93\n100,GRAPH RAG,GLOBAL SENSEMAKING QUESTIONS,Graph RAG is designed to handle global sensemaking questions over large datasets,93\n126,GRAPH RAG,LOW-LEVEL COMMUNITY SUMMARIES,Low-level community summaries are generated in the Graph RAG approach,93\n132,GRAPH RAG,RAG SYSTEMS,Graph RAG is a specific implementation of RAG systems,93\n145,GRAPH RAG,DECISION,Graph RAG is determined to be the winner based on the decision metric,93\n152,GRAPH RAG,WIN RATE,Win rate is used to measure the success rate of Graph RAG in providing comprehensive and diverse answers,93\n162,GRAPH RAG,\"GAO ET AL., 2023\",\"Graph RAG incorporates concepts from Gao et al., 2023\",93\n173,GRAPH RAG,COMMUNITY ANSWERS,Graph RAG generates community answers in parallel,93\n178,GRAPH RAG,EMBEDDING-BASED MATCHING,Graph RAG can operate using embedding-based matching,93\n102,GRAPH RAG,HTTPS://AKA.MS/GRAPHRAG,The open-source implementation of Graph RAG will be available at this URL,92\n103,GRAPH RAG,ENTITY KNOWLEDGE GRAPH,Graph RAG uses an entity knowledge graph to index text,92\n106,GRAPH RAG,KNOWLEDGE GRAPH,Graph RAG uses a knowledge graph for global summarization,92\n122,GRAPH RAG,ACTIVITY-CENTERED SENSEMAKING,Activity-centered sensemaking is used to evaluate the Graph RAG approach,92\n127,GRAPH RAG,INTERMEDIATE-LEVEL COMMUNITY SUMMARIES,Intermediate-level community summaries are generated in the Graph RAG approach,92\n129,GRAPH RAG,PIPELINE,The Graph RAG approach involves a specific pipeline for processing and summarizing text,92\n148,GRAPH RAG,SOURCE TEXTS,Graph RAG is compared with source texts for answer comprehensiveness and diversity,92\n154,GRAPH RAG,SELF-MEMORY (SELFMEM),Graph RAG incorporates the concept of self-memory,92\n157,GRAPH RAG,MULTI-DOCUMENT SUMMARIZATION,Graph RAG incorporates concepts used in multi-document summarization,92\n159,GRAPH RAG,HIERARCHICAL INDEX,Graph RAG uses a hierarchical index,92\n174,GRAPH RAG,GRAPH-FREE APPROACH,Graph RAG is compared to a graph-free approach for global summarization,92\n175,GRAPH RAG,MAP-REDUCE SUMMARIZATION,Graph RAG is compared to map-reduce summarization,92\n179,GRAPH RAG,HYBRID RAG SCHEMES,Graph RAG can be part of hybrid RAG schemes,92\n180,GRAPH RAG,MAP-REDUCE SUMMARIZATION MECHANISMS,Graph RAG can employ map-reduce summarization mechanisms,92\n181,GRAPH RAG,COMMUNITY HIERARCHY,Graph RAG can extend operations across the community hierarchy,92\n114,GRAPH RAG,ACTIVITY-CENTERED SENSEMAKING QUESTIONS,Activity-centered sensemaking questions are used to evaluate the Graph RAG approach,91\n182,GRAPH RAG,ALONSO,Alonso contributed to the work on Graph RAG,91\n151,GRAPH RAG,ANSWER COMPREHENSIVENESS,Answer comprehensiveness is used to evaluate the performance of Graph RAG,91\n163,GRAPH RAG,\"CHENG ET AL., 2024\",\"Graph RAG incorporates concepts from Cheng et al., 2024\",91\n121,GRAPH RAG,CORPUS,Graph RAG uses a corpus for analysis and summarization,91\n117,GRAPH RAG,DATA FLOW,Data flow describes the high-level process of the Graph RAG approach,91\n118,GRAPH RAG,DESIGN PARAMETERS,Design parameters are key settings in the Graph RAG approach and significantly influence the Graph RAG approach and pipeline.,91\n153,GRAPH RAG,ELEMENT EXTRACTION PROMPTS,Element extraction prompts are used in Graph RAG to retain specific details in the index,91\n184,GRAPH RAG,ENTITY-BASED GRAPH INDEX,Graph RAG uses an entity-based graph index,91\n156,GRAPH RAG,FEDERATED RETRIEVAL-GENERATION (FEB4RAG),Graph RAG incorporates the concept of federated retrieval-generation,91\n168,GRAPH RAG,\"FENG ET AL., 2023\",\"Graph RAG incorporates concepts from Feng et al., 2023\",91\n112,GRAPH RAG,GLOBAL MAP-REDUCE SUMMARIZATION,Graph RAG is compared to global map-reduce summarization in the evaluation,91\n98,GRAPH RAG,QFS,Graph RAG is proposed as a method to combine the strengths of RAG and QFS,91\n101,GRAPH RAG,PYTHON,Graph RAG is implemented in Python.,91\n115,GRAPH RAG,HIERARCHICAL LEVEL,Hierarchical level of community summaries is varied to evaluate the Graph RAG approach,91\n116,GRAPH RAG,TOKEN COSTS,\"The \"\"Graph RAG\"\" approach is evaluated in terms of its performance by considering \"\"Token Costs.\"\" Token costs are measured to assess the efficiency of the Graph RAG method, indicating that the computational expense associated with processing tokens is a critical factor in determining the overall effectiveness of this approach.\",91\n120,GRAPH RAG,QUERY,Graph RAG aims to answer specific queries,91\n123,GRAPH RAG,REAL-WORLD DATASETS,Real-world datasets are used to evaluate the Graph RAG approach,91\n124,GRAPH RAG,HIERARCHICAL LEVEL OF COMMUNITY SUMMARIES,Hierarchical level of community summaries is varied to evaluate the Graph RAG approach,91\n125,GRAPH RAG,SOURCE TEXT SUMMARIZATION,Graph RAG is compared to source text summarization in the evaluation,91\n128,GRAPH RAG,HIGH-LEVEL COMMUNITY SUMMARIES,High-level community summaries are generated in the Graph RAG approach,91\n130,GRAPH RAG,TECHNIQUES,Techniques are specific methods used in the Graph RAG approach,91\n131,GRAPH RAG,IMPLEMENTATION DETAILS,Implementation details are specific configurations used in the Graph RAG approach,91\n140,GRAPH RAG,MULTI-STAGE,Graph RAG is a multi-stage mechanism,91\n150,GRAPH RAG,ROOT-LEVEL SUMMARIES,Root-level summaries are used in the Graph RAG analysis,91\n155,GRAPH RAG,ITERATIVE RETRIEVAL-GENERATION (ITER-RETGEN),Graph RAG incorporates the concept of iterative retrieval-generation,91\n160,GRAPH RAG,TREE OF CLARIFICATIONS,Graph RAG incorporates the concept of a tree of clarifications,91\n164,GRAPH RAG,\"MAO ET AL., 2020\",\"Graph RAG incorporates concepts from Mao et al., 2020\",91\n165,GRAPH RAG,\"SHAO ET AL., 2023\",\"Graph RAG incorporates concepts from Shao et al., 2023\",91\n166,GRAPH RAG,\"WANG ET AL., 2024\",\"Graph RAG incorporates concepts from Wang et al., 2024\",91\n167,GRAPH RAG,\"SU ET AL., 2020\",\"Graph RAG incorporates concepts from Su et al., 2020\",91\n169,GRAPH RAG,\"TRIVEDI ET AL., 2022\",\"Graph RAG incorporates concepts from Trivedi et al., 2022\",91\n170,GRAPH RAG,\"KHATTAB ET AL., 2022\",\"Graph RAG incorporates concepts from Khattab et al., 2022\",91\n171,GRAPH RAG,\"SARTHI ET AL., 2024\",\"Graph RAG incorporates concepts from Sarthi et al., 2024\",91\n172,GRAPH RAG,\"KIM ET AL., 2023\",\"Graph RAG incorporates concepts from Kim et al., 2023\",91\n176,GRAPH RAG,RICH TEXT ANNOTATIONS,Graph RAG uses rich text annotations,91\n183,GRAPH RAG,LOCAL GRAPH RAG APPROACHES,Graph RAG includes local graph RAG approaches,91\n233,DIVERSITY,PODCAST TRANSCRIPTS,Diversity is used to evaluate the variety in the generated answers for podcast transcripts,26\n232,DIVERSITY,NEWS ARTICLE DATASET,Diversity is used to evaluate the variety in the generated answers for news articles,23\n231,DIVERSITY,LLM EVALUATOR,The LLM evaluator assesses answers based on the diversity metric,18\n234,DIVERSITY,CONTEXT WINDOW SIZE,The smallest context window size (8k) performed comparably with larger context sizes on diversity,16\n235,DIVERSITY,FINAL EVALUATION,The final evaluation prioritized diversity in answers,10\n236,DIVERSITY,GLOBAL APPROACHES,Global approaches achieved higher diversity win rates,10\n502,TEXT SUMMARIZATION,CONDITIONS,Text summarization is one of the conditions compared in the analysis,6\n504,SEMANTIC SEARCH RAG,CONDITIONS,Semantic search RAG is one of the conditions compared in the analysis,5\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 19 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n73,KNOWLEDGE GRAPH,\"A knowledge graph is a structured representation of information, utilized in the Graph RAG approach for summarization. This structured representation of knowledge is specifically employed in the Graph RAG approach for global summarization, highlighting its role in organizing and integrating information to facilitate comprehensive and coherent summaries.\",2\n104,MODULARITY,Modularity is an inherent quality of graphs that allows them to be partitioned into communities of closely-related nodes,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n106,GRAPH RAG,KNOWLEDGE GRAPH,Graph RAG uses a knowledge graph for global summarization,92\n351,KNOWLEDGE GRAPH,MODULARITY,Modularity is an inherent quality of knowledge graphs,3\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 8 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 8 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n258,QUESTION,\"The entity \"\"QUESTION\"\" refers to a specific query used in the evaluation process, particularly as a metric to evaluate the generated responses by asking specific questions. This approach is commonly employed in the domain of Natural Language Processing and Information Retrieval to assess the quality and relevance of responses generated by various models or systems.\",2\n259,PUBLIC FIGURES,\"Public figures are individuals who have gained fame or notoriety in various sectors such as entertainment, sports, and digital media. These individuals are well-known in the entertainment industry and are frequently mentioned across various articles. Their prominence in public discourse spans multiple domains, reflecting their influence and recognition in society.\",5\n303,CONTROVERSIES,Controversies are events or issues involving public figures that generate public debate and impact public discourse.,2\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n90,LLM,QUESTION,LLM-generated responses are evaluated using specific questions,34\n565,PUBLIC FIGURES,ANSWER 1,Answer 1 covers a wide range of public figures from different sectors of the entertainment industry.,16\n566,PUBLIC FIGURES,ANSWER 2,\"Answer 2 focuses on a smaller group of public figures, primarily from the music industry and sports.\",15\n612,ANSWER 1,CONTROVERSIES,Answer 1 provides insights into controversies involving public figures and their impact on public discourse.,13\n564,PUBLIC FIGURES,ENTERTAINMENT ARTICLES,Public figures are repeatedly mentioned across various entertainment articles,11\n567,PUBLIC FIGURES,CONTROVERSIES,Controversies involve public figures and impact public discourse.,7\n563,QUESTION,PUBLIC FIGURES,The question asks about public figures mentioned in entertainment articles,7\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 8 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 8 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n34,TEXT CHUNKS,\"TEXT CHUNKS are segments of text that are embedded into a vector space for analysis. These segments are extracted from source documents and are used for processing in the Graph RAG (Retrieval-Augmented Generation) approach. By embedding these text chunks into a vector space, they can be analyzed more effectively, facilitating various natural language processing and information retrieval tasks.\",5\n129,CHUNK SIZE,Chunk size refers to the length of text chunks used in the extraction process,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n68,LLM,TEXT CHUNKS,Text chunks are processed using LLM to extract elements of a graph index,37\n218,SOURCE DOCUMENTS,TEXT CHUNKS,\"In the domain of Natural Language Processing and Information Retrieval, \"\"SOURCE DOCUMENTS\"\" and \"\"TEXT CHUNKS\"\" are closely related entities. Text chunks are extracted from source documents, specifically for processing in the Graph RAG (Retrieval-Augmented Generation) approach. This method involves breaking down the source documents into smaller, manageable pieces, or text chunks, which can then be effectively utilized in various computational processes, enhancing the efficiency and accuracy of information retrieval and generation tasks.\",10\n254,TEXT CHUNKS,ELEMENT INSTANCES,Element instances are extracted from text chunks,9\n255,TEXT CHUNKS,ENTITY REFERENCES,Entity references are extracted from text chunks during processing,8\n256,TEXT CHUNKS,CHUNK SIZE,Chunk size refers to the length of text chunks used in the extraction process,6\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 7 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 7 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n115,LOW-LEVEL COMMUNITY SUMMARIES,LOW-LEVEL COMMUNITY SUMMARIES are a type of community summary used in the News dataset for analysis. These summaries provide a detailed overview of the source text and are generated from lower hierarchical levels of the community in the knowledge graph.,3\n23,ENTITY KNOWLEDGE GRAPH,An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents,2\n116,INTERMEDIATE-LEVEL COMMUNITY SUMMARIES,\"INTERMEDIATE-LEVEL COMMUNITY SUMMARIES are summaries that provide a mid-level overview of the source text. These summaries are generated from intermediate hierarchical levels of the community in the knowledge graph, offering a balanced perspective that captures essential details without overwhelming the reader with excessive information. This approach ensures that the summaries are both informative and concise, making them valuable for understanding the structural dynamics and relationships within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval.\",2\n24,SOURCE DOCUMENTS,\"Source documents are the original texts from which information is extracted, retrieved, or summarized. These documents serve as the foundational input texts for various processing tasks, including the Graph RAG (Retrieval-Augmented Generation) approach. In this context, source documents are critical for extracting and analyzing information, ensuring that the data used in computational linguistics and information retrieval tasks is accurate and comprehensive.\",5\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n126,GRAPH RAG,LOW-LEVEL COMMUNITY SUMMARIES,Low-level community summaries are generated in the Graph RAG approach,93\n103,GRAPH RAG,ENTITY KNOWLEDGE GRAPH,Graph RAG uses an entity knowledge graph to index text,92\n127,GRAPH RAG,INTERMEDIATE-LEVEL COMMUNITY SUMMARIES,Intermediate-level community summaries are generated in the Graph RAG approach,92\n407,LOW-LEVEL COMMUNITY SUMMARIES,NEWS DATASET,Low-level community summaries are derived from the News dataset for analysis,18\n218,SOURCE DOCUMENTS,TEXT CHUNKS,\"In the domain of Natural Language Processing and Information Retrieval, \"\"SOURCE DOCUMENTS\"\" and \"\"TEXT CHUNKS\"\" are closely related entities. Text chunks are extracted from source documents, specifically for processing in the Graph RAG (Retrieval-Augmented Generation) approach. This method involves breaking down the source documents into smaller, manageable pieces, or text chunks, which can then be effectively utilized in various computational processes, enhancing the efficiency and accuracy of information retrieval and generation tasks.\",10\n220,SOURCE DOCUMENTS,LOW-LEVEL COMMUNITY SUMMARIES,Low-level community summaries are derived from source documents,8\n217,ENTITY KNOWLEDGE GRAPH,SOURCE DOCUMENTS,An entity knowledge graph is derived from source documents,7\n219,SOURCE DOCUMENTS,INTERMEDIATE-LEVEL COMMUNITY SUMMARIES,Intermediate-level community summaries are derived from source documents,7\n221,SOURCE DOCUMENTS,DOCUMENT CORPUS,Document corpus consists of source documents being processed,7\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 4 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 4 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n120,ENTITY REFERENCES,\"Entity references are mentions of entities within text chunks, extracted during the processing\",3\n121,RECALL,Recall is a metric used to measure the completeness of entity extraction from text chunks,1\n122,PRECISION,Precision is a metric used to measure the accuracy of entity extraction from text chunks,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n255,TEXT CHUNKS,ENTITY REFERENCES,Entity references are extracted from text chunks during processing,8\n422,ENTITY REFERENCES,RECALL,Recall measures the completeness of entity references extracted from text chunks,4\n423,ENTITY REFERENCES,PRECISION,Precision measures the accuracy of entity references extracted from text chunks,4\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 4 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 4 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n133,DOCUMENT CORPUS,Document corpus refers to the collection of documents being processed in the Graph RAG approach,2\n132,DOMAIN,Domain refers to the specific area of knowledge or field to which the document corpus belongs,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n221,SOURCE DOCUMENTS,DOCUMENT CORPUS,Document corpus consists of source documents being processed,7\n430,DOMAIN,DOCUMENT CORPUS,Domain refers to the specific area of knowledge of the document corpus,3\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 4 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 4 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n15,COMMUNITY SUMMARIES,\"Community summaries are generated summaries of data clusters or communities, used to answer queries and tailored to the domain. These summaries are pre-generated for groups of closely-related entities, particularly in the Graph RAG approach, and are derived from community-generated content to compare with source texts. They are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks. Community summaries provide report-like insights into each community within a hierarchical structure, which is useful for understanding the dataset. They are generated from modular communities in the knowledge graph and cover different levels of each graph community hierarchy, including root-level communities in an entity-based graph index. Additionally, these summaries act as a kind of self-memory for generation-augmented retrieval.\",26\n38,COMMUNITY ANSWERS,\"COMMUNITY ANSWERS are query-focused summaries derived from community summaries. These answers are generated in parallel from the community summaries and are designed to respond to user queries effectively. Essentially, COMMUNITY ANSWERS serve as responses that synthesize information from community summaries to provide concise and relevant answers to specific questions posed by users.\",3\n187,CHUNK,Chunks are segments of community summaries divided into pre-specified token sizes,1\n105,COMMUNITY DESCRIPTIONS,Descriptions generated from modular communities in the knowledge graph,1\n184,SUMMARY DETAIL,Summary detail refers to the level of detail provided in a summary,1\n185,SCOPE,Scope refers to the range or extent of information covered in a summary,1\n192,SUMMARY DETAIL AND SCOPE,Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking,1\n421,ROOT-LEVEL COMMUNITIES,The top-level communities in a hierarchical structure,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n99,GRAPH RAG,COMMUNITY SUMMARIES,\"Graph RAG is an approach that utilizes community summaries in various capacities to enhance its functionality. Specifically, community summaries are generated within the Graph RAG framework and are employed to generate partial responses. These summaries are also used to compare against source texts, serving as a form of self-memory for the system. By leveraging community summaries, Graph RAG aims to improve the comprehensiveness and diversity of answers. Additionally, Graph RAG incorporates summaries of root-level communities within an entity-based graph index, further enriching its analytical capabilities.\",116\n173,GRAPH RAG,COMMUNITY ANSWERS,Graph RAG generates community answers in parallel,93\n208,COMMUNITY SUMMARIES,PODCAST DATASET,Community summaries are derived from the Podcast dataset for analysis,41\n209,COMMUNITY SUMMARIES,NEWS DATASET,Community summaries are derived from the News dataset for analysis,41\n192,COMMUNITY SUMMARIES,HIERARCHICAL COMMUNITY STRUCTURE,Community summaries are created for each level in the hierarchical community structure,33\n198,COMMUNITY SUMMARIES,GLOBAL ANSWER,Global answers are generated from community summaries,33\n201,COMMUNITY SUMMARIES,SENSEMAKING,Community summaries are used for sensemaking,33\n193,COMMUNITY SUMMARIES,GLOBAL SUMMARIZATION,Community summaries are useful for understanding the global structure and semantics of the dataset,31\n196,COMMUNITY SUMMARIES,SUB-COMMUNITIES,Community summaries are generated from sub-communities,31\n195,COMMUNITY SUMMARIES,ROOT COMMUNITIES,Community summaries are generated from root communities,30\n207,COMMUNITY SUMMARIES,GRAPH,Community summaries are part of the graph community hierarchy,30\n188,COMMUNITY SUMMARIES,COMMUNITY ANSWERS,\"Community answers are created from community summaries. These answers are generated by synthesizing information from the summaries provided by the community, ensuring that the responses are comprehensive and reflective of the collective knowledge and insights within the community.\",29\n187,COMMUNITY SUMMARIES,GRAPH COMMUNITIES,Community summaries are created from graph communities,29\n204,COMMUNITY SUMMARIES,CHUNKS,Community summaries are divided into chunks,29\n206,COMMUNITY SUMMARIES,INTERMEDIATE ANSWERS,Intermediate answers are generated from community summaries,29\n186,COMMUNITY SUMMARIES,PARTIAL RESPONSE,Community summaries are used to generate partial responses,28\n189,COMMUNITY SUMMARIES,DOMAIN-TAILORED SUMMARIZATION,Domain-tailored summarization is used to create community summaries,28\n191,COMMUNITY SUMMARIES,PARTIAL ANSWERS,Partial answers are generated from community summaries,28\n194,COMMUNITY SUMMARIES,GLOBAL QUERIES,Community summaries are used to answer global queries,28\n197,COMMUNITY SUMMARIES,LLM CONTEXT WINDOW,Community summaries are added to the LLM context window until the token limit is reached,28\n205,COMMUNITY SUMMARIES,USER QUERY,Community summaries are prepared to answer user queries,28\n202,COMMUNITY SUMMARIES,CHUNK,Community summaries are divided into chunks of pre-specified token size,27\n190,COMMUNITY SUMMARIES,COMMUNITY DESCRIPTIONS,Community descriptions are generated from community summaries,27\n199,COMMUNITY SUMMARIES,SUMMARY DETAIL,The level of summary detail affects the content of community summaries,27\n200,COMMUNITY SUMMARIES,SCOPE,The scope of information affects the content of community summaries,27\n203,COMMUNITY SUMMARIES,SUMMARY DETAIL AND SCOPE,Summary detail and scope affect the content of community summaries,27\n210,COMMUNITY SUMMARIES,ROOT-LEVEL COMMUNITIES,Summaries of root-level communities are used in Graph RAG,27\n269,COMMUNITY ANSWERS,GLOBAL ANSWER,Global answer is created from community answers,10\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 1 second. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 1 second. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n216,MT-BENCH,\"MT-BENCH is a benchmark dataset designed for open-domain question answering, specifically targeting explicit fact retrieval. It serves as a critical tool in evaluating the performance of models in retrieving accurate and relevant information from a broad range of topics. MT-Bench is prominently featured in the academic paper titled \"\"Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena,\"\" where it is utilized to assess the capabilities of large language models in a structured and rigorous manner. This dataset plays a significant role in advancing the field of Natural Language Processing and Information Retrieval by providing a standardized metric for model comparison and evaluation.\",12\n223,ZHENG ET AL.,Authors associated with the MT-Bench dataset,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n495,MT-BENCH,\"LI, Z.\",\"Li, Z. is an author of the paper that discusses MT-Bench\",32\n491,MT-BENCH,\"ZHUANG, S.\",\"Zhuang, S. is an author of the paper that discusses MT-Bench\",27\n488,MT-BENCH,\"ZHENG, L.\",\"Zheng, L. is an author of the paper that discusses MT-Bench\",24\n489,MT-BENCH,\"CHIANG, W.-L.\",\"Chiang, W.-L. is an author of the paper that discusses MT-Bench\",24\n490,MT-BENCH,\"SHENG, Y.\",\"Sheng, Y. is an author of the paper that discusses MT-Bench\",24\n492,MT-BENCH,\"WU, Z.\",\"Wu, Z. is an author of the paper that discusses MT-Bench\",24\n493,MT-BENCH,\"ZHUANG, Y.\",\"Zhuang, Y. is an author of the paper that discusses MT-Bench\",24\n494,MT-BENCH,\"LIN, Z.\",\"Lin, Z. is an author of the paper that discusses MT-Bench\",24\n496,MT-BENCH,\"LI, D.\",\"Li, D. is an author of the paper that discusses MT-Bench\",24\n497,MT-BENCH,\"XING, E.\",\"Xing, E. is an author of the paper that discusses MT-Bench\",24\n498,MT-BENCH,CHATBOT ARENA,\"MT-Bench and Chatbot Arena are both tools used in the paper \"\"Judging llm-as-a-judge with mt-bench and chatbot arena\"\"\",23\n487,MT-BENCH,ZHENG ET AL.,Zheng et al. are the authors associated with the MT-Bench dataset,13\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 1 second. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 1 second. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n280,TRAVIS KELCE,\"Travis Kelce is a public figure frequently mentioned in entertainment articles, known for his contributions to the sports industry and his high-profile personal life.\",6\n293,SPORTS,\"The entity \"\"SPORTS\"\" refers to a sector within the entertainment industry that encompasses athletic events and competitions. This sector includes public figures involved in sports, such as athletes, coaches, and sports commentators.\",4\n\n\n-----Claims-----\nhuman_readable_id,subject_id,type,status,description\n34,TRAVIS KELCE,PUBLIC INTEREST,TRUE,Travis Kelce is frequently mentioned in entertainment articles due to his high-profile status and the publicā€™s interest in his career and personal life.\n39,TRAVIS KELCE,FREQUENT MENTIONS,TRUE,Travis Kelce is repeatedly mentioned across various entertainment articles.\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n142,GRAPH RAG,TRAVIS KELCE,Graph RAG mentions Travis Kelce as a prominent public figure,96\n574,ENTERTAINMENT INDUSTRY,TRAVIS KELCE,Travis Kelce is a significant figure in the entertainment industry,25\n585,ENTERTAINMENT INDUSTRY,SPORTS,Sports is a sector within the entertainment industry,23\n394,NAIVE RAG,TRAVIS KELCE,Naive RAG mentions Travis Kelce as a public figure,19\n598,TRAVIS KELCE,ANSWER 2,Travis Kelce is one of the specific public figures mentioned in Answer 2.,16\n608,SPORTS,ANSWER 1,Answer 1 includes public figures from the sports sector.,15\n609,SPORTS,ANSWER 2,Answer 2 focuses on public figures primarily from the sports sector.,14\n570,ENTERTAINMENT ARTICLES,TRAVIS KELCE,Travis Kelce is frequently mentioned in entertainment articles,12\n599,TRAVIS KELCE,SPORTS,Travis Kelce is a public figure in the sports sector.,10\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 1 second. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 1 second. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n108,PARTIAL ANSWERS,Intermediate answers generated from community summaries before being combined into a final global answer,2\n109,FINAL GLOBAL ANSWER,The comprehensive answer generated by combining all relevant partial answers,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n191,COMMUNITY SUMMARIES,PARTIAL ANSWERS,Partial answers are generated from community summaries,28\n405,PARTIAL ANSWERS,FINAL GLOBAL ANSWER,Final global answer is generated by combining all relevant partial answers,3\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 20 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 20 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n194,CHUNKS,Chunks are segments of community summaries divided based on a pre-specified token size,3\n209,TOKEN SIZE,The pre-specified size of tokens used to divide community summaries into chunks,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n84,LLM,CHUNKS,LLM generates intermediate answers and scores for each chunk,35\n204,COMMUNITY SUMMARIES,CHUNKS,Community summaries are divided into chunks,29\n470,CHUNKS,TOKEN SIZE,Chunks are divided based on a pre-specified token size,4\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 17 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 17 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n195,HELPFULNESS SCORE,A score between 0-100 generated by the LLM to indicate how helpful an answer is in addressing the target question,3\n39,GLOBAL ANSWER,GLOBAL ANSWER is a comprehensive response generated from multiple community summaries to answer a user query. It is the final query-focused summary produced from all relevant community summaries. The final answer is generated by combining intermediate community answers based on their helpfulness scores.,7\n208,INTERMEDIATE ANSWERS,Answers generated for each chunk of community summaries,3\n186,USER QUERY,\"A \"\"USER QUERY\"\" is a question or inquiry posed by a user seeking information, which the system aims to answer.\",2\n210,CONTEXT WINDOW,\"The \"\"CONTEXT WINDOW\"\" refers to a window of text used to generate answers, constrained by token size. The size of the context window is consistent across all conditions, ensuring uniformity in answer generation processes.\",2\n239,PROMPTS,\"The prompts used for answer generation, which are the same across all conditions with minor modifications\",1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n85,LLM,HELPFULNESS SCORE,LLM generates a helpfulness score for each answer,35\n198,COMMUNITY SUMMARIES,GLOBAL ANSWER,Global answers are generated from community summaries,33\n206,COMMUNITY SUMMARIES,INTERMEDIATE ANSWERS,Intermediate answers are generated from community summaries,29\n205,COMMUNITY SUMMARIES,USER QUERY,Community summaries are prepared to answer user queries,28\n214,QUERY-FOCUSED SUMMARIZATION,GLOBAL ANSWER,Query-focused summarization is used to produce the global answer,12\n271,GLOBAL ANSWER,HELPFULNESS SCORE,Global answer is generated by sorting intermediate answers based on helpfulness scores,10\n272,GLOBAL ANSWER,INTERMEDIATE ANSWERS,Intermediate answers are combined to form the global answer,10\n269,COMMUNITY ANSWERS,GLOBAL ANSWER,Global answer is created from community answers,10\n273,GLOBAL ANSWER,CONTEXT WINDOW,The final context window is used to generate the global answer,9\n270,GLOBAL ANSWER,USER QUERY,Global answers are generated in response to user queries,9\n471,HELPFULNESS SCORE,INTERMEDIATE ANSWERS,Helpfulness scores are assigned to intermediate answers,6\n482,CONTEXT WINDOW,PROMPTS,The size of the context window and the prompts used for answer generation are the same across all conditions,3\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 15 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 15 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n339,MULTI-DOCUMENT SUMMARIZATION,A method that combines multiple concepts for summarizing multiple documents,2\n360,MULTI-DOCUMENT SUMMARIZATION (CAIRE-COVID),A system that combines multiple concepts for multi-document summarization,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n157,GRAPH RAG,MULTI-DOCUMENT SUMMARIZATION,Graph RAG incorporates concepts used in multi-document summarization,92\n631,MULTI-DOCUMENT SUMMARIZATION,MULTI-DOCUMENT SUMMARIZATION (CAIRE-COVID),CAiRE-COVID is a system for multi-document summarization,3\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 12 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 12 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n88,COMMUNITY DETECTION ALGORITHMS,\"COMMUNITY DETECTION ALGORITHMS are algorithms used to partition a graph into communities of nodes with stronger connections to one another. These algorithms are designed to identify modular communities of closely-related nodes within a graph, thereby revealing the underlying structure and relationships within the network.\",6\n46,LEIDEN,\"LEIDEN is a community detection algorithm renowned for its efficiency in recovering hierarchical community structures. It is widely used to partition graphs into modular communities, effectively grouping elements within a graph index. The algorithm's ability to identify and organize these communities makes it a valuable tool in the analysis of complex networks, particularly within the domains of Natural Language Processing and Information Retrieval.\",7\n174,FIGURE 3,A visual representation of graph communities detected using the Leiden algorithm,1\n53,TRAAG,Traag is an author who contributed to the development of the Leiden community detection method,1\n66,TRAAG ET AL.,\"Traag et al. are the authors who developed the Leiden algorithm, a method renowned for its efficiency in recovering hierarchical community structures. This algorithm is widely recognized in the field of Natural Language Processing and Information Retrieval for its ability to accurately detect and map out complex community dynamics.\",1\n89,LOUVAIN,Louvain is a community detection algorithm used to partition graphs into modular communities,1\n163,FORTUNATO,An author who has conducted surveys on community detection algorithms,1\n164,JIN ET AL.,Authors who have conducted surveys on community detection algorithms,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n107,GRAPH RAG,COMMUNITY DETECTION ALGORITHMS,Community detection algorithms are used in the Graph RAG approach to partition graphs,96\n356,COMMUNITY DETECTION ALGORITHMS,GRAPH INDEX,Community detection algorithms are used to partition the graph index into communities,24\n289,LEIDEN,MULTIHOP-RAG,The Leiden algorithm is used to detect graph communities in the MultiHop-RAG,15\n288,LEIDEN,HIERARCHICAL COMMUNITY STRUCTURE,Leiden is known for its ability to recover hierarchical community structures efficiently,14\n279,GRAPH RAG PIPELINE,LEIDEN,Leiden method is used in the graph RAG pipeline for community detection,14\n287,LEIDEN,COMMUNITY DETECTION ALGORITHMS,\"Leiden is a specific type of community detection algorithm used in various analytical pipelines. It is designed to identify and map out the structural dynamics within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval. The algorithm is known for its efficiency and accuracy in detecting community structures, making it a valuable tool for researchers and practitioners in the field.\",13\n290,LEIDEN,FIGURE 3,Figure 3 shows graph communities detected using the Leiden algorithm,8\n285,LEIDEN,TRAAG,Traag contributed to the development of the Leiden method,8\n286,LEIDEN,TRAAG ET AL.,Traag et al. are the authors of the Leiden algorithm and developed the Leiden method.,8\n355,COMMUNITY DETECTION ALGORITHMS,LOUVAIN,Louvain is a type of community detection algorithm,7\n357,COMMUNITY DETECTION ALGORITHMS,FORTUNATO,Fortunato has conducted surveys on community detection algorithms,7\n358,COMMUNITY DETECTION ALGORITHMS,JIN ET AL.,Jin et al. have conducted surveys on community detection algorithms,7\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 9 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 9 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n15,COMMUNITY SUMMARIES,\"Community summaries are generated summaries of data clusters or communities, used to answer queries and tailored to the domain. These summaries are pre-generated for groups of closely-related entities, particularly in the Graph RAG approach, and are derived from community-generated content to compare with source texts. They are randomly shuffled and divided into chunks of pre-specified token size to ensure relevant information is distributed across chunks. Community summaries provide report-like insights into each community within a hierarchical structure, which is useful for understanding the dataset. They are generated from modular communities in the knowledge graph and cover different levels of each graph community hierarchy, including root-level communities in an entity-based graph index. Additionally, these summaries act as a kind of self-memory for generation-augmented retrieval.\",26\n38,COMMUNITY ANSWERS,\"COMMUNITY ANSWERS are query-focused summaries derived from community summaries. These answers are generated in parallel from the community summaries and are designed to respond to user queries effectively. Essentially, COMMUNITY ANSWERS serve as responses that synthesize information from community summaries to provide concise and relevant answers to specific questions posed by users.\",3\n25,PARTIAL RESPONSE,A partial response is an intermediate answer generated from community summaries before being combined into a final response,2\n108,PARTIAL ANSWERS,Intermediate answers generated from community summaries before being combined into a final global answer,2\n187,CHUNK,Chunks are segments of community summaries divided into pre-specified token sizes,1\n105,COMMUNITY DESCRIPTIONS,Descriptions generated from modular communities in the knowledge graph,1\n184,SUMMARY DETAIL,Summary detail refers to the level of detail provided in a summary,1\n185,SCOPE,Scope refers to the range or extent of information covered in a summary,1\n192,SUMMARY DETAIL AND SCOPE,Summary detail and scope refer to the balance of detail and range of information in community summaries for sensemaking,1\n421,ROOT-LEVEL COMMUNITIES,The top-level communities in a hierarchical structure,1\n109,FINAL GLOBAL ANSWER,The comprehensive answer generated by combining all relevant partial answers,1\n26,FINAL RESPONSE,A final response is the comprehensive answer generated after combining all partial responses,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n99,GRAPH RAG,COMMUNITY SUMMARIES,\"Graph RAG is an approach that utilizes community summaries in various capacities to enhance its functionality. Specifically, community summaries are generated within the Graph RAG framework and are employed to generate partial responses. These summaries are also used to compare against source texts, serving as a form of self-memory for the system. By leveraging community summaries, Graph RAG aims to improve the comprehensiveness and diversity of answers. Additionally, Graph RAG incorporates summaries of root-level communities within an entity-based graph index, further enriching its analytical capabilities.\",116\n173,GRAPH RAG,COMMUNITY ANSWERS,Graph RAG generates community answers in parallel,93\n208,COMMUNITY SUMMARIES,PODCAST DATASET,Community summaries are derived from the Podcast dataset for analysis,41\n209,COMMUNITY SUMMARIES,NEWS DATASET,Community summaries are derived from the News dataset for analysis,41\n192,COMMUNITY SUMMARIES,HIERARCHICAL COMMUNITY STRUCTURE,Community summaries are created for each level in the hierarchical community structure,33\n198,COMMUNITY SUMMARIES,GLOBAL ANSWER,Global answers are generated from community summaries,33\n201,COMMUNITY SUMMARIES,SENSEMAKING,Community summaries are used for sensemaking,33\n193,COMMUNITY SUMMARIES,GLOBAL SUMMARIZATION,Community summaries are useful for understanding the global structure and semantics of the dataset,31\n196,COMMUNITY SUMMARIES,SUB-COMMUNITIES,Community summaries are generated from sub-communities,31\n195,COMMUNITY SUMMARIES,ROOT COMMUNITIES,Community summaries are generated from root communities,30\n207,COMMUNITY SUMMARIES,GRAPH,Community summaries are part of the graph community hierarchy,30\n188,COMMUNITY SUMMARIES,COMMUNITY ANSWERS,\"Community answers are created from community summaries. These answers are generated by synthesizing information from the summaries provided by the community, ensuring that the responses are comprehensive and reflective of the collective knowledge and insights within the community.\",29\n187,COMMUNITY SUMMARIES,GRAPH COMMUNITIES,Community summaries are created from graph communities,29\n204,COMMUNITY SUMMARIES,CHUNKS,Community summaries are divided into chunks,29\n206,COMMUNITY SUMMARIES,INTERMEDIATE ANSWERS,Intermediate answers are generated from community summaries,29\n186,COMMUNITY SUMMARIES,PARTIAL RESPONSE,Community summaries are used to generate partial responses,28\n189,COMMUNITY SUMMARIES,DOMAIN-TAILORED SUMMARIZATION,Domain-tailored summarization is used to create community summaries,28\n191,COMMUNITY SUMMARIES,PARTIAL ANSWERS,Partial answers are generated from community summaries,28\n194,COMMUNITY SUMMARIES,GLOBAL QUERIES,Community summaries are used to answer global queries,28\n197,COMMUNITY SUMMARIES,LLM CONTEXT WINDOW,Community summaries are added to the LLM context window until the token limit is reached,28\n205,COMMUNITY SUMMARIES,USER QUERY,Community summaries are prepared to answer user queries,28\n202,COMMUNITY SUMMARIES,CHUNK,Community summaries are divided into chunks of pre-specified token size,27\n190,COMMUNITY SUMMARIES,COMMUNITY DESCRIPTIONS,Community descriptions are generated from community summaries,27\n199,COMMUNITY SUMMARIES,SUMMARY DETAIL,The level of summary detail affects the content of community summaries,27\n200,COMMUNITY SUMMARIES,SCOPE,The scope of information affects the content of community summaries,27\n203,COMMUNITY SUMMARIES,SUMMARY DETAIL AND SCOPE,Summary detail and scope affect the content of community summaries,27\n210,COMMUNITY SUMMARIES,ROOT-LEVEL COMMUNITIES,Summaries of root-level communities are used in Graph RAG,27\n269,COMMUNITY ANSWERS,GLOBAL ANSWER,Global answer is created from community answers,10\n405,PARTIAL ANSWERS,FINAL GLOBAL ANSWER,Final global answer is generated by combining all relevant partial answers,3\n222,PARTIAL RESPONSE,FINAL RESPONSE,Partial responses are summarized to generate a final response,3\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 7 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 7 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n404,HYBRID RAG SCHEMES,RAG schemes that combine embedding-based matching with other approaches,2\n417,COMMUNITY REPORTS,Reports generated from community summaries,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n179,GRAPH RAG,HYBRID RAG SCHEMES,Graph RAG can be part of hybrid RAG schemes,92\n654,HYBRID RAG SCHEMES,COMMUNITY REPORTS,Hybrid RAG schemes combine embedding-based matching against community reports,3\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 5 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 5 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n181,LLM CONTEXT WINDOW,The LLM context window is the token limit within which summaries are added for processing by a language model,2\n183,TOKEN LIMIT,The token limit is the maximum number of tokens that can be processed in a single context window by a language model,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n197,COMMUNITY SUMMARIES,LLM CONTEXT WINDOW,Community summaries are added to the LLM context window until the token limit is reached,28\n467,LLM CONTEXT WINDOW,TOKEN LIMIT,The token limit defines the maximum number of tokens in the LLM context window,3\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 3 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 3 seconds. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n510,\"ES, S.\",\"Es, S. is an author of the paper \"\"Ragas: Automated evaluation of retrieval augmented generation\"\"\",2\n512,\"ESPINOSA-ANKE, L.\",\"Espinosa-Anke, L. is an author of the paper \"\"Ragas: Automated evaluation of retrieval augmented generation\"\"\",2\n511,\"JAMES, J.\",\"James, J. is an author of the paper \"\"Ragas: Automated evaluation of retrieval augmented generation\"\"\",2\n513,\"SCHOCKAERT, S.\",\"Schockaert, S. is an author of the paper \"\"Ragas: Automated evaluation of retrieval augmented generation\"\"\",2\n530,ARXIV:2309.15217,\"The arXiv identifier for the paper \"\"Ragas: Automated evaluation of retrieval augmented generation\"\"\",4\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n300,ARXIV,\"ES, S.\",\"Es, S. published the paper \"\"Ragas: Automated evaluation of retrieval augmented generation\"\" on arXiv\",41\n302,ARXIV,\"ESPINOSA-ANKE, L.\",\"Espinosa-Anke, L. published the paper \"\"Ragas: Automated evaluation of retrieval augmented generation\"\" on arXiv\",41\n301,ARXIV,\"JAMES, J.\",\"James, J. published the paper \"\"Ragas: Automated evaluation of retrieval augmented generation\"\" on arXiv\",41\n303,ARXIV,\"SCHOCKAERT, S.\",\"Schockaert, S. published the paper \"\"Ragas: Automated evaluation of retrieval augmented generation\"\" on arXiv\",41\n826,\"ES, S.\",ARXIV:2309.15217,\"Es, S. is an author of the paper with arXiv identifier 2309.15217\",6\n827,\"JAMES, J.\",ARXIV:2309.15217,\"James, J. is an author of the paper with arXiv identifier 2309.15217\",6\n828,\"ESPINOSA-ANKE, L.\",ARXIV:2309.15217,\"Espinosa-Anke, L. is an author of the paper with arXiv identifier 2309.15217\",6\n829,\"SCHOCKAERT, S.\",ARXIV:2309.15217,\"Schockaert, S. is an author of the paper with arXiv identifier 2309.15217\",6\n\nOutput:"}} +{"type": "error", "data": "Error Invoking LLM", "stack": "Traceback (most recent call last):\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/base/base_llm.py\", line 53, in _invoke\n output = await self._execute_llm(input, **kwargs)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/graphrag/llm/openai/openai_chat_llm.py\", line 53, in _execute_llm\n completion = await self.client.chat.completions.create(\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/resources/chat/completions.py\", line 1339, in create\n return await self._post(\n ^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1816, in post\n return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1510, in request\n return await self._request(\n ^^^^^^^^^^^^^^^^^^^^\n File \"/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/openai/_base_client.py\", line 1611, in _request\n raise self._make_status_error_from_response(err.response) from None\nopenai.RateLimitError: Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 1 second. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}\n", "source": "Error code: 429 - {'error': {'code': '429', 'message': 'Requests to the ChatCompletions_Create Operation under Azure OpenAI API version 2024-04-01-preview have exceeded token rate limit of your current OpenAI S0 pricing tier. Please retry after 1 second. Please go here: https://aka.ms/oai/quotaincrease if you would like to further increase the default rate limit.'}}", "details": {"input": "\nYou are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the \"Natural Language Processing and Information Retrieval\" domain.\n\n# Goal\nWrite a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the \"Natural Language Processing and Information Retrieval\" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims.\nThe analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact.\n\nDomain: \"Natural Language Processing and Information Retrieval\"\nText: results,\ni.e., that na ĀØıve RAG produces the most direct responses across all comparisons.\n9Podcast Transcripts News Articles\nC0 C1 C2 C3 TS C0 C1 C2 C3 TS\nUnits 34 367 969 1310 1669 55 555 1797 2142 3197\nTokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694\n% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100\nTable 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre-\nsponding token counts, and percentage of the maximum token count. Map-reduce summarization of\nsource texts is the most , 21(5):88ā€“92.\nKoesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data\nsensemaking behaviours. International journal of human-computer studies , 146:102562.\nKuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search\nof needles in a 11m haystack: Recurrent memory finds what llms miss.\nLangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/.\nLaskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via\nincorporating query relevance and transfer learning with transformer models. In Advances in\nArtificial Intelligence: system for covid-19 scholarly\ninformation management. arXiv preprint arXiv:2005.03975 .\nTang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for\nmulti-hop queries. arXiv preprint arXiv:2401.15391 .\nTouvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S.,\nBhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models.\narXiv preprint arXiv:2307.09288 .\nTraag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing\nwell . arXiv preprint arXiv:2306.04136 .\nBan, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing\nlarge language models for advanced causal discovery from data.\nBaumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo-\nrating query relevance, multi-document coverage, and summary length constraints into seq2seq\nmodels. arXiv preprint arXiv:1801.07704 .\nBlondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of\ncommunities in large networks. Journal of statistical mechanics: theory and experiment ,\n2008(10):P10008.\nBrown, T., Mann, B., Ryder, N., Subbia are the same across\nall six conditions (except for minor modifications to reference styles to match the types of context\ninformation used). Conditions only differ in how the contents of the context window are created.\nThe graph index supporting conditions C0-C3was created using our generic prompts for entity and\nrelationship extraction only, with entity types and few-shot examples tailored to the domain of the\ndata. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the\nPodcast dataset and 0 gleanings for the News dataset.\n3.4 Metrics\nLLMs have been shown to be good evaluators of natural language generation, achieving state-of-\nthe-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al.,\n2024). While this approach can generate reference-based metrics when gold standard answers are\nknown, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using\na method designed to scale to very large datasets. These summaries are independently useful in\ntheir own right as a way to understand the global structure and semantics of the dataset, and may\nthemselves be used to make sense of a corpus in the absence of a question. For example, a user\nmay scan through community summaries at one level looking for general themes of interest, then\nfollow links to the reports at the lower level that provide more details for each of the subtopics. Here,\nhowever, we focus on their utility as part of a graph-based index used for answering global queries.\nCommunity summaries are generated in the following way:\n4(a) Root communities at level 0 (b) Sub-communities at level 1\nFigure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the\nMultiHop-RAG (T LLM-generated assessment.\n7Question Which public figures are repeatedly mentioned across various entertainment articles?\nGraph\nRAGOverview of Prominent Public Figures in Entertainment\nThe entertainment industry is vast and diverse, encompassing film, television, music, sports, and\ndigital media. Certain public figures stand out due to their significant contributions and influence\nacross these sectors. The following summary highlights key individuals who are repeatedly\nmentioned in various entertainment articles, reflecting their impact and presence within the industry.\nActors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...]\nAthletes and Coaches [...]Influencers and Entrepreneurs [...]\nThe repeated mention of these figures in entertainment articles signifies their ongoing relevance and\nthe publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from\nshaping cultural narratives in film and television to driving trends in music and digital media. These\nindividuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and\nnot just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent\nyears, however, such distinctions between summarization tasks that are abstractive versus extractive,\ngeneric versus query-focused, and single-document versus multi-document, have become less rele-\nvant. While early applications of the transformer architecture showed substantial improvements on\nthe state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu\nand Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam\net al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse\nanswers, we therefore used a fixed context window size of 8k tokens for the final evaluation.\n3.6 Results\nThe indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast\ndataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the\nnumber of community summaries at different levels of each graph community hierarchy.\nGlobal approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out-\nperformed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across\ndatasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83%\nfor Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82%\nand 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short\ndescriptions of the target datasets. Questions target global understanding rather than specific details.\n3 Evaluation\n3.1 Datasets\nWe selected two datasets in the one million token range, each equivalent to about 10 novels of text\nand representative of the kind of corpora that users may encounter in their real world activities:\nā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott,\nMicrosoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669\nƗ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens).\nā€¢News articles . Benchmark dataset comprising news articles published from September\n2013 to December 2023 in a range of categories, including entertainment, business, sports,\ntechnology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ\n600-token text chunks, with 100 view the role of policy and regulation\nQuestions :\n1. Which episodes deal primarily with tech policy and government regulation?\n2. How do guests perceive the impact of privacy laws on technology development?\n3. Do any guests discuss the balance between innovation and ethical considerations?\n4. What are the suggested changes to current policies mentioned by the guests?\n5. Are collaborations between tech companies and governments discussed and how?\nNews\narticlesUser : Educator incorporating current affairs into curricula\nTask: Teaching about health and wellness\nQuestions :\n1. What current topics in health can be integrated into health education curricula?\n2. How do news articles address the concepts of preventive medicine and wellness?\n3. Are there examples of health articles that contradict each other, and if so, why?\n4. What insights can be gleaned about public health priorities based on news coverage?\n5. How can educators use the dataset to highlight the importance of health literacy?\nTable 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry\nand sports, and relies heavily on a single source for data, which makes it less diverse in perspectives\nand insights.\nEmpowerment: Winner=1 (Graph RAG)\nAnswer 1 is better because it provides a comprehensive and structured overview of public figures\nacross various sectors of the entertainment industry, including film, television, music, sports, and\ndigital media. It lists multiple individuals, providing specific examples of their contributions and the\ncontext in which they are mentioned in entertainment articles, along with references to data reports\nfor each claim. This approach helps the reader understand the breadth of the topic and make informed\njudgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures\nand primarily discusses their personal lives and relationships, which may not provide as broad an\nunderstanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety\nof Answer 1 final round of\nquery-focused summarization over all community summaries reporting relevance to that query.\nā€œa motivated, continuous effort to understand connections (which can be among people, places, and\nevents) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting\nhuman-led sensemaking over entire text corpora, however, needs a way for people to both apply and\nrefine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature.\nRetrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering\nuser questions over entire datasets, but it is designed for situations where these answers are contained\nlocally within regions of text whose retrieval provides sufficient grounding for the generation task.\nInstead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in\nparticular examples provided to the LLM for in-context learning (Brown et al., 2020).\n3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people,\nplaces, and organizations is generally applicable, domains with specialized knowledge (e.g., science,\nmedicine, law) will benefit from few-shot examples specialized to those domains. We also support\na secondary extraction prompt for any additional covariates we would like to associate with the\nextracted node instances. Our default covariate prompt aims to extract claims linked to detected\nentities, including the subject, object, type, description, source text span, and start and end dates.\nTo balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a\nspecified maximum, to encourage the LLM to detect any additional entities it may have missed\non prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess\nwhether all entities resource-intensive approach requiring the highest number of context tokens.\nRoot-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x).\nCommunity summaries vs. source texts. When comparing community summaries to source texts\nusing Graph RAG, community summaries generally provided a small but consistent improvement\nin answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level\nsummaries in the Podcast dataset and low-level community summaries in the News dataset achieved\ncomprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for\nPodcast intermediate-level summaries and 60% for News low-level community summaries. Table 3\nalso illustrates the scalability advantages of Graph RAG compared to source text summarization: for\nlow-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while\nfor root-level community summaries ( C0), it required over 97% fewer tokens.\nRole:. The content of this report includes an overview of the community's key entities and relationships.\n\n# Report Structure\nThe report should include the following sections:\n- TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title.\n- SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities.\n- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval.\n- RATING EXPLANATION: Give a single sentence explanation of the rating.\n- DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive.\n\nReturn output as a well-formed JSON-formatted string with the following format. Don't use any unnecessary escape sequences. The output should be a single JSON object that can be parsed by json.loads.\n {\n \"title\": \"\",\n \"summary\": \"\",\n \"rating\": ,\n \"rating_explanation\": \"\"\n \"findings\": \"[{\"summary\":\"\", \"explanation\": \"\", \"explanation\": \" (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records.\nEach paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use \"NONE\" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing..\n\nExample paragraph with references added:\nThis is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)]\n\n# Example Input\n-----------\nText:\n\nEntities\n\nid,entity,description\n5,ABILA CITY PARK,Abila City Park is the location of the POK rally\n\nRelationships\n\nid,source,target,description\n37,ABILA CITY PARK,POK RALLY,Abila City Park is the location of the POK rally\n38,ABILA CITY PARK,POK,POK is holding a rally in Abila City Park\n39,ABILA CITY PARK,POKRALLY,The POKRally is taking place at Abila City Park\n40,ABILA CITY PARK,CENTRAL BULLETIN,Central Bulletin is reporting on the POK rally taking place in Abila City Park\n\nOutput:\n{\n \"title\": \"Abila City Park and POK Rally\",\n \"summary\": \"The community revolves around the Abila City Park, which is the location of the POK rally. The park has relationships with POK, POKRALLY, and Central Bulletin, all\nof which are associated with the rally event.\",\n \"rating\": 5.0,\n \"rating_explanation\": \"The impact rating is moderate due to the potential for unrest or conflict during the POK rally.\",\n \"findings\": [\n {\n \"summary\": \"Abila City Park as the central location\",\n \"explanation\": \"Abila City Park is the central entity in this community, serving as the location for the POK rally. This park is the common link between all other\nentities, suggesting its significance in the community. The park's association with the rally could potentially lead to issues such as public disorder or conflict, depending on the\nnature of the rally and the reactions it provokes. [records: Entities (5), Relationships (37, 38, 39, 40)]\"\n },\n {\n \"summary\": \"POK's role in the community\",\n \"explanation\": \"POK is another key entity in this community, being the organizer of the rally at Abila City Park. The nature of POK and its rally could be a potential\nsource of threat, depending on their objectives and the reactions they provoke. The relationship between POK and the park is crucial in understanding the dynamics of this community.\n[records: Relationships (38)]\"\n },\n {\n \"summary\": \"POKRALLY as a significant event\",\n \"explanation\": \"The POKRALLY is a significant event taking place at Abila City Park. This event is a key factor in the community's dynamics and could be a potential\nsource of threat, depending on the nature of the rally and the reactions it provokes. The relationship between the rally and the park is crucial in understanding the dynamics of this\ncommunity. [records: Relationships (39)]\"\n },\n {\n \"summary\": \"Role of Central Bulletin\",\n \"explanation\": \"Central Bulletin is reporting on the POK rally taking place in Abila City Park. This suggests that the event has attracted media attention, which could\namplify its impact on the community. The role of Central Bulletin could be significant in shaping public perception of the event and the entities involved. [records: Relationships\n(40)]\"\n }\n ]\n\n}\n\n# Real Data\n\nUse the following text for your answer. Do not make anything up in your answer.\n\nText:\n-----Entities-----\nhuman_readable_id,title,description,degree\n115,LOW-LEVEL COMMUNITY SUMMARIES,LOW-LEVEL COMMUNITY SUMMARIES are a type of community summary used in the News dataset for analysis. These summaries provide a detailed overview of the source text and are generated from lower hierarchical levels of the community in the knowledge graph.,3\n23,ENTITY KNOWLEDGE GRAPH,An entity knowledge graph is a graph-based representation of entities and their relationships derived from source documents,2\n116,INTERMEDIATE-LEVEL COMMUNITY SUMMARIES,\"INTERMEDIATE-LEVEL COMMUNITY SUMMARIES are summaries that provide a mid-level overview of the source text. These summaries are generated from intermediate hierarchical levels of the community in the knowledge graph, offering a balanced perspective that captures essential details without overwhelming the reader with excessive information. This approach ensures that the summaries are both informative and concise, making them valuable for understanding the structural dynamics and relationships within specialized communities, particularly in the domain of Natural Language Processing and Information Retrieval.\",2\n34,TEXT CHUNKS,\"TEXT CHUNKS are segments of text that are embedded into a vector space for analysis. These segments are extracted from source documents and are used for processing in the Graph RAG (Retrieval-Augmented Generation) approach. By embedding these text chunks into a vector space, they can be analyzed more effectively, facilitating various natural language processing and information retrieval tasks.\",5\n24,SOURCE DOCUMENTS,\"Source documents are the original texts from which information is extracted, retrieved, or summarized. These documents serve as the foundational input texts for various processing tasks, including the Graph RAG (Retrieval-Augmented Generation) approach. In this context, source documents are critical for extracting and analyzing information, ensuring that the data used in computational linguistics and information retrieval tasks is accurate and comprehensive.\",5\n120,ENTITY REFERENCES,\"Entity references are mentions of entities within text chunks, extracted during the processing\",3\n133,DOCUMENT CORPUS,Document corpus refers to the collection of documents being processed in the Graph RAG approach,2\n129,CHUNK SIZE,Chunk size refers to the length of text chunks used in the extraction process,1\n121,RECALL,Recall is a metric used to measure the completeness of entity extraction from text chunks,1\n122,PRECISION,Precision is a metric used to measure the accuracy of entity extraction from text chunks,1\n132,DOMAIN,Domain refers to the specific area of knowledge or field to which the document corpus belongs,1\n\n\n-----Relationships-----\nhuman_readable_id,source,target,description,rank\n126,GRAPH RAG,LOW-LEVEL COMMUNITY SUMMARIES,Low-level community summaries are generated in the Graph RAG approach,93\n103,GRAPH RAG,ENTITY KNOWLEDGE GRAPH,Graph RAG uses an entity knowledge graph to index text,92\n127,GRAPH RAG,INTERMEDIATE-LEVEL COMMUNITY SUMMARIES,Intermediate-level community summaries are generated in the Graph RAG approach,92\n68,LLM,TEXT CHUNKS,Text chunks are processed using LLM to extract elements of a graph index,37\n407,LOW-LEVEL COMMUNITY SUMMARIES,NEWS DATASET,Low-level community summaries are derived from the News dataset for analysis,18\n218,SOURCE DOCUMENTS,TEXT CHUNKS,\"In the domain of Natural Language Processing and Information Retrieval, \"\"SOURCE DOCUMENTS\"\" and \"\"TEXT CHUNKS\"\" are closely related entities. Text chunks are extracted from source documents, specifically for processing in the Graph RAG (Retrieval-Augmented Generation) approach. This method involves breaking down the source documents into smaller, manageable pieces, or text chunks, which can then be effectively utilized in various computational processes, enhancing the efficiency and accuracy of information retrieval and generation tasks.\",10\n254,TEXT CHUNKS,ELEMENT INSTANCES,Element instances are extracted from text chunks,9\n255,TEXT CHUNKS,ENTITY REFERENCES,Entity references are extracted from text chunks during processing,8\n220,SOURCE DOCUMENTS,LOW-LEVEL COMMUNITY SUMMARIES,Low-level community summaries are derived from source documents,8\n221,SOURCE DOCUMENTS,DOCUMENT CORPUS,Document corpus consists of source documents being processed,7\n217,ENTITY KNOWLEDGE GRAPH,SOURCE DOCUMENTS,An entity knowledge graph is derived from source documents,7\n219,SOURCE DOCUMENTS,INTERMEDIATE-LEVEL COMMUNITY SUMMARIES,Intermediate-level community summaries are derived from source documents,7\n256,TEXT CHUNKS,CHUNK SIZE,Chunk size refers to the length of text chunks used in the extraction process,6\n422,ENTITY REFERENCES,RECALL,Recall measures the completeness of entity references extracted from text chunks,4\n423,ENTITY REFERENCES,PRECISION,Precision measures the accuracy of entity references extracted from text chunks,4\n430,DOMAIN,DOCUMENT CORPUS,Domain refers to the specific area of knowledge of the document corpus,3\n\nOutput:"}} diff --git a/graphfleet/prompts/community_report.txt b/graphfleet/prompts/community_report.txt index 83a3b35c5..9ed7da025 100644 --- a/graphfleet/prompts/community_report.txt +++ b/graphfleet/prompts/community_report.txt @@ -1,81 +1,174 @@ -<<<<<<< HEAD -You are an expert in Artificial Intelligence and Productivity Research. You are skilled at analyzing the impact and application of generative AI in professional workflows, including software development and multilingual contexts. You are adept at helping people understand the evaluation metrics for AI performance and the integration of AI tools like GitHub Copilot in real-world settings. +You are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the "Natural Language Processing and Information Retrieval" domain. # Goal -Write a comprehensive assessment report of a community taking on the role of a A community analyst that is evaluating the impact and application of generative AI in professional workflows, including software development, multilingual contexts, and productivity studies. The analysis will include the evaluation metrics for AI performance and the integration of AI tools like GitHub Copilot in real-world settings. The report will be used to inform decision-makers about significant developments associated with the community and their potential impact. - -Domain: **Artificial Intelligence and Productivity Research** -Text: The tasks studied in the lab thus far have tended to be those for which researchers hypothesized generative AI would perform well. This was, in fact, the focus of most of the studies presented in the first AI and Productivity report we published (Cambon et al. 2023). Actual information work, however, often includes a huge variety of tasks and much of the unstructured and informal work in peopleā€™s jobs is not yet directly supported by the first-generation of generative AI tools. Software developer workflows, for example, involve far more than the hands-on coding supported by GitHub Copilot (Meyer et al. 2017). The ability to shed light on generative AI's productivity dynamics in the natural complexity of entire workflows is a key advantage of field studies of generative AIā€™s productivity impacts, and a major reason we hope to see many more field studies emerging in the literature Liu, N. F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F., and Liang, P. (2023). Lost in the middle: How language models use long contexts. arXiv:2307.03172. Liu, Y. and Lapata, M. (2019). Hierarchical transformers for multi-document summarization. arXiv preprint arXiv:1905.13164. LlamaIndex (2024). LlamaIndex Knowledge Graph Index. https://docs.llamaindex.ai/en/stable/examples/index_structs/knowledge_graph/KnowledgeGraphDemo.html. Manakul, P., Liusie, A., and Gales, M. J. (2023). Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models. arXiv preprint ar Generative AI in Real-World Workplaces The Second Microsoft Report on AI and Productivity Research - -Editors: -Sonia Jaffe, Neha Parikh Shah, Jenna Butler, Alex Farach, Alexia Cambon, Brent Hecht, Michael Schwarz, and Jaime Teevan -Contributing Researchers: -Reid Andersen, Margarita Bermejo-Cano, James Bono, Georg Buscher, Chacha Chen, Steven Clarke, Scott Counts, Eleanor Dillon, Ben Edelman, Ulrike Gruber-Gremlich, Cory Hilke, Ben Hanrahan, Sandra Ho, Brian Houck, Mansi Khemka, Viktor Kewenig, Madeline Kleiner, Eric Knudsen, Sathish Manivannan, Max Meijer, Jennifer Neville, Nam Ngo, Donald Ngwe, Ried Peckham, Sida Peng, Nora Presson, Nagu Rangan, the dominance of majority languages: in interviews conducted by other researchers at Microsoft, some people reported changing the language in which meetings were held to one where Copilot was more effective. This effect might shrink or go away as model performance in other languages improves, and improving model performance in non-English languages is a major direction of research at Microsoft and around the world (e.g., Ahuja et al. 2023). Impact of Generative AI on Metacognition (Lev Tankelevitch, Viktor Kewenig, Auste Simkute, Ava Elizabeth Scott, Advait Sarkar, Abigail Sellen, and Sean Rintel) More details available in The Metacognitive Demands and Opportunities of Generative AI (Tankelevitch, Kewenig et al. 2024) Metacognitive demandā€”the effort needed for monitoring and controlling of one used it regularly or they lacked training or manager support for use. A Selection of New Lab Studies While the above research focuses on the use of generative AI in the wild, we are also exploring in a lab setting some of the important trends that real-world use highlights. Given AIā€™s impact appears to vary by role and function, several of these lab studies explore this, diving more deeply into software development and extending the analysis to other important roles like sales and security. Further, because Copilot is deployed globally, weā€™re also starting to see variation across languages, and thus present research studies looking at AI in multilingual contexts. Finally, the complex trade-offs people are starting to make to incorporate AI into their work practices suggests the cognitive mechanisms underlying its use are important to understand, and we share some early work in that space as well. Comparing the Effect of will, looking forward, be substantially redesigned to better integrate AI. Furthermore, generative AI is still under development and the tools that make use of it are improving rapidly. This means not only that the long-term effects of AI on productivity will differ from those observed in the short-term, but that we are likely to continue to see differences between local task effects and more global productivity effects. Research should try to capture and inform changes in workflows, task design, and business processes in addition to productivity effects for fixed tasks. - -One result seen in the above studies and those in our prior work is the common disconnect between the time savings people report from Copilot use and the actual time savings measured. This has been observed not only across studies, where survey measures about time saved tend to be larger than telemetry-based measures, but also within a given study where researchers win across all four metrics. -Our head-to-head measures computed using an LLM evaluator are as follows: -ā€¢Comprehensiveness. How much detail does the answer provide to cover all aspects and details of the question? -ā€¢Diversity. How varied and rich is the answer in providing different perspectives and insights on the question? -ā€¢Empowerment. How well does the answer help the reader understand and make informed judgments about the topic? -ā€¢Directness. How specifically and clearly does the answer address the question? -For our evaluation, the LLM is provided with the question, target metric, and a pair of answers, and asked to assess which answer is better according to the metric, as well as why. It returns the winner if one exists, otherwise a tie if they are fundamentally similar and the differences are negligible. -To account for the stochasticity of LLMs, we run each comparison five times and use mean scores. -Table 2 shows an example of view the role of policy and regulation -Questions: +Write a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the "Natural Language Processing and Information Retrieval" domain, given a list of entities that belong to the community as well as their relationships and optional associated claims. +The analysis will be used to inform researchers and practitioners about significant trends, key contributors, and emerging technologies within the community and their potential impact. + +Domain: "Natural Language Processing and Information Retrieval" +Text: results, +i.e., that na ĀØıve RAG produces the most direct responses across all comparisons. +9Podcast Transcripts News Articles +C0 C1 C2 C3 TS C0 C1 C2 C3 TS +Units 34 367 969 1310 1669 55 555 1797 2142 3197 +Tokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694 +% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100 +Table 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre- +sponding token counts, and percentage of the maximum token count. Map-reduce summarization of +source texts is the most , 21(5):88ā€“92. +Koesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data +sensemaking behaviours. International journal of human-computer studies , 146:102562. +Kuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search +of needles in a 11m haystack: Recurrent memory finds what llms miss. +LangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/. +Laskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via +incorporating query relevance and transfer learning with transformer models. In Advances in +Artificial Intelligence: system for covid-19 scholarly +information management. arXiv preprint arXiv:2005.03975 . +Tang, Y . and Yang, Y . (2024). MultiHop-RAG: Benchmarking retrieval-augmented generation for +multi-hop queries. arXiv preprint arXiv:2401.15391 . +Touvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y ., Bashlykov, N., Batra, S., +Bhargava, P., Bhosale, S., et al. (2023). Llama 2: Open foundation and fine-tuned chat models. +arXiv preprint arXiv:2307.09288 . +Traag, V . A., Waltman, L., and Van Eck, N. J. (2019). From Louvain to Leiden: guaranteeing +well . arXiv preprint arXiv:2306.04136 . +Ban, T., Chen, L., Wang, X., and Chen, H. (2023). From query tools to causal architects: Harnessing +large language models for advanced causal discovery from data. +Baumel, T., Eyal, M., and Elhadad, M. (2018). Query focused abstractive summarization: Incorpo- +rating query relevance, multi-document coverage, and summary length constraints into seq2seq +models. arXiv preprint arXiv:1801.07704 . +Blondel, V . D., Guillaume, J.-L., Lambiotte, R., and Lefebvre, E. (2008). Fast unfolding of +communities in large networks. Journal of statistical mechanics: theory and experiment , +2008(10):P10008. +Brown, T., Mann, B., Ryder, N., Subbia are the same across +all six conditions (except for minor modifications to reference styles to match the types of context +information used). Conditions only differ in how the contents of the context window are created. +The graph index supporting conditions C0-C3was created using our generic prompts for entity and +relationship extraction only, with entity types and few-shot examples tailored to the domain of the +data. The graph indexing process used a context window size of 600 tokens with 1 gleaning for the +Podcast dataset and 0 gleanings for the News dataset. +3.4 Metrics +LLMs have been shown to be good evaluators of natural language generation, achieving state-of- +the-art or competitive results compared against human judgements (Wang et al., 2023a; Zheng et al., +2024). While this approach can generate reference-based metrics when gold standard answers are +known, it is also capable of measuring the qualities of generated texts (e.g., fluency) in The next step is to create report-like summaries of each community in the Leiden hierarchy, using +a method designed to scale to very large datasets. These summaries are independently useful in +their own right as a way to understand the global structure and semantics of the dataset, and may +themselves be used to make sense of a corpus in the absence of a question. For example, a user +may scan through community summaries at one level looking for general themes of interest, then +follow links to the reports at the lower level that provide more details for each of the subtopics. Here, +however, we focus on their utility as part of a graph-based index used for answering global queries. +Community summaries are generated in the following way: +4(a) Root communities at level 0 (b) Sub-communities at level 1 +Figure 3: Graph communities detected using the Leiden algorithm (Traag et al., 2019) over the +MultiHop-RAG (T LLM-generated assessment. +7Question Which public figures are repeatedly mentioned across various entertainment articles? +Graph +RAGOverview of Prominent Public Figures in Entertainment +The entertainment industry is vast and diverse, encompassing film, television, music, sports, and +digital media. Certain public figures stand out due to their significant contributions and influence +across these sectors. The following summary highlights key individuals who are repeatedly +mentioned in various entertainment articles, reflecting their impact and presence within the industry. +Actors and Directors [...]Public Figures in Controversy [...]Musicians and Executives [...] +Athletes and Coaches [...]Influencers and Entrepreneurs [...] +The repeated mention of these figures in entertainment articles signifies their ongoing relevance and +the publicā€™s interest in their work. Their influence spans across various aspects of entertainment, from +shaping cultural narratives in film and television to driving trends in music and digital media. These +individuals not only contribute to their respective fields but also influence the broader cultural , query-focused abstractive summarization that generates natural language summaries and +not just concatenated excerpts (Baumel et al., 2018; Laskar et al., 2020; Yao et al., 2017) . In recent +years, however, such distinctions between summarization tasks that are abstractive versus extractive, +generic versus query-focused, and single-document versus multi-document, have become less rele- +vant. While early applications of the transformer architecture showed substantial improvements on +the state-of-the-art for all such summarization tasks (Goodwin et al., 2020; Laskar et al., 2022; Liu +and Lapata, 2019), these tasks are now trivialized by modern LLMs, including the GPT (Achiam +et al., 2023; Brown et al., 2020), Llama (Touvron et al., 2023), and Gemini (Anil et al., diverse +answers, we therefore used a fixed context window size of 8k tokens for the final evaluation. +3.6 Results +The indexing process resulted in a graph consisting of 8564 nodes and 20691 edges for the Podcast +dataset, and a larger graph of 15754 nodes and 19520 edges for the News dataset. Table 3 shows the +number of community summaries at different levels of each graph community hierarchy. +Global approaches vs. na ĀØıve RAG . As shown in Figure 4, global approaches consistently out- +performed the na ĀØıve RAG ( SS) approach in both comprehensiveness and diversity metrics across +datasets. Specifically, global approaches achieved comprehensiveness win rates between 72-83% +for Podcast transcripts and 72-80% for News articles, while diversity win rates ranged from 75-82% +and 62-71% respectively. Our use of directness as a validity test also achieved the expected generated by the LLM based on short +descriptions of the target datasets. Questions target global understanding rather than specific details. +3 Evaluation +3.1 Datasets +We selected two datasets in the one million token range, each equivalent to about 10 novels of text +and representative of the kind of corpora that users may encounter in their real world activities: +ā€¢Podcast transcripts . Compiled transcripts of podcast conversations between Kevin Scott, +Microsoft CTO, and other technology leaders (Behind the Tech, Scott, 2024). Size: 1669 +Ɨ600-token text chunks, with 100-token overlaps between chunks ( āˆ¼1 million tokens). +ā€¢News articles . Benchmark dataset comprising news articles published from September +2013 to December 2023 in a range of categories, including entertainment, business, sports, +technology, health, and science (MultiHop-RAG; Tang and Yang, 2024). Size: 3197 Ɨ +600-token text chunks, with 100 view the role of policy and regulation +Questions : 1. Which episodes deal primarily with tech policy and government regulation? 2. How do guests perceive the impact of privacy laws on technology development? 3. Do any guests discuss the balance between innovation and ethical considerations? 4. What are the suggested changes to current policies mentioned by the guests? 5. Are collaborations between tech companies and governments discussed and how? -News articlesUser: Educator incorporating current affairs into curricula +News +articlesUser : Educator incorporating current affairs into curricula Task: Teaching about health and wellness -Questions: +Questions : 1. What current topics in health can be integrated into health education curricula? 2. How do news articles address the concepts of preventive medicine and wellness? 3. Are there examples of health articles that contradict each other, and if so, why? 4. What insights can be gleaned about public health priorities based on news coverage? 5. How can educators use the dataset to highlight the importance of health literacy? -Table 1: Examples of potential users, tasks, and questions -landscape, often becoming central figures in social discussions and public discourse. -NaĆÆve -RAGPublic figures who are repeatedly mentioned across various entertainment articles include Taylor Swift, Travis Kelce, Britney Spears, and Justin Timberlake. These individuals have been highlighted for various reasons, ranging from their professional achievements to their personal lives. -Taylor Swift [...]Travis Kelce [...]Britney Spears [...]Justin Timberlake [...] -These figures are frequently covered due to their high-profile status and the publicā€™s interest in their careers and personal lives. Their activities, whether in music, sports, or personal relationships, have significant cultural and economic impacts, as evidenced by the media coverage and public reactions. -LLM -DecisionComprehensiveness: Winner=1 (Graph RAG) -Answer 1 is better because it provides a more comprehensive and detailed list of public figures from a wider range of entertainment sectors, including film, television, music, sports, gaming, and digital language. Researchers explored Copilot in multilingual contexts, examining how Copilot can facilitate collaboration between colleagues with different native languages. - -First, researchers asked 77 native Japanese speakers to review a meeting recorded in English. Half the participants had to watch and listen to the video. The other half could use Copilot Meeting Recap, which gave them an AI meeting summary as well as a chatbot to answer questions about the meeting. Then, researchers asked 83 other native Japanese speakers to review a similar meeting, following the same script, but this time held in Japanese by native Japanese speakers. Again, half of participants had access to Copilot. - -For the meeting in English, participants with Copilot answered 16.4% more multiple-choice questions about the meeting correctly, and they were more than twice as likely to get a perfect score. Moreover, in comparing accuracy between the two scenarios, people ang and Yang, 2024) dataset as indexed. Circles represent entity nodes with size proportional to their degree. Node layout was performed via OpenORD (Martin et al., 2011) and Force Atlas 2 (Jacomy et al., 2014). Node colors represent entity communities, shown at two levels of hierarchical clustering: (a) Level 0, corresponding to the hierarchical partition with maximum modularity, and (b) Level 1, which reveals internal structure within these root-level communities. -ā€¢Leaf-level communities. The element summaries of a leaf-level community (nodes, edges, covariates) are prioritized and then iteratively added to the LLM context window until the token limit is reached. The prioritization is as follows: for each community edge in decreasing order of combined source and target node degree (i.e., overall prominence), add descriptions of the source node, target node, linked covari AG incorporates multiple concepts related to other systems. For example, our community summaries are a kind of self-memory (Selfmem, Cheng et al., 2024) for generation-augmented retrieval (GAR, Mao et al., 2020) that facilitates future generation cycles, while our parallel generation of community answers from these summaries is a kind of iterative (Iter-RetGen, Shao et al., 2023) or federated (FeB4RAG, Wang et al., 2024) retrieval-generation strategy. Other systems have also combined these concepts for multi-document summarization (CAiRE-COVID, Su et al., 2020) and multi-hop question answering (ITRG, Feng et al., 2023; IR-CoT, Trivedi et al., 2022; DSP, Khattab et al., 2022). Our use of a hierarchical index and summarization also bears need for thorough validation and human oversight. Job security was a worry for 10% of respondents, reflecting fears of AI encroaching on their roles. - -These learnings suggest developers view AI as helpful to improving aspects of their workflows, even as they remain uncertain of AIā€™s promise and concerned about threats to their job security. To mitigate the negative effects of this uncertainty on productivity and innovation, and to maintain developersā€™ trust and satisfaction, organizations may identify ways to integrate AI into developers' workflows effectively. These may include acknowledging and addressing concerns and offering training programs. -Problem-Solving Styles and Confidence Generating Prompts for GitHub Copilot (Steven Clarke and Ben Hanrahan) -This study explored how developersā€™ problem-solving styles influence their confidence when generating prompts for GitHub Copilot. The authors hypothesized that variations in developersā€™ problem-solving approaches and workstyles would significantly influence their specific methods, actions, or outcomes. The importance score, measured by a Random Forest statistical model, should be interpreted relatively, as it shows how much each feature helps in predicting AI power usage compared to others. Higher scores indicate greater importance. In this analysis, scores range from 361 to 882, highlighting the significant factors influencing AI power user classification within this dataset and model. - -As with all surveys of this type, it is important to view all the above results through the lens of the limitations of the methodology. While the analysis reveals significant associations, causation cannot be conclusively established due to the observational nature of the data. Similarly, self-selection bias, response bias, and unmeasured confounding variables such as workplace culture and managerial support could influence the outcomes. -Copilot Usage in the Workplace Survey (Alexia Cambon, Alex Farach, Margarita Bermejo as ā€œTranslation and language learning,ā€ ā€œCreative writing and editing,ā€ and ā€œProgramming and scripting.ā€ Overall, 72.9% of the Copilot conversations are in knowledge work domains compared to 37% of Bing Search sessions. The researchers also used GPT-4 to directly classify whether the task associated with each Copilot conversation and or search session was knowledge work (instead of classifying based on the category) and see a similar pattern. - -Researchers then used GPT-4 to classify the main task associated with each conversation or search sessions according to Anderson and Krathwohlā€™s Taxonomy (Anderson and Krathwohl 2001), which defines six categories from lowest complexity (for a human) to highest: Remember, Understand, Apply, Analyze, Evaluate, and Create. Over three-quarters of traditional search sessions, but less than half of Copilot conversations were for ā€œRemember. The content of this report includes an overview of the community's key entities and relationships. -======= -You are an expert in Organizational Sociology and Artificial Intelligence. You are skilled at analyzing social structures, community dynamics, and the impact of technological advancements on workplace environments. You are adept at helping people understand the relationships and organizational frameworks within communities, particularly in the context of how AI influences productivity and interactions in professional settings. - -# Goal -Write a comprehensive assessment report of a community taking on the role of a A community analyst that is examining the impact of generative AI on productivity in workplace environments, given a list of entities that belong to the community as well as their relationships and optional associated claims. -The analysis will be used to inform decision-makers about significant developments associated with the community and their potential impact.. The content of this report includes an overview of the community's key entities and relationships. ->>>>>>> origin/main +Table 1: Examples of potential users, tasks, and questions public figures, primarily from the music industry +and sports, and relies heavily on a single source for data, which makes it less diverse in perspectives +and insights. +Empowerment: Winner=1 (Graph RAG) +Answer 1 is better because it provides a comprehensive and structured overview of public figures +across various sectors of the entertainment industry, including film, television, music, sports, and +digital media. It lists multiple individuals, providing specific examples of their contributions and the +context in which they are mentioned in entertainment articles, along with references to data reports +for each claim. This approach helps the reader understand the breadth of the topic and make informed +judgments without being misled. In contrast, Answer 2 focuses on a smaller group of public figures +and primarily discusses their personal lives and relationships, which may not provide as broad an +understanding of the topic. While Answer 2 also cites sources, it does not match the depth and variety +of Answer 1 final round of +query-focused summarization over all community summaries reporting relevance to that query. +ā€œa motivated, continuous effort to understand connections (which can be among people, places, and +events) in order to anticipate their trajectories and act effectively ā€ (Klein et al., 2006a). Supporting +human-led sensemaking over entire text corpora, however, needs a way for people to both apply and +refine their mental model of the data (Klein et al., 2006b) by asking questions of a global nature. +Retrieval-augmented generation (RAG, Lewis et al., 2020) is an established approach to answering +user questions over entire datasets, but it is designed for situations where these answers are contained +locally within regions of text whose retrieval provides sufficient grounding for the generation task. +Instead, a more appropriate task framing is query-focused summarization (QFS, Dang, 2006), and in +particular examples provided to the LLM for in-context learning (Brown et al., 2020). +3For example, while our default prompt extracting the broad class of ā€œnamed entitiesā€ like people, +places, and organizations is generally applicable, domains with specialized knowledge (e.g., science, +medicine, law) will benefit from few-shot examples specialized to those domains. We also support +a secondary extraction prompt for any additional covariates we would like to associate with the +extracted node instances. Our default covariate prompt aims to extract claims linked to detected +entities, including the subject, object, type, description, source text span, and start and end dates. +To balance the needs of efficiency and quality, we use multiple rounds of ā€œgleaningsā€, up to a +specified maximum, to encourage the LLM to detect any additional entities it may have missed +on prior extraction rounds. This is a multi-stage process in which we first ask the LLM to assess +whether all entities resource-intensive approach requiring the highest number of context tokens. +Root-level community summaries ( C0) require dramatically fewer tokens per query (9x-43x). +Community summaries vs. source texts. When comparing community summaries to source texts +using Graph RAG, community summaries generally provided a small but consistent improvement +in answer comprehensiveness and diversity, except for root-level summaries. Intermediate-level +summaries in the Podcast dataset and low-level community summaries in the News dataset achieved +comprehensiveness win rates of 57% and 64%, respectively. Diversity win rates were 57% for +Podcast intermediate-level summaries and 60% for News low-level community summaries. Table 3 +also illustrates the scalability advantages of Graph RAG compared to source text summarization: for +low-level community summaries ( C3), Graph RAG required 26-33% fewer context tokens, while +for root-level community summaries ( C0), it required over 97% fewer tokens. +Role:. The content of this report includes an overview of the community's key entities and relationships. # Report Structure The report should include the following sections: - TITLE: community's name that represents its key entities - title should be short but specific. When possible, include representative named entities in the title. - SUMMARY: An executive summary of the community's overall structure, how its entities are related to each other, and significant points associated with its entities. -<<<<<<< HEAD -- REPORT RATING: A float score between 0-10 that represents the relevance of the text to the impact and application of generative AI in professional workflows, including software development and multilingual contexts, evaluation metrics for AI performance, and the integration of AI tools like GitHub Copilot in real-world settings, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in promoting understanding and advancement in the field of AI and productivity research. -======= -- REPORT RATING: A float score between 0-10 that represents the relevance of the text to the impact of AI on workplace productivity, organizational dynamics, and the integration of AI tools in professional settings, with 1 being trivial or irrelevant and 10 being highly significant, insightful, and impactful in understanding and improving workplace efficiency and interactions. ->>>>>>> origin/main +- REPORT RATING: A float score between 0-10 that represents the relevance of the text to computational linguistics, data science, and the structural dynamics within specialized communities, with 1 being trivial or irrelevant and 10 being highly significant, impactful, and actionable in advancing the understanding of Natural Language Processing and Information Retrieval. - RATING EXPLANATION: Give a single sentence explanation of the rating. - DETAILED FINDINGS: A list of 5-10 key insights about the community. Each insight should have a short summary followed by multiple paragraphs of explanatory text grounded according to the grounding rules below. Be comprehensive. @@ -90,7 +183,7 @@ Return output as a well-formed JSON-formatted string with the following format. # Grounding Rules After each paragraph, add data record reference if the content of the paragraph was derived from one or more data records. Reference is in the format of [records: (, ... ()]. If there are more than 10 data records, show the top 10 most relevant records. -Each paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use "NONE" if there are no related roles or records. Everything should be in The primary language of the provided text is "English.". +Each paragraph should contain multiple sentences of explanation and concrete examples with specific named entities. All paragraphs must have these references at the start and end. Use "NONE" if there are no related roles or records. Everything should be in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing.. Example paragraph with references added: This is a paragraph of the output text [records: Entities (1, 2, 3), Claims (2, 5), Relationships (10, 12)] diff --git a/graphfleet/prompts/entity_extraction.txt b/graphfleet/prompts/entity_extraction.txt index 448e76c43..57f8e9a14 100644 --- a/graphfleet/prompts/entity_extraction.txt +++ b/graphfleet/prompts/entity_extraction.txt @@ -18,9 +18,9 @@ For each pair of related entities, extract the following information: - relationship_strength: a numeric score indicating strength of the relationship between the source entity and target entity Format each relationship as ("relationship"{tuple_delimiter}{tuple_delimiter}{tuple_delimiter}{tuple_delimiter}) -3. Return output in The primary language of the provided text is "English." as a single list of all the entities and relationships identified in steps 1 and 2. Use **{record_delimiter}** as the list delimiter. +3. Return output in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing. as a single list of all the entities and relationships identified in steps 1 and 2. Use **{record_delimiter}** as the list delimiter. -4. If you have to translate into The primary language of the provided text is "English.", just translate the descriptions, nothing else! +4. If you have to translate into The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing., just translate the descriptions, nothing else! 5. When finished, output {completion_delimiter}. @@ -30,118 +30,75 @@ Format each relationship as ("relationship"{tuple_delimiter}{tupl Example 1: text: -<<<<<<< HEAD - he tasks studied in the lab thus far have tended to -be those for which researchers hypothesized generative AI would -perform well . This was, in fact, the focus of most of the studies -presented in the first AI and Productivity report we published -(Cambon et al. 2023) . Actual information work , however, often -includes a huge variety of tasks and much of the unstructured and -informal work in peopleā€™s jobs is not yet directly supported by the -first-generation of generative AI tools. Software developer -workflows , for example, involve far more than the hands-on coding supported by GitHub Copilot (Meyer et al. 2017). The ability to -shed light on generative AI's productivity dynamics in the natural -complexity of entire workflows is a key advantage of field studies -of generative AIā€™s productivity impacts , and a major reason we -hope to see many more field studies emerging in the literature + results, +i.e., that na ĀØıve RAG produces the most direct responses across all comparisons. +9Podcast Transcripts News Articles +C0 C1 C2 C3 TS C0 C1 C2 C3 TS +Units 34 367 969 1310 1669 55 555 1797 2142 3197 +Tokens 26657 225756 565720 746100 1014611 39770 352641 980898 1140266 1707694 +% Max 2.6 22.2 55.8 73.5 100 2.3 20.7 57.4 66.8 100 +Table 3: Number of context units (community summaries for C0-C3 and text chunks for TS), corre- +sponding token counts, and percentage of the maximum token count. Map-reduce summarization of +source texts is the most ------------------------ output: -("entity"{tuple_delimiter}LAB{tuple_delimiter}LOCATION, RESEARCH ENVIRONMENT{tuple_delimiter}The lab is where tasks are studied to hypothesize the performance of generative AI) +("entity"{tuple_delimiter}RAG{tuple_delimiter}METHOD{tuple_delimiter}RAG (Retrieval-Augmented Generation) is a method that produces direct responses in text generation tasks) {record_delimiter} -("entity"{tuple_delimiter}GENERATIVE AI{tuple_delimiter}TECHNOLOGY, TOOL{tuple_delimiter}Generative AI refers to artificial intelligence systems that can generate content, such as text, based on input data) +("entity"{tuple_delimiter}PODCAST TRANSCRIPTS{tuple_delimiter}DATASET{tuple_delimiter}A dataset consisting of transcripts from podcasts used for analysis) {record_delimiter} -("entity"{tuple_delimiter}AI AND PRODUCTIVITY REPORT{tuple_delimiter}DOCUMENT, PUBLICATION{tuple_delimiter}A report published by Cambon et al. in 2023 focusing on the performance of generative AI in various tasks) +("entity"{tuple_delimiter}NEWS ARTICLES{tuple_delimiter}DATASET{tuple_delimiter}A dataset consisting of news articles used for analysis) {record_delimiter} -("entity"{tuple_delimiter}INFORMATION WORK{tuple_delimiter}ACTIVITY, TASK{tuple_delimiter}Information work includes a variety of tasks, often unstructured and informal, that are part of people's jobs) +("entity"{tuple_delimiter}C0{tuple_delimiter}CATEGORY{tuple_delimiter}A category or cluster used in the analysis, representing a specific subset of the data) {record_delimiter} -("entity"{tuple_delimiter}FIRST-GENERATION GENERATIVE AI TOOLS{tuple_delimiter}TECHNOLOGY, TOOL{tuple_delimiter}The initial versions of generative AI tools that support specific tasks but not the full range of unstructured work) +("entity"{tuple_delimiter}C1{tuple_delimiter}CATEGORY{tuple_delimiter}A category or cluster used in the analysis, representing a specific subset of the data) {record_delimiter} -("entity"{tuple_delimiter}SOFTWARE DEVELOPER WORKFLOWS{tuple_delimiter}ACTIVITY, TASK{tuple_delimiter}The comprehensive set of activities involved in software development, beyond just coding) +("entity"{tuple_delimiter}C2{tuple_delimiter}CATEGORY{tuple_delimiter}A category or cluster used in the analysis, representing a specific subset of the data) {record_delimiter} -("entity"{tuple_delimiter}GITHUB COPILOT{tuple_delimiter}TECHNOLOGY, TOOL{tuple_delimiter}A generative AI tool that assists with hands-on coding tasks for software developers) +("entity"{tuple_delimiter}C3{tuple_delimiter}CATEGORY{tuple_delimiter}A category or cluster used in the analysis, representing a specific subset of the data) {record_delimiter} -("entity"{tuple_delimiter}FIELD STUDIES{tuple_delimiter}RESEARCH METHOD, STUDY TYPE{tuple_delimiter}Studies conducted in natural settings to observe the real-world impacts of generative AI on productivity) +("entity"{tuple_delimiter}TS{tuple_delimiter}CATEGORY{tuple_delimiter}A category or cluster used in the analysis, representing a specific subset of the data) {record_delimiter} -("entity"{tuple_delimiter}PRODUCTIVITY DYNAMICS{tuple_delimiter}CONCEPT, PHENOMENON{tuple_delimiter}The various factors and interactions that affect productivity when using generative AI in real-world workflows) +("entity"{tuple_delimiter}UNITS{tuple_delimiter}METRIC{tuple_delimiter}The number of context units, such as community summaries or text chunks, used in the analysis) {record_delimiter} -("entity"{tuple_delimiter}LITERATURE{tuple_delimiter}BODY OF WORK, RESEARCH{tuple_delimiter}The collection of academic and professional studies and publications on a given topic) +("entity"{tuple_delimiter}TOKENS{tuple_delimiter}METRIC{tuple_delimiter}The number of tokens, or individual words, used in the analysis) {record_delimiter} -("relationship"{tuple_delimiter}LAB{tuple_delimiter}GENERATIVE AI{tuple_delimiter}The lab is where tasks are studied to hypothesize the performance of generative AI{tuple_delimiter}7) +("entity"{tuple_delimiter}% MAX{tuple_delimiter}METRIC{tuple_delimiter}The percentage of the maximum token count used in the analysis) {record_delimiter} -("relationship"{tuple_delimiter}AI AND PRODUCTIVITY REPORT{tuple_delimiter}GENERATIVE AI{tuple_delimiter}The report focuses on the performance of generative AI in various tasks{tuple_delimiter}8) +("relationship"{tuple_delimiter}RAG{tuple_delimiter}PODCAST TRANSCRIPTS{tuple_delimiter}RAG is used to produce direct responses from podcast transcripts{tuple_delimiter}7) {record_delimiter} -("relationship"{tuple_delimiter}INFORMATION WORK{tuple_delimiter}FIRST-GENERATION GENERATIVE AI TOOLS{tuple_delimiter}First-generation generative AI tools do not yet directly support much of the unstructured and informal information work{tuple_delimiter}6) +("relationship"{tuple_delimiter}RAG{tuple_delimiter}NEWS ARTICLES{tuple_delimiter}RAG is used to produce direct responses from news articles{tuple_delimiter}7) {record_delimiter} -("relationship"{tuple_delimiter}SOFTWARE DEVELOPER WORKFLOWS{tuple_delimiter}GITHUB COPILOT{tuple_delimiter}GitHub Copilot supports hands-on coding, which is a part of software developer workflows{tuple_delimiter}9) +("relationship"{tuple_delimiter}C0{tuple_delimiter}PODCAST TRANSCRIPTS{tuple_delimiter}C0 is a category used in the analysis of podcast transcripts{tuple_delimiter}5) {record_delimiter} -("relationship"{tuple_delimiter}FIELD STUDIES{tuple_delimiter}PRODUCTIVITY DYNAMICS{tuple_delimiter}Field studies help shed light on the productivity dynamics of generative AI in real-world settings{tuple_delimiter}8) +("relationship"{tuple_delimiter}C1{tuple_delimiter}PODCAST TRANSCRIPTS{tuple_delimiter}C1 is a category used in the analysis of podcast transcripts{tuple_delimiter}5) {record_delimiter} -("relationship"{tuple_delimiter}FIELD STUDIES{tuple_delimiter}LITERATURE{tuple_delimiter}The hope is to see more field studies emerging in the literature to understand generative AI's productivity impacts{tuple_delimiter}7) -======= - (2024). The -Use of Generative Search Engines for Knowledge Work and Complex -Tasks. arXiv preprint. https://doi.org/10.48550/arXiv.2404.04268 . -Tankelevitch *, L., Kewenig *, V., Simkute, A., Scott, A. E., Sarkar, A., Sellen, A., & -Rintel, S. (2024). The Metacognitive Demands and Opportunities of Generative -AI. In Proceedings of the CHI Conference on Human Factors in Computing -Systems (pp. 1-24). https://doi.org/10.1145/3613904.3642902 . -Taraborelli, D. (2015). The Sum of All Human Knowledge in the Age of Machines: A -New Research Agenda for Wikimedia. ICWSM -15 Workshop on Wikipedia, a -Social Pedia: Research Challenges and Opportunities . ------------------------- -output: -("entity"{tuple_delimiter}GENERATIVE SEARCH ENGINES{tuple_delimiter}TECHNOLOGY{tuple_delimiter}Generative search engines are AI-driven tools used for knowledge work and complex tasks) -{record_delimiter} -("entity"{tuple_delimiter}KNOWLEDGE WORK{tuple_delimiter}ACTIVITY{tuple_delimiter}Knowledge work involves tasks that require significant cognitive effort and expertise) -{record_delimiter} -("entity"{tuple_delimiter}COMPLEX TASKS{tuple_delimiter}ACTIVITY{tuple_delimiter}Complex tasks are activities that require advanced problem-solving and critical thinking skills) -{record_delimiter} -("entity"{tuple_delimiter}ARXIV{tuple_delimiter}PLATFORM{tuple_delimiter}arXiv is an open-access repository for research papers) -{record_delimiter} -("entity"{tuple_delimiter}CHI CONFERENCE{tuple_delimiter}EVENT{tuple_delimiter}The CHI Conference on Human Factors in Computing Systems is an annual event focusing on human-computer interaction) -{record_delimiter} -("entity"{tuple_delimiter}WIKIMEDIA{tuple_delimiter}ORGANIZATION{tuple_delimiter}Wikimedia is a non-profit organization that supports Wikipedia and other free knowledge projects) -{record_delimiter} -("entity"{tuple_delimiter}TANKELEVITCH, L.{tuple_delimiter}PERSON{tuple_delimiter}L. Tankelevitch is an author of a paper on generative search engines for knowledge work and complex tasks) +("relationship"{tuple_delimiter}C2{tuple_delimiter}PODCAST TRANSCRIPTS{tuple_delimiter}C2 is a category used in the analysis of podcast transcripts{tuple_delimiter}5) {record_delimiter} -("entity"{tuple_delimiter}KEWENIG, V.{tuple_delimiter}PERSON{tuple_delimiter}V. Kewenig is an author of a paper on generative search engines for knowledge work and complex tasks) +("relationship"{tuple_delimiter}C3{tuple_delimiter}PODCAST TRANSCRIPTS{tuple_delimiter}C3 is a category used in the analysis of podcast transcripts{tuple_delimiter}5) {record_delimiter} -("entity"{tuple_delimiter}SIMKUTE, A.{tuple_delimiter}PERSON{tuple_delimiter}A. Simkute is an author of a paper on generative search engines for knowledge work and complex tasks) +("relationship"{tuple_delimiter}TS{tuple_delimiter}PODCAST TRANSCRIPTS{tuple_delimiter}TS is a category used in the analysis of podcast transcripts{tuple_delimiter}5) {record_delimiter} -("entity"{tuple_delimiter}SCOTT, A. E.{tuple_delimiter}PERSON{tuple_delimiter}A. E. Scott is an author of a paper on generative search engines for knowledge work and complex tasks) +("relationship"{tuple_delimiter}C0{tuple_delimiter}NEWS ARTICLES{tuple_delimiter}C0 is a category used in the analysis of news articles{tuple_delimiter}5) {record_delimiter} -("entity"{tuple_delimiter}SARKAR, A.{tuple_delimiter}PERSON{tuple_delimiter}A. Sarkar is an author of a paper on generative search engines for knowledge work and complex tasks) +("relationship"{tuple_delimiter}C1{tuple_delimiter}NEWS ARTICLES{tuple_delimiter}C1 is a category used in the analysis of news articles{tuple_delimiter}5) {record_delimiter} -("entity"{tuple_delimiter}SELLEN, A.{tuple_delimiter}PERSON{tuple_delimiter}A. Sellen is an author of a paper on generative search engines for knowledge work and complex tasks) +("relationship"{tuple_delimiter}C2{tuple_delimiter}NEWS ARTICLES{tuple_delimiter}C2 is a category used in the analysis of news articles{tuple_delimiter}5) {record_delimiter} -("entity"{tuple_delimiter}RINTEL, S.{tuple_delimiter}PERSON{tuple_delimiter}S. Rintel is an author of a paper on generative search engines for knowledge work and complex tasks) +("relationship"{tuple_delimiter}C3{tuple_delimiter}NEWS ARTICLES{tuple_delimiter}C3 is a category used in the analysis of news articles{tuple_delimiter}5) {record_delimiter} -("entity"{tuple_delimiter}TARABORELLI, D.{tuple_delimiter}PERSON{tuple_delimiter}D. Taraborelli is an author of a paper on the impact of machines on human knowledge) +("relationship"{tuple_delimiter}TS{tuple_delimiter}NEWS ARTICLES{tuple_delimiter}TS is a category used in the analysis of news articles{tuple_delimiter}5) {record_delimiter} -("entity"{tuple_delimiter}ICWSM{tuple_delimiter}EVENT{tuple_delimiter}ICWSM is a conference focusing on social media research) +("relationship"{tuple_delimiter}UNITS{tuple_delimiter}PODCAST TRANSCRIPTS{tuple_delimiter}Units are used to measure the context in podcast transcripts{tuple_delimiter}6) {record_delimiter} -("relationship"{tuple_delimiter}GENERATIVE SEARCH ENGINES{tuple_delimiter}KNOWLEDGE WORK{tuple_delimiter}Generative search engines are used for knowledge work{tuple_delimiter}8) +("relationship"{tuple_delimiter}UNITS{tuple_delimiter}NEWS ARTICLES{tuple_delimiter}Units are used to measure the context in news articles{tuple_delimiter}6) {record_delimiter} -("relationship"{tuple_delimiter}GENERATIVE SEARCH ENGINES{tuple_delimiter}COMPLEX TASKS{tuple_delimiter}Generative search engines are used for complex tasks{tuple_delimiter}8) +("relationship"{tuple_delimiter}TOKENS{tuple_delimiter}PODCAST TRANSCRIPTS{tuple_delimiter}Tokens are used to measure the word count in podcast transcripts{tuple_delimiter}6) {record_delimiter} -("relationship"{tuple_delimiter}TANKELEVITCH, L.{tuple_delimiter}GENERATIVE SEARCH ENGINES{tuple_delimiter}L. Tankelevitch authored a paper on generative search engines{tuple_delimiter}7) +("relationship"{tuple_delimiter}TOKENS{tuple_delimiter}NEWS ARTICLES{tuple_delimiter}Tokens are used to measure the word count in news articles{tuple_delimiter}6) {record_delimiter} -("relationship"{tuple_delimiter}KEWENIG, V.{tuple_delimiter}GENERATIVE SEARCH ENGINES{tuple_delimiter}V. Kewenig authored a paper on generative search engines{tuple_delimiter}7) +("relationship"{tuple_delimiter}% MAX{tuple_delimiter}PODCAST TRANSCRIPTS{tuple_delimiter}% Max is used to measure the percentage of maximum token count in podcast transcripts{tuple_delimiter}6) {record_delimiter} -("relationship"{tuple_delimiter}SIMKUTE, A.{tuple_delimiter}GENERATIVE SEARCH ENGINES{tuple_delimiter}A. Simkute authored a paper on generative search engines{tuple_delimiter}7) -{record_delimiter} -("relationship"{tuple_delimiter}SCOTT, A. E.{tuple_delimiter}GENERATIVE SEARCH ENGINES{tuple_delimiter}A. E. Scott authored a paper on generative search engines{tuple_delimiter}7) -{record_delimiter} -("relationship"{tuple_delimiter}SARKAR, A.{tuple_delimiter}GENERATIVE SEARCH ENGINES{tuple_delimiter}A. Sarkar authored a paper on generative search engines{tuple_delimiter}7) -{record_delimiter} -("relationship"{tuple_delimiter}SELLEN, A.{tuple_delimiter}GENERATIVE SEARCH ENGINES{tuple_delimiter}A. Sellen authored a paper on generative search engines{tuple_delimiter}7) -{record_delimiter} -("relationship"{tuple_delimiter}RINTEL, S.{tuple_delimiter}GENERATIVE SEARCH ENGINES{tuple_delimiter}S. Rintel authored a paper on generative search engines{tuple_delimiter}7) -{record_delimiter} -("relationship"{tuple_delimiter}TARABORELLI, D.{tuple_delimiter}WIKIMEDIA{tuple_delimiter}D. Taraborelli authored a paper on Wikimedia's research agenda{tuple_delimiter}6) -{record_delimiter} -("relationship"{tuple_delimiter}ICWSM{tuple_delimiter}WIKIMEDIA{tuple_delimiter}Wikimedia's research agenda was presented at the ICWSM conference{tuple_delimiter}5) ->>>>>>> origin/main +("relationship"{tuple_delimiter}% MAX{tuple_delimiter}NEWS ARTICLES{tuple_delimiter}% Max is used to measure the percentage of maximum token count in news articles{tuple_delimiter}6) {completion_delimiter} ############################# @@ -149,121 +106,100 @@ output: Example 2: text: -<<<<<<< HEAD -Liu, N. F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F., and Liang, P. (2023). Lost -in the middle: How language models use long contexts. arXiv:2307.03172. -Liu, Y . and Lapata, M. (2019). Hierarchical transformers for multi-document summarization. arXiv -preprint arXiv:1905.13164 . -LlamaIndex (2024). LlamaIndex Knowledge Graph Index. https://docs .llamaindex .ai/en/stable/ -examples/index structs/knowledge graph/KnowledgeGraphDemo .html. -Manakul, P., Liusie, A., and Gales, M. J. (2023). Selfcheckgpt: Zero-resource black-box hallucina- -tion detection for generative large language models. arXiv preprint ar + , 21(5):88ā€“92. +Koesten, L., Gregory, K., Groth, P., and Simperl, E. (2021). Talking datasetsā€“understanding data +sensemaking behaviours. International journal of human-computer studies , 146:102562. +Kuratov, Y ., Bulatov, A., Anokhin, P., Sorokin, D., Sorokin, A., and Burtsev, M. (2024). In search +of needles in a 11m haystack: Recurrent memory finds what llms miss. +LangChain (2024). Langchain graphs. https://python .langchain .com/docs/use cases/graph/. +Laskar, M. T. R., Hoque, E., and Huang, J. (2020). Query focused abstractive summarization via +incorporating query relevance and transfer learning with transformer models. In Advances in +Artificial Intelligence: ------------------------ output: -("entity"{tuple_delimiter}LIU, N. F.{tuple_delimiter}PERSON{tuple_delimiter}Liu, N. F. is an author of the paper "Lost in the middle: How language models use long contexts" published in 2023) +("entity"{tuple_delimiter}KOESTEN, L.{tuple_delimiter}PERSON{tuple_delimiter}Koesten, L. is an author of the paper "Talking datasetsā€“understanding data sensemaking behaviours") {record_delimiter} -("entity"{tuple_delimiter}LIN, K.{tuple_delimiter}PERSON{tuple_delimiter}Lin, K. is an author of the paper "Lost in the middle: How language models use long contexts" published in 2023) +("entity"{tuple_delimiter}GREGORY, K.{tuple_delimiter}PERSON{tuple_delimiter}Gregory, K. is an author of the paper "Talking datasetsā€“understanding data sensemaking behaviours") {record_delimiter} -("entity"{tuple_delimiter}HEWITT, J.{tuple_delimiter}PERSON{tuple_delimiter}Hewitt, J. is an author of the paper "Lost in the middle: How language models use long contexts" published in 2023) +("entity"{tuple_delimiter}GROTH, P.{tuple_delimiter}PERSON{tuple_delimiter}Groth, P. is an author of the paper "Talking datasetsā€“understanding data sensemaking behaviours") {record_delimiter} -("entity"{tuple_delimiter}PARANJAPE, A.{tuple_delimiter}PERSON{tuple_delimiter}Paranjape, A. is an author of the paper "Lost in the middle: How language models use long contexts" published in 2023) +("entity"{tuple_delimiter}SIMPERL, E.{tuple_delimiter}PERSON{tuple_delimiter}Simperl, E. is an author of the paper "Talking datasetsā€“understanding data sensemaking behaviours") {record_delimiter} -("entity"{tuple_delimiter}BEVILACQUA, M.{tuple_delimiter}PERSON{tuple_delimiter}Bevilacqua, M. is an author of the paper "Lost in the middle: How language models use long contexts" published in 2023) +("entity"{tuple_delimiter}INTERNATIONAL JOURNAL OF HUMAN-COMPUTER STUDIES{tuple_delimiter}PUBLICATION{tuple_delimiter}The journal where the paper "Talking datasetsā€“understanding data sensemaking behaviours" was published) {record_delimiter} -("entity"{tuple_delimiter}PETRONI, F.{tuple_delimiter}PERSON{tuple_delimiter}Petroni, F. is an author of the paper "Lost in the middle: How language models use long contexts" published in 2023) +("entity"{tuple_delimiter}KURATOV, Y.{tuple_delimiter}PERSON{tuple_delimiter}Kuratov, Y. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss") {record_delimiter} -("entity"{tuple_delimiter}LIANG, P.{tuple_delimiter}PERSON{tuple_delimiter}Liang, P. is an author of the paper "Lost in the middle: How language models use long contexts" published in 2023) +("entity"{tuple_delimiter}BULATOV, A.{tuple_delimiter}PERSON{tuple_delimiter}Bulatov, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss") {record_delimiter} -("entity"{tuple_delimiter}LIU, Y.{tuple_delimiter}PERSON{tuple_delimiter}Liu, Y. is an author of the paper "Hierarchical transformers for multi-document summarization" published in 2019) +("entity"{tuple_delimiter}ANOKHIN, P.{tuple_delimiter}PERSON{tuple_delimiter}Anokhin, P. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss") {record_delimiter} -("entity"{tuple_delimiter}LAPATA, M.{tuple_delimiter}PERSON{tuple_delimiter}Lapata, M. is an author of the paper "Hierarchical transformers for multi-document summarization" published in 2019) +("entity"{tuple_delimiter}SOROKIN, D.{tuple_delimiter}PERSON{tuple_delimiter}Sorokin, D. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss") {record_delimiter} -("entity"{tuple_delimiter}LLAMAINDEX{tuple_delimiter}ORGANIZATION{tuple_delimiter}LlamaIndex is an organization that developed the LlamaIndex Knowledge Graph Index, with documentation available online) +("entity"{tuple_delimiter}SOROKIN, A.{tuple_delimiter}PERSON{tuple_delimiter}Sorokin, A. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss") {record_delimiter} -("entity"{tuple_delimiter}MANAKUL, P.{tuple_delimiter}PERSON{tuple_delimiter}Manakul, P. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" published in 2023) +("entity"{tuple_delimiter}BURTSEV, M.{tuple_delimiter}PERSON{tuple_delimiter}Burtsev, M. is an author of the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss") {record_delimiter} -("entity"{tuple_delimiter}LIUSIE, A.{tuple_delimiter}PERSON{tuple_delimiter}Liusie, A. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" published in 2023) +("entity"{tuple_delimiter}LANGCHAIN{tuple_delimiter}ORGANIZATION{tuple_delimiter}LangChain is an organization that developed Langchain graphs) {record_delimiter} -("entity"{tuple_delimiter}GALES, M. J.{tuple_delimiter}PERSON{tuple_delimiter}Gales, M. J. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models" published in 2023) +("entity"{tuple_delimiter}LANGCHAIN GRAPHS{tuple_delimiter}TECHNOLOGY{tuple_delimiter}Langchain graphs is a technology developed by LangChain) {record_delimiter} -("entity"{tuple_delimiter}LOST IN THE MIDDLE: HOW LANGUAGE MODELS USE LONG CONTEXTS{tuple_delimiter}DOCUMENT{tuple_delimiter}A paper published in 2023 by Liu, N. F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F., and Liang, P.) +("entity"{tuple_delimiter}LASKAR, M. T. R.{tuple_delimiter}PERSON{tuple_delimiter}Laskar, M. T. R. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models") {record_delimiter} -("entity"{tuple_delimiter}HIERARCHICAL TRANSFORMERS FOR MULTI-DOCUMENT SUMMARIZATION{tuple_delimiter}DOCUMENT{tuple_delimiter}A paper published in 2019 by Liu, Y. and Lapata, M.) +("entity"{tuple_delimiter}HOQUE, E.{tuple_delimiter}PERSON{tuple_delimiter}Hoque, E. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models") {record_delimiter} -("entity"{tuple_delimiter}SELFCHECKGPT: ZERO-RESOURCE BLACK-BOX HALLUCINATION DETECTION FOR GENERATIVE LARGE LANGUAGE MODELS{tuple_delimiter}DOCUMENT{tuple_delimiter}A paper published in 2023 by Manakul, P., Liusie, A., and Gales, M. J.) +("entity"{tuple_delimiter}HUANG, J.{tuple_delimiter}PERSON{tuple_delimiter}Huang, J. is an author of the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models") {record_delimiter} -("relationship"{tuple_delimiter}LIU, N. F.{tuple_delimiter}LOST IN THE MIDDLE: HOW LANGUAGE MODELS USE LONG CONTEXTS{tuple_delimiter}Liu, N. F. is an author of the paper "Lost in the middle: How language models use long contexts"{tuple_delimiter}9) +("entity"{tuple_delimiter}ADVANCES IN ARTIFICIAL INTELLIGENCE{tuple_delimiter}PUBLICATION{tuple_delimiter}The conference where the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models" was presented) {record_delimiter} -("relationship"{tuple_delimiter}LIN, K.{tuple_delimiter}LOST IN THE MIDDLE: HOW LANGUAGE MODELS USE LONG CONTEXTS{tuple_delimiter}Lin, K. is an author of the paper "Lost in the middle: How language models use long contexts"{tuple_delimiter}9) +("relationship"{tuple_delimiter}KOESTEN, L.{tuple_delimiter}GREGORY, K.{tuple_delimiter}Koesten, L. and Gregory, K. co-authored the paper "Talking datasetsā€“understanding data sensemaking behaviours"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}HEWITT, J.{tuple_delimiter}LOST IN THE MIDDLE: HOW LANGUAGE MODELS USE LONG CONTEXTS{tuple_delimiter}Hewitt, J. is an author of the paper "Lost in the middle: How language models use long contexts"{tuple_delimiter}9) +("relationship"{tuple_delimiter}KOESTEN, L.{tuple_delimiter}GROTH, P.{tuple_delimiter}Koesten, L. and Groth, P. co-authored the paper "Talking datasetsā€“understanding data sensemaking behaviours"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}PARANJAPE, A.{tuple_delimiter}LOST IN THE MIDDLE: HOW LANGUAGE MODELS USE LONG CONTEXTS{tuple_delimiter}Paranjape, A. is an author of the paper "Lost in the middle: How language models use long contexts"{tuple_delimiter}9) +("relationship"{tuple_delimiter}KOESTEN, L.{tuple_delimiter}SIMPERL, E.{tuple_delimiter}Koesten, L. and Simperl, E. co-authored the paper "Talking datasetsā€“understanding data sensemaking behaviours"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}BEVILACQUA, M.{tuple_delimiter}LOST IN THE MIDDLE: HOW LANGUAGE MODELS USE LONG CONTEXTS{tuple_delimiter}Bevilacqua, M. is an author of the paper "Lost in the middle: How language models use long contexts"{tuple_delimiter}9) +("relationship"{tuple_delimiter}GREGORY, K.{tuple_delimiter}GROTH, P.{tuple_delimiter}Gregory, K. and Groth, P. co-authored the paper "Talking datasetsā€“understanding data sensemaking behaviours"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}PETRONI, F.{tuple_delimiter}LOST IN THE MIDDLE: HOW LANGUAGE MODELS USE LONG CONTEXTS{tuple_delimiter}Petroni, F. is an author of the paper "Lost in the middle: How language models use long contexts"{tuple_delimiter}9) +("relationship"{tuple_delimiter}GREGORY, K.{tuple_delimiter}SIMPERL, E.{tuple_delimiter}Gregory, K. and Simperl, E. co-authored the paper "Talking datasetsā€“understanding data sensemaking behaviours"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}LIANG, P.{tuple_delimiter}LOST IN THE MIDDLE: HOW LANGUAGE MODELS USE LONG CONTEXTS{tuple_delimiter}Liang, P. is an author of the paper "Lost in the middle: How language models use long contexts"{tuple_delimiter}9) +("relationship"{tuple_delimiter}GROTH, P.{tuple_delimiter}SIMPERL, E.{tuple_delimiter}Groth, P. and Simperl, E. co-authored the paper "Talking datasetsā€“understanding data sensemaking behaviours"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}LIU, Y.{tuple_delimiter}HIERARCHICAL TRANSFORMERS FOR MULTI-DOCUMENT SUMMARIZATION{tuple_delimiter}Liu, Y. is an author of the paper "Hierarchical transformers for multi-document summarization"{tuple_delimiter}9) +("relationship"{tuple_delimiter}KURATOV, Y.{tuple_delimiter}BULATOV, A.{tuple_delimiter}Kuratov, Y. and Bulatov, A. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}LAPATA, M.{tuple_delimiter}HIERARCHICAL TRANSFORMERS FOR MULTI-DOCUMENT SUMMARIZATION{tuple_delimiter}Lapata, M. is an author of the paper "Hierarchical transformers for multi-document summarization"{tuple_delimiter}9) +("relationship"{tuple_delimiter}KURATOV, Y.{tuple_delimiter}ANOKHIN, P.{tuple_delimiter}Kuratov, Y. and Anokhin, P. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}MANAKUL, P.{tuple_delimiter}SELFCHECKGPT: ZERO-RESOURCE BLACK-BOX HALLUCINATION DETECTION FOR GENERATIVE LARGE LANGUAGE MODELS{tuple_delimiter}Manakul, P. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models"{tuple_delimiter}9) +("relationship"{tuple_delimiter}KURATOV, Y.{tuple_delimiter}SOROKIN, D.{tuple_delimiter}Kuratov, Y. and Sorokin, D. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}LIUSIE, A.{tuple_delimiter}SELFCHECKGPT: ZERO-RESOURCE BLACK-BOX HALLUCINATION DETECTION FOR GENERATIVE LARGE LANGUAGE MODELS{tuple_delimiter}Liusie, A. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models"{tuple_delimiter}9) +("relationship"{tuple_delimiter}KURATOV, Y.{tuple_delimiter}SOROKIN, A.{tuple_delimiter}Kuratov, Y. and Sorokin, A. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}GALES, M. J.{tuple_delimiter}SELFCHECKGPT: ZERO-RESOURCE BLACK-BOX HALLUCINATION DETECTION FOR GENERATIVE LARGE LANGUAGE MODELS{tuple_delimiter}Gales, M. J. is an author of the paper "Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models"{tuple_delimiter}9) -======= -: The Second Microsoft Report on AI and Productivity -Research. Microsoft. -1 INTRODUCTION -There is tremendous interest in how AI can increase peopleā€™s -productivity at work. To help meet this interest, in December 2023, -Microsoft released a first AI and Productivity Report (Cambon et -al. 2023) synthesizing the results of many Microsoft studies on AI -and productivity. These studies contributed to a large and growing -literature from around the world and a wide variety of disciplines. -Although there are exceptions, this literature largely points to a -broad conclusion: Generative AI tools have the potential to -introduce a substantial step -function increase in productivity for tasks performed by information workers (e.g. , Noy and Zhang -2023; Dellā€™Acqua et al. 2023; Brynjolfsson et al. 2023; Peng et al. -2023). - -However, much of this ------------------------- -output: -("entity"{tuple_delimiter}MICROSOFT{tuple_delimiter}ORGANIZATION{tuple_delimiter}Microsoft is a technology company that released the AI and Productivity Report) +("relationship"{tuple_delimiter}KURATOV, Y.{tuple_delimiter}BURTSEV, M.{tuple_delimiter}Kuratov, Y. and Burtsev, M. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("entity"{tuple_delimiter}AI AND PRODUCTIVITY REPORT{tuple_delimiter}DOCUMENT{tuple_delimiter}The AI and Productivity Report is a document released by Microsoft in December 2023, synthesizing results of studies on AI and productivity) +("relationship"{tuple_delimiter}BULATOV, A.{tuple_delimiter}ANOKHIN, P.{tuple_delimiter}Bulatov, A. and Anokhin, P. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("entity"{tuple_delimiter}CAMBON ET AL. 2023{tuple_delimiter}STUDY{tuple_delimiter}A study included in the AI and Productivity Report, authored by Cambon and others in 2023) +("relationship"{tuple_delimiter}BULATOV, A.{tuple_delimiter}SOROKIN, D.{tuple_delimiter}Bulatov, A. and Sorokin, D. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("entity"{tuple_delimiter}NOY AND ZHANG 2023{tuple_delimiter}STUDY{tuple_delimiter}A study referenced in the AI and Productivity Report, authored by Noy and Zhang in 2023) +("relationship"{tuple_delimiter}BULATOV, A.{tuple_delimiter}SOROKIN, A.{tuple_delimiter}Bulatov, A. and Sorokin, A. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("entity"{tuple_delimiter}DELLā€™ACQUA ET AL. 2023{tuple_delimiter}STUDY{tuple_delimiter}A study referenced in the AI and Productivity Report, authored by Dellā€™Acqua and others in 2023) +("relationship"{tuple_delimiter}BULATOV, A.{tuple_delimiter}BURTSEV, M.{tuple_delimiter}Bulatov, A. and Burtsev, M. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("entity"{tuple_delimiter}BRYNJOLFSSON ET AL. 2023{tuple_delimiter}STUDY{tuple_delimiter}A study referenced in the AI and Productivity Report, authored by Brynjolfsson and others in 2023) +("relationship"{tuple_delimiter}ANOKHIN, P.{tuple_delimiter}SOROKIN, D.{tuple_delimiter}Anokhin, P. and Sorokin, D. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("entity"{tuple_delimiter}PENG ET AL. 2023{tuple_delimiter}STUDY{tuple_delimiter}A study referenced in the AI and Productivity Report, authored by Peng and others in 2023) +("relationship"{tuple_delimiter}ANOKHIN, P.{tuple_delimiter}SOROKIN, A.{tuple_delimiter}Anokhin, P. and Sorokin, A. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("entity"{tuple_delimiter}GENERATIVE AI TOOLS{tuple_delimiter}TECHNOLOGY{tuple_delimiter}Generative AI tools are technologies that have the potential to significantly increase productivity for information workers) +("relationship"{tuple_delimiter}ANOKHIN, P.{tuple_delimiter}BURTSEV, M.{tuple_delimiter}Anokhin, P. and Burtsev, M. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}MICROSOFT{tuple_delimiter}AI AND PRODUCTIVITY REPORT{tuple_delimiter}Microsoft released the AI and Productivity Report in December 2023{tuple_delimiter}9) +("relationship"{tuple_delimiter}SOROKIN, D.{tuple_delimiter}SOROKIN, A.{tuple_delimiter}Sorokin, D. and Sorokin, A. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}AI AND PRODUCTIVITY REPORT{tuple_delimiter}CAMBON ET AL. 2023{tuple_delimiter}The AI and Productivity Report includes the study by Cambon et al. 2023{tuple_delimiter}8) +("relationship"{tuple_delimiter}SOROKIN, D.{tuple_delimiter}BURTSEV, M.{tuple_delimiter}Sorokin, D. and Burtsev, M. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}AI AND PRODUCTIVITY REPORT{tuple_delimiter}NOY AND ZHANG 2023{tuple_delimiter}The AI and Productivity Report references the study by Noy and Zhang 2023{tuple_delimiter}8) +("relationship"{tuple_delimiter}SOROKIN, A.{tuple_delimiter}BURTSEV, M.{tuple_delimiter}Sorokin, A. and Burtsev, M. co-authored the paper "In search of needles in a 11m haystack: Recurrent memory finds what llms miss"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}AI AND PRODUCTIVITY REPORT{tuple_delimiter}DELLā€™ACQUA ET AL. 2023{tuple_delimiter}The AI and Productivity Report references the study by Dellā€™Acqua et al. 2023{tuple_delimiter}8) +("relationship"{tuple_delimiter}LANGCHAIN{tuple_delimiter}LANGCHAIN GRAPHS{tuple_delimiter}LangChain developed Langchain graphs{tuple_delimiter}9) {record_delimiter} -("relationship"{tuple_delimiter}AI AND PRODUCTIVITY REPORT{tuple_delimiter}BRYNJOLFSSON ET AL. 2023{tuple_delimiter}The AI and Productivity Report references the study by Brynjolfsson et al. 2023{tuple_delimiter}8) +("relationship"{tuple_delimiter}LASKAR, M. T. R.{tuple_delimiter}HOQUE, E.{tuple_delimiter}Laskar, M. T. R. and Hoque, E. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}AI AND PRODUCTIVITY REPORT{tuple_delimiter}PENG ET AL. 2023{tuple_delimiter}The AI and Productivity Report references the study by Peng et al. 2023{tuple_delimiter}8) +("relationship"{tuple_delimiter}LASKAR, M. T. R.{tuple_delimiter}HUANG, J.{tuple_delimiter}Laskar, M. T. R. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models"{tuple_delimiter}8) {record_delimiter} -("relationship"{tuple_delimiter}GENERATIVE AI TOOLS{tuple_delimiter}INFORMATION WORKERS{tuple_delimiter}Generative AI tools have the potential to significantly increase productivity for information workers{tuple_delimiter}7) ->>>>>>> origin/main +("relationship"{tuple_delimiter}HOQUE, E.{tuple_delimiter}HUANG, J.{tuple_delimiter}Hoque, E. and Huang, J. co-authored the paper "Query focused abstractive summarization via incorporating query relevance and transfer learning with transformer models"{tuple_delimiter}8) {completion_delimiter} ############################# diff --git a/graphfleet/prompts/summarize_descriptions.txt b/graphfleet/prompts/summarize_descriptions.txt index 252d99c31..4260c51b7 100644 --- a/graphfleet/prompts/summarize_descriptions.txt +++ b/graphfleet/prompts/summarize_descriptions.txt @@ -1,14 +1,10 @@ -<<<<<<< HEAD -You are an expert in Artificial Intelligence and Productivity Research. You are skilled at analyzing the impact and application of generative AI in professional workflows, including software development and multilingual contexts. You are adept at helping people understand the evaluation metrics for AI performance and the integration of AI tools like GitHub Copilot in real-world settings. -======= -You are an expert in Organizational Sociology and Artificial Intelligence. You are skilled at analyzing social structures, community dynamics, and the impact of technological advancements on workplace environments. You are adept at helping people understand the relationships and organizational frameworks within communities, particularly in the context of how AI influences productivity and interactions in professional settings. ->>>>>>> origin/main +You are an expert in computational linguistics and data science. You are skilled at analyzing complex text data, mapping out relationships, and understanding the structural dynamics within specialized communities. You are adept at helping people identify the relations and structure of the community of interest, specifically within the "Natural Language Processing and Information Retrieval" domain. Using your expertise, you're asked to generate a comprehensive summary of the data provided below. Given one or two entities, and a list of descriptions, all related to the same entity or group of entities. -Please concatenate all of these into a single, concise description in The primary language of the provided text is "English.". Make sure to include information collected from all the descriptions. +Please concatenate all of these into a single, concise description in The primary language of the provided text is **English**. The text includes technical terms, references to academic papers, and other content that is characteristic of English-language academic and technical writing.. Make sure to include information collected from all the descriptions. If the provided descriptions are contradictory, please resolve the contradictions and provide a single, coherent summary. -Make sure it is written in third person, and include the entity names so we the have full context. +Make sure it is written in third person, and include the entity names so we have the full context. Enrich it as much as you can with relevant information from the nearby text, this is very important. diff --git a/graphfleet/read_env.py b/graphfleet/read_env.py index b9eceb6fd..fa28fea76 100644 --- a/graphfleet/read_env.py +++ b/graphfleet/read_env.py @@ -1,4 +1,3 @@ -<<<<<<< HEAD """ Module for reading environment variables and configuration files. @@ -22,15 +21,6 @@ def load_dotenv( interpolate: bool = True, encoding: str | None = "utf-8" ): -======= -from dotenv import load_dotenv -import os -import yaml -import re - -# Load environment variables from .env file -load_dotenv() ->>>>>>> origin/main # Function to replace placeholders with environment variables def replace_placeholders(content): diff --git a/graphfleet/settings.yaml b/graphfleet/settings.yaml index 39274f93b..81a7f231f 100644 --- a/graphfleet/settings.yaml +++ b/graphfleet/settings.yaml @@ -7,11 +7,11 @@ llm: model_supports_json: true # recommended if this is available for your model. max_tokens: 4000 # request_timeout: 180.0 - api_base: https://gpt-4o-fr.openai.azure.com - api_version: "2023-05-15" + api_base: https://fleet-openai.openai.azure.com + api_version: "2024-04-01-preview" # organization: deployment_name: gpt-4o - tokens_per_minute: 150_000 # set a leaky bucket throttle + tokens_per_minute: 150000 # set a leaky bucket throttle requests_per_minute: 10_000 # set a leaky bucket throttle # max_retries: 10 max_retry_wait: 10.0 @@ -34,12 +34,12 @@ embeddings: api_key: ${GRAPHRAG_API_KEY} type: azure_openai_embedding # or openai_embedding model: text-embedding-ada-002 - api_base: https://gpt-4o-fr.openai.azure.com - api_version: "2023-05-15" + api_base: https://fleet-openai.openai.azure.com + api_version: "2024-04-01-preview" # organization: deployment_name: text-embedding-ada-002 - tokens_per_minute: 150_000 # set a leaky bucket throttle - requests_per_minute: 10_000 # set a leaky bucket throttle + tokens_per_minute: 250000 # set a leaky bucket throttle + requests_per_minute: 10000 # set a leaky bucket throttle max_retries: 10 # max_retry_wait: 10.0 sleep_on_rate_limit_recommendation: true # whether to sleep when azure suggests wait-times diff --git a/graphfleet/tools/convert_pdf_txt.py b/graphfleet/tools/convert_pdf_txt.py new file mode 100644 index 000000000..4826ba385 --- /dev/null +++ b/graphfleet/tools/convert_pdf_txt.py @@ -0,0 +1,28 @@ +import os +from pathlib import Path +from pypdf import PdfReader + +def convert_pdf_to_txt(pdf_path, txt_path): + with open(pdf_path, 'rb') as file: + pdf_reader = PdfReader(file) + text = '' + for page in pdf_reader.pages: + text += page.extract_text() + + with open(txt_path, 'w', encoding='utf-8') as txt_file: + txt_file.write(text) + +def main(): + pdf_dir = Path('graphfleet/data/pdf') + txt_dir = Path('graphfleet/data') + + # Create the output directory if it doesn't exist + txt_dir.mkdir(parents=True, exist_ok=True) + + for pdf_file in pdf_dir.glob('*.pdf'): + txt_file = txt_dir / f"{pdf_file.stem}.txt" + print(f"Converting {pdf_file} to {txt_file}") + convert_pdf_to_txt(pdf_file, txt_file) + +if __name__ == "__main__": + main() \ No newline at end of file diff --git a/graphfleet/tools/lib_path.py b/graphfleet/tools/lib_path.py new file mode 100644 index 000000000..fdfbfe971 --- /dev/null +++ b/graphfleet/tools/lib_path.py @@ -0,0 +1,12 @@ +## This script is used to get the path to the graphrag library. + + +import sys +import os + + +# Add the graphfleet directory to the Python path +sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), '..'))) + +from graphfleet.libs import graphrag +print(os.path.dirname(graphrag.__file__)) \ No newline at end of file diff --git a/notebook/Global Search Notebook.ipynb b/notebook/Global Search Notebook.ipynb new file mode 100644 index 000000000..5d1cc0269 --- /dev/null +++ b/notebook/Global Search Notebook.ipynb @@ -0,0 +1,631 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": 1, + "metadata": {}, + "outputs": [ + { + "data": { + "text/plain": [ + "'\\nCopyright (c) Microsoft Corporation.\\n'" + ] + }, + "execution_count": 1, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "# Copyright (c) 2024 Microsoft Corporation.\n", + "# Licensed under the MIT License." + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "metadata": {}, + "outputs": [], + "source": [ + "import os\n", + "\n", + "import pandas as pd\n", + "import tiktoken\n", + "\n", + "from graphrag.query.indexer_adapters import read_indexer_entities, read_indexer_reports\n", + "from graphrag.query.llm.oai.chat_openai import ChatOpenAI\n", + "from graphrag.query.llm.oai.typing import OpenaiApiType\n", + "from graphrag.query.structured_search.global_search.community_context import (\n", + " GlobalCommunityContext,\n", + ")\n", + "from graphrag.query.structured_search.global_search.search import GlobalSearch" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Global Search example\n", + "\n", + "Global search method generates answers by searching over all AI-generated community reports in a map-reduce fashion. This is a resource-intensive method, but often gives good responses for questions that require an understanding of the dataset as a whole (e.g. What are the most significant values of the herbs mentioned in this notebook?)." + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### LLM setup" + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "metadata": {}, + "outputs": [], + "source": [ + "api_key = os.environ[\"GRAPHRAG_API_KEY\"]\n", + "llm_model = os.environ[\"GRAPHRAG_LLM_MODEL\"]\n", + "embedding_model = os.environ[\"GRAPHRAG_EMBEDDING_MODEL\"]\n", + "api_base = os.environ[\"GRAPHRAG_API_BASE\"]\n", + "api_version = os.environ[\"GRAPHRAG_API_VERSION\"]\n", + "\n", + "llm = ChatOpenAI(\n", + " api_key=api_key,\n", + " api_base=api_base,\n", + " api_version=api_version,\n", + " model=llm_model,\n", + " api_type=OpenaiApiType.AzureOpenAI, # OpenaiApiType.OpenAI or OpenaiApiType.AzureOpenAI\n", + " max_retries=20,\n", + ")\n", + "\n", + "token_encoder = tiktoken.get_encoding(\"cl100k_base\")" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### Load community reports as context for global search\n", + "\n", + "- Load all community reports in the `create_final_community_reports` table from the ire-indexing engine, to be used as context data for global search.\n", + "- Load entities from the `create_final_nodes` and `create_final_entities` tables from the ire-indexing engine, to be used for calculating community weights for context ranking. Note that this is optional (if no entities are provided, we will not calculate community weights and only use the `rank` attribute in the community reports table for context ranking)" + ] + }, + { + "cell_type": "code", + "execution_count": 3, + "metadata": {}, + "outputs": [], + "source": [ + "# parquet files generated from indexing pipeline\n", + "INPUT_DIR = \"../graphfleet/output/graphindex/artifacts\"\n", + "COMMUNITY_REPORT_TABLE = \"create_final_community_reports\"\n", + "ENTITY_TABLE = \"create_final_nodes\"\n", + "ENTITY_EMBEDDING_TABLE = \"create_final_entities\"\n", + "\n", + "# community level in the Leiden community hierarchy from which we will load the community reports\n", + "# higher value means we use reports from more fine-grained communities (at the cost of higher computation cost)\n", + "COMMUNITY_LEVEL = 2" + ] + }, + { + "cell_type": "code", + "execution_count": 4, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Total report count: 102\n", + "Report count after filtering by community level 2: 83\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
communityfull_contentlevelranktitlerank_explanationsummaryfindingsfull_content_jsonid
0100# Natural Language Processing and Information ...28.5Natural Language Processing and Information Re...The rating is high due to the significant impa...The community of 'Natural Language Processing ...[{'explanation': 'Community summaries play a p...{\\n \"title\": \"Natural Language Processing a...8c20fd3b-1450-4841-a64a-18429d05d5d3
1101# Natural Language Processing and Information ...29.0Natural Language Processing and Information Re...The text is highly significant and impactful i...The community of Natural Language Processing (...[{'explanation': 'Recent advancements in trans...{\\n \"title\": \"Natural Language Processing a...cea459eb-b1cf-46cd-aabb-0ce4f3c184b7
271# Natural Language Processing and Information ...29.0Natural Language Processing and Information Re...The text is highly significant and impactful i...The community of 'Natural Language Processing ...[{'explanation': 'The 'NEWS ARTICLES' dataset ...{\\n \"title\": \"Natural Language Processing a...1458f6ef-6da1-486e-94d1-3070e4e280ff
372# Natural Language Processing and Information ...29.0Natural Language Processing and Information Re...The text is highly significant and impactful i...The community of 'Natural Language Processing ...[{'explanation': 'The 'PODCAST TRANSCRIPTS' da...{\\n \"title\": \"Natural Language Processing a...bcb2afbf-6b77-4e7c-ad5f-845d873cd10d
473# Graph RAG and Community Summarization in NLP...29.0Graph RAG and Community Summarization in NLP a...The rating is high due to the significant impa...The community revolves around the Graph RAG sy...[{'explanation': 'Graph RAG is a pivotal syste...{\\n \"title\": \"Graph RAG and Community Summa...6a14f684-f45d-4563-88f7-18a2cac9b5cb
\n", + "
" + ], + "text/plain": [ + " community full_content level rank \\\n", + "0 100 # Natural Language Processing and Information ... 2 8.5 \n", + "1 101 # Natural Language Processing and Information ... 2 9.0 \n", + "2 71 # Natural Language Processing and Information ... 2 9.0 \n", + "3 72 # Natural Language Processing and Information ... 2 9.0 \n", + "4 73 # Graph RAG and Community Summarization in NLP... 2 9.0 \n", + "\n", + " title \\\n", + "0 Natural Language Processing and Information Re... \n", + "1 Natural Language Processing and Information Re... \n", + "2 Natural Language Processing and Information Re... \n", + "3 Natural Language Processing and Information Re... \n", + "4 Graph RAG and Community Summarization in NLP a... \n", + "\n", + " rank_explanation \\\n", + "0 The rating is high due to the significant impa... \n", + "1 The text is highly significant and impactful i... \n", + "2 The text is highly significant and impactful i... \n", + "3 The text is highly significant and impactful i... \n", + "4 The rating is high due to the significant impa... \n", + "\n", + " summary \\\n", + "0 The community of 'Natural Language Processing ... \n", + "1 The community of Natural Language Processing (... \n", + "2 The community of 'Natural Language Processing ... \n", + "3 The community of 'Natural Language Processing ... \n", + "4 The community revolves around the Graph RAG sy... \n", + "\n", + " findings \\\n", + "0 [{'explanation': 'Community summaries play a p... \n", + "1 [{'explanation': 'Recent advancements in trans... \n", + "2 [{'explanation': 'The 'NEWS ARTICLES' dataset ... \n", + "3 [{'explanation': 'The 'PODCAST TRANSCRIPTS' da... \n", + "4 [{'explanation': 'Graph RAG is a pivotal syste... \n", + "\n", + " full_content_json \\\n", + "0 {\\n \"title\": \"Natural Language Processing a... \n", + "1 {\\n \"title\": \"Natural Language Processing a... \n", + "2 {\\n \"title\": \"Natural Language Processing a... \n", + "3 {\\n \"title\": \"Natural Language Processing a... \n", + "4 {\\n \"title\": \"Graph RAG and Community Summa... \n", + "\n", + " id \n", + "0 8c20fd3b-1450-4841-a64a-18429d05d5d3 \n", + "1 cea459eb-b1cf-46cd-aabb-0ce4f3c184b7 \n", + "2 1458f6ef-6da1-486e-94d1-3070e4e280ff \n", + "3 bcb2afbf-6b77-4e7c-ad5f-845d873cd10d \n", + "4 6a14f684-f45d-4563-88f7-18a2cac9b5cb " + ] + }, + "execution_count": 4, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "entity_df = pd.read_parquet(f\"{INPUT_DIR}/{ENTITY_TABLE}.parquet\")\n", + "report_df = pd.read_parquet(f\"{INPUT_DIR}/{COMMUNITY_REPORT_TABLE}.parquet\")\n", + "entity_embedding_df = pd.read_parquet(f\"{INPUT_DIR}/{ENTITY_EMBEDDING_TABLE}.parquet\")\n", + "\n", + "reports = read_indexer_reports(report_df, entity_df, COMMUNITY_LEVEL)\n", + "entities = read_indexer_entities(entity_df, entity_embedding_df, COMMUNITY_LEVEL)\n", + "print(f\"Total report count: {len(report_df)}\")\n", + "print(\n", + " f\"Report count after filtering by community level {COMMUNITY_LEVEL}: {len(reports)}\"\n", + ")\n", + "report_df.head()" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "#### Build global context based on community reports" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "metadata": {}, + "outputs": [], + "source": [ + "context_builder = GlobalCommunityContext(\n", + " community_reports=reports,\n", + " entities=entities, # default to None if you don't want to use community weights for ranking\n", + " token_encoder=token_encoder,\n", + ")" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "#### Perform global search" + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "metadata": {}, + "outputs": [], + "source": [ + "context_builder_params = {\n", + " \"use_community_summary\": False, # False means using full community reports. True means using community short summaries.\n", + " \"shuffle_data\": True,\n", + " \"include_community_rank\": True,\n", + " \"min_community_rank\": 0,\n", + " \"community_rank_name\": \"rank\",\n", + " \"include_community_weight\": True,\n", + " \"community_weight_name\": \"occurrence weight\",\n", + " \"normalize_community_weight\": True,\n", + " \"max_tokens\": 12_000, # change this based on the token limit you have on your model (if you are using a model with 8k limit, a good setting could be 5000)\n", + " \"context_name\": \"Reports\",\n", + "}\n", + "\n", + "map_llm_params = {\n", + " \"max_tokens\": 1000,\n", + " \"temperature\": 0.0,\n", + " \"response_format\": {\"type\": \"json_object\"},\n", + "}\n", + "\n", + "reduce_llm_params = {\n", + " \"max_tokens\": 2000, # change this based on the token limit you have on your model (if you are using a model with 8k limit, a good setting could be 1000-1500)\n", + " \"temperature\": 0.0,\n", + "}" + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "metadata": {}, + "outputs": [], + "source": [ + "search_engine = GlobalSearch(\n", + " llm=llm,\n", + " context_builder=context_builder,\n", + " token_encoder=token_encoder,\n", + " max_data_tokens=12_000, # change this based on the token limit you have on your model (if you are using a model with 8k limit, a good setting could be 5000)\n", + " map_llm_params=map_llm_params,\n", + " reduce_llm_params=reduce_llm_params,\n", + " allow_general_knowledge=False, # set this to True will add instruction to encourage the LLM to incorporate general knowledge in the response, which may increase hallucinations, but could be useful in some use cases.\n", + " json_mode=True, # set this to False if your LLM model does not support JSON mode.\n", + " context_builder_params=context_builder_params,\n", + " concurrent_coroutines=32,\n", + " response_type=\"multiple paragraphs\", # free form text describing the response type and format, can be anything, e.g. prioritized list, single paragraph, multiple paragraphs, multiple-page report\n", + ")" + ] + }, + { + "cell_type": "code", + "execution_count": 9, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "### Purpose of GraphRAG\n", + "\n", + "GraphRAG (Graph-based Retrieval-Augmented Generation) is designed to enhance the capabilities of retrieval-augmented generation tasks by integrating retrieval mechanisms with generation models. This approach allows for more comprehensive and contextually relevant summaries by retrieving pertinent information from large datasets before generating the final summary [Data: Reports (57, 22, 70, 68, 71)].\n", + "\n", + "### Key Features and Benefits\n", + "\n", + "1. **Graph-Based Text Indexing**:\n", + " GraphRAG leverages graph-based text indexing to partition data, facilitating global summarization. This method enhances the efficiency and effectiveness of text data processing and summarization by utilizing the natural modularity of graphs [Data: Reports (26)].\n", + "\n", + "2. **Comprehensive and Diverse Responses**:\n", + " By employing graph-based indexing and summarization techniques, GraphRAG ensures that the generated responses are both comprehensive and diverse, outperforming traditional summarization methods [Data: Reports (17, 22, 70, 71)].\n", + "\n", + "3. **Integration with Large Language Models (LLMs)**:\n", + " GraphRAG utilizes LLMs to construct a graph-based text index, enabling the generation of summaries and the answering of queries. This integration enhances the system's ability to process and synthesize complex text data, making it a powerful tool for information retrieval and natural language processing tasks [Data: Reports (77, 87)].\n", + "\n", + "4. **Hierarchical Community Summaries**:\n", + " The system uses hierarchical community structures to answer user questions over entire datasets. This approach is designed for situations where answers are contained locally within regions of text whose retrieval provides sufficient grounding for the generation task [Data: Reports (65, 73)].\n", + "\n", + "5. **Efficiency and Scalability**:\n", + " GraphRAG demonstrates significant scalability and efficiency advantages compared to traditional source text summarization. The system requires fewer context tokens for community summaries, particularly at the root level (C0), making it a resource-efficient approach [Data: Reports (73, 22)].\n", + "\n", + "### Applications\n", + "\n", + "GraphRAG is particularly useful for answering user questions over large datasets by leveraging graph-based indexing and summarization techniques. This method ensures that the generated responses are both comprehensive and diverse, outperforming traditional summarization methods [Data: Reports (17, 22, 70, 71)]. It is also employed to enhance the capabilities of large language models (LLMs) by enabling them to retrieve pertinent information from external knowledge sources, thereby generating and assessing text more effectively [Data: Reports (26)].\n", + "\n", + "### Conclusion\n", + "\n", + "In summary, GraphRAG is a pivotal system in the NLP and IR community, utilizing graph-based text indexing and hierarchical community summaries to enhance the retrieval and generation of information. Its ability to provide comprehensive, diverse, and contextually relevant summaries makes it a valuable tool for handling large datasets and answering complex user queries efficiently [Data: Reports (57, 22, 70, 68, 71, +more)].\n" + ] + } + ], + "source": [ + "result = await search_engine.asearch(\n", + " \"What is the purpose of GraphRAG?\"\n", + ")\n", + "\n", + "print(result.response)" + ] + }, + { + "cell_type": "code", + "execution_count": 10, + "metadata": {}, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
idtitleoccurrence weightcontentrank
023Natural Language Processing and Information Re...0.444444# Natural Language Processing and Information ...9.0
173Graph RAG and Community Summarization in NLP a...0.333333# Graph RAG and Community Summarization in NLP...9.0
278Few-Shot Examples and Their Role in Enhancing ...0.166667# Few-Shot Examples and Their Role in Enhancin...9.0
335Natural Language Processing and Information Re...0.111111# Natural Language Processing and Information ...9.0
480Natural Language Processing and Information Re...0.111111# Natural Language Processing and Information ...9.0
..................
7882Retrieval-Generation Synergy in Large Language...0.055556# Retrieval-Generation Synergy in Large Langua...9.0
7940LLAMA and Its Impact on Natural Language Proce...0.055556# LLAMA and Its Impact on Natural Language Pro...9.0
8030Natural Language Processing and Information Re...0.055556# Natural Language Processing and Information ...9.0
81101Natural Language Processing and Information Re...0.055556# Natural Language Processing and Information ...9.0
821Natural Language Processing and Information Re...0.055556# Natural Language Processing and Information ...9.0
\n", + "

83 rows Ɨ 5 columns

\n", + "
" + ], + "text/plain": [ + " id title occurrence weight \\\n", + "0 23 Natural Language Processing and Information Re... 0.444444 \n", + "1 73 Graph RAG and Community Summarization in NLP a... 0.333333 \n", + "2 78 Few-Shot Examples and Their Role in Enhancing ... 0.166667 \n", + "3 35 Natural Language Processing and Information Re... 0.111111 \n", + "4 80 Natural Language Processing and Information Re... 0.111111 \n", + ".. ... ... ... \n", + "78 82 Retrieval-Generation Synergy in Large Language... 0.055556 \n", + "79 40 LLAMA and Its Impact on Natural Language Proce... 0.055556 \n", + "80 30 Natural Language Processing and Information Re... 0.055556 \n", + "81 101 Natural Language Processing and Information Re... 0.055556 \n", + "82 1 Natural Language Processing and Information Re... 0.055556 \n", + "\n", + " content rank \n", + "0 # Natural Language Processing and Information ... 9.0 \n", + "1 # Graph RAG and Community Summarization in NLP... 9.0 \n", + "2 # Few-Shot Examples and Their Role in Enhancin... 9.0 \n", + "3 # Natural Language Processing and Information ... 9.0 \n", + "4 # Natural Language Processing and Information ... 9.0 \n", + ".. ... ... \n", + "78 # Retrieval-Generation Synergy in Large Langua... 9.0 \n", + "79 # LLAMA and Its Impact on Natural Language Pro... 9.0 \n", + "80 # Natural Language Processing and Information ... 9.0 \n", + "81 # Natural Language Processing and Information ... 9.0 \n", + "82 # Natural Language Processing and Information ... 9.0 \n", + "\n", + "[83 rows x 5 columns]" + ] + }, + "execution_count": 10, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "# inspect the data used to build the context for the LLM responses\n", + "result.context_data[\"reports\"]" + ] + }, + { + "cell_type": "code", + "execution_count": 11, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "LLM calls: 7. LLM tokens: 75225\n" + ] + } + ], + "source": [ + "# inspect number of LLM calls and tokens\n", + "print(f\"LLM calls: {result.llm_calls}. LLM tokens: {result.prompt_tokens}\")" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.11.2" + } + }, + "nbformat": 4, + "nbformat_minor": 2 +} diff --git a/notebook/Local Search Notebook.ipynb b/notebook/Local Search Notebook.ipynb new file mode 100644 index 000000000..4e2a8d81e --- /dev/null +++ b/notebook/Local Search Notebook.ipynb @@ -0,0 +1,1503 @@ +{ + "cells": [ + { + "cell_type": "code", + "execution_count": 1, + "metadata": {}, + "outputs": [], + "source": [ + "# Copyright (c) 2024 Microsoft Corporation.\n", + "# Licensed under the MIT License." + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "metadata": {}, + "outputs": [], + "source": [ + "import os\n", + "\n", + "import pandas as pd\n", + "import tiktoken\n", + "\n", + "from graphrag.query.context_builder.entity_extraction import EntityVectorStoreKey\n", + "from graphrag.query.indexer_adapters import (\n", + " read_indexer_covariates,\n", + " read_indexer_entities,\n", + " read_indexer_relationships,\n", + " read_indexer_reports,\n", + " read_indexer_text_units,\n", + ")\n", + "from graphrag.query.input.loaders.dfs import (\n", + " store_entity_semantic_embeddings,\n", + ")\n", + "from graphrag.query.llm.oai.chat_openai import ChatOpenAI\n", + "from graphrag.query.llm.oai.embedding import OpenAIEmbedding\n", + "from graphrag.query.llm.oai.typing import OpenaiApiType\n", + "from graphrag.query.question_gen.local_gen import LocalQuestionGen\n", + "from graphrag.query.structured_search.local_search.mixed_context import (\n", + " LocalSearchMixedContext,\n", + ")\n", + "from graphrag.query.structured_search.local_search.search import LocalSearch\n", + "from graphrag.vector_stores.lancedb import LanceDBVectorStore" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Local Search Example\n", + "\n", + "Local search method generates answers by combining relevant data from the AI-extracted knowledge-graph with text chunks of the raw documents. This method is suitable for questions that require an understanding of specific entities mentioned in the documents (e.g. What are the healing properties of chamomile?)." + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### Load text units and graph data tables as context for local search\n", + "\n", + "- In this test we first load indexing outputs from parquet files to dataframes, then convert these dataframes into collections of data objects aligning with the knowledge model." + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### Load tables to dataframes" + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "metadata": {}, + "outputs": [], + "source": [ + "INPUT_DIR = \"../graphfleet/output/graphindex/artifacts\"\n", + "LANCEDB_URI = f\"{INPUT_DIR}/lancedb\"\n", + "\n", + "COMMUNITY_REPORT_TABLE = \"create_final_community_reports\"\n", + "ENTITY_TABLE = \"create_final_nodes\"\n", + "ENTITY_EMBEDDING_TABLE = \"create_final_entities\"\n", + "RELATIONSHIP_TABLE = \"create_final_relationships\"\n", + "COVARIATE_TABLE = \"create_final_covariates\"\n", + "TEXT_UNIT_TABLE = \"create_final_text_units\"\n", + "COMMUNITY_LEVEL = 2" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "#### Read entities" + ] + }, + { + "cell_type": "code", + "execution_count": 3, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Entity count: 2208\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
leveltitletypedescriptionsource_iddegreehuman_readable_ididgraph_embeddingcommunitysizeentity_typetop_level_node_idxy
00DARREN EDGEPERSONDarren Edge is an author of the paper \"From Lo...e8d83e6e7a7c0f57b218cef24976b74580b45241d70f0e43fca764df95b2b81f77NoneNoneNaNNoneb45241d70f0e43fca764df95b2b81f77NaNNaN
10HA TRINHPERSONHa Trinh is an author of the paper \"From Local...e8d83e6e7a7c0f57b218cef24976b745814119fd06010c494caa07f439b333f4c5NoneNoneNaNNone4119fd06010c494caa07f439b333f4c5NaNNaN
20NEWMAN CHENGPERSONNewman Cheng is an author of the paper \"From L...e8d83e6e7a7c0f57b218cef24976b74582d3835bf3dda84ead99deadbeac5d0d7dNoneNoneNaNNoned3835bf3dda84ead99deadbeac5d0d7dNaNNaN
30JOSHUA BRADLEYPERSONJoshua Bradley is an author of the paper \"From...e8d83e6e7a7c0f57b218cef24976b74583077d2820ae1845bcbb1803379a3d1eaeNoneNoneNaNNone077d2820ae1845bcbb1803379a3d1eaeNaNNaN
40ALEX CHAOPERSONAlex Chao is an author of the paper \"From Loca...e8d83e6e7a7c0f57b218cef24976b745843671ea0dd4e84c1a9b02c5ab2c8f4bacNoneNoneNaNNone3671ea0dd4e84c1a9b02c5ab2c8f4bacNaNNaN
\n", + "
" + ], + "text/plain": [ + " level title type \\\n", + "0 0 DARREN EDGE PERSON \n", + "1 0 HA TRINH PERSON \n", + "2 0 NEWMAN CHENG PERSON \n", + "3 0 JOSHUA BRADLEY PERSON \n", + "4 0 ALEX CHAO PERSON \n", + "\n", + " description \\\n", + "0 Darren Edge is an author of the paper \"From Lo... \n", + "1 Ha Trinh is an author of the paper \"From Local... \n", + "2 Newman Cheng is an author of the paper \"From L... \n", + "3 Joshua Bradley is an author of the paper \"From... \n", + "4 Alex Chao is an author of the paper \"From Loca... \n", + "\n", + " source_id degree human_readable_id \\\n", + "0 e8d83e6e7a7c0f57b218cef24976b745 8 0 \n", + "1 e8d83e6e7a7c0f57b218cef24976b745 8 1 \n", + "2 e8d83e6e7a7c0f57b218cef24976b745 8 2 \n", + "3 e8d83e6e7a7c0f57b218cef24976b745 8 3 \n", + "4 e8d83e6e7a7c0f57b218cef24976b745 8 4 \n", + "\n", + " id graph_embedding community size \\\n", + "0 b45241d70f0e43fca764df95b2b81f77 None None NaN \n", + "1 4119fd06010c494caa07f439b333f4c5 None None NaN \n", + "2 d3835bf3dda84ead99deadbeac5d0d7d None None NaN \n", + "3 077d2820ae1845bcbb1803379a3d1eae None None NaN \n", + "4 3671ea0dd4e84c1a9b02c5ab2c8f4bac None None NaN \n", + "\n", + " entity_type top_level_node_id x y \n", + "0 None b45241d70f0e43fca764df95b2b81f77 NaN NaN \n", + "1 None 4119fd06010c494caa07f439b333f4c5 NaN NaN \n", + "2 None d3835bf3dda84ead99deadbeac5d0d7d NaN NaN \n", + "3 None 077d2820ae1845bcbb1803379a3d1eae NaN NaN \n", + "4 None 3671ea0dd4e84c1a9b02c5ab2c8f4bac NaN NaN " + ] + }, + "execution_count": 3, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "# read nodes table to get community and degree data\n", + "entity_df = pd.read_parquet(f\"{INPUT_DIR}/{ENTITY_TABLE}.parquet\")\n", + "entity_embedding_df = pd.read_parquet(f\"{INPUT_DIR}/{ENTITY_EMBEDDING_TABLE}.parquet\")\n", + "\n", + "entities = read_indexer_entities(entity_df, entity_embedding_df, COMMUNITY_LEVEL)\n", + "\n", + "# load description embeddings to an in-memory lancedb vectorstore\n", + "# to connect to a remote db, specify url and port values.\n", + "description_embedding_store = LanceDBVectorStore(\n", + " collection_name=\"entity_description_embeddings\",\n", + ")\n", + "description_embedding_store.connect(db_uri=LANCEDB_URI)\n", + "entity_description_embeddings = store_entity_semantic_embeddings(\n", + " entities=entities, vectorstore=description_embedding_store\n", + ")\n", + "\n", + "print(f\"Entity count: {len(entity_df)}\")\n", + "entity_df.head()" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "#### Read relationships" + ] + }, + { + "cell_type": "code", + "execution_count": 4, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Relationship count: 1179\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
sourcetargetweightdescriptiontext_unit_idsidhuman_readable_idsource_degreetarget_degreerank
0DARREN EDGEHA TRINH1.0Darren Edge and Ha Trinh co-authored the paper...[e8d83e6e7a7c0f57b218cef24976b745]28b7457ca5dc4a38a488946a3f8e207e08816
1DARREN EDGENEWMAN CHENG1.0Darren Edge and Newman Cheng co-authored the p...[e8d83e6e7a7c0f57b218cef24976b745]8029a14d15404e6db95ddf5e2bf9fc1518816
2DARREN EDGEJOSHUA BRADLEY1.0Darren Edge and Joshua Bradley co-authored the...[e8d83e6e7a7c0f57b218cef24976b745]389314ca89d445888c8d4985864dd73328816
3DARREN EDGEALEX CHAO1.0Darren Edge and Alex Chao co-authored the pape...[e8d83e6e7a7c0f57b218cef24976b745]87fe1462b9064d5692641ab48e82630138816
4DARREN EDGEAPURVA MODY1.0Darren Edge and Apurva Mody co-authored the pa...[e8d83e6e7a7c0f57b218cef24976b745]a55175ac57014df696ca09d0def9604b48816
\n", + "
" + ], + "text/plain": [ + " source target weight \\\n", + "0 DARREN EDGE HA TRINH 1.0 \n", + "1 DARREN EDGE NEWMAN CHENG 1.0 \n", + "2 DARREN EDGE JOSHUA BRADLEY 1.0 \n", + "3 DARREN EDGE ALEX CHAO 1.0 \n", + "4 DARREN EDGE APURVA MODY 1.0 \n", + "\n", + " description \\\n", + "0 Darren Edge and Ha Trinh co-authored the paper... \n", + "1 Darren Edge and Newman Cheng co-authored the p... \n", + "2 Darren Edge and Joshua Bradley co-authored the... \n", + "3 Darren Edge and Alex Chao co-authored the pape... \n", + "4 Darren Edge and Apurva Mody co-authored the pa... \n", + "\n", + " text_unit_ids id \\\n", + "0 [e8d83e6e7a7c0f57b218cef24976b745] 28b7457ca5dc4a38a488946a3f8e207e \n", + "1 [e8d83e6e7a7c0f57b218cef24976b745] 8029a14d15404e6db95ddf5e2bf9fc15 \n", + "2 [e8d83e6e7a7c0f57b218cef24976b745] 389314ca89d445888c8d4985864dd733 \n", + "3 [e8d83e6e7a7c0f57b218cef24976b745] 87fe1462b9064d5692641ab48e826301 \n", + "4 [e8d83e6e7a7c0f57b218cef24976b745] a55175ac57014df696ca09d0def9604b \n", + "\n", + " human_readable_id source_degree target_degree rank \n", + "0 0 8 8 16 \n", + "1 1 8 8 16 \n", + "2 2 8 8 16 \n", + "3 3 8 8 16 \n", + "4 4 8 8 16 " + ] + }, + "execution_count": 4, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "relationship_df = pd.read_parquet(f\"{INPUT_DIR}/{RELATIONSHIP_TABLE}.parquet\")\n", + "relationships = read_indexer_relationships(relationship_df)\n", + "\n", + "print(f\"Relationship count: {len(relationship_df)}\")\n", + "relationship_df.head()" + ] + }, + { + "cell_type": "code", + "execution_count": 5, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Claim records: 201\n" + ] + } + ], + "source": [ + "# NOTE: covariates are turned off by default, because they generally need prompt tuning to be valuable\n", + "# Please see the GRAPHRAG_CLAIM_* settings\n", + "covariate_df = pd.read_parquet(f\"{INPUT_DIR}/{COVARIATE_TABLE}.parquet\")\n", + "\n", + "claims = read_indexer_covariates(covariate_df)\n", + "\n", + "print(f\"Claim records: {len(claims)}\")\n", + "covariates = {\"claims\": claims}" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "#### Read community reports" + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Report records: 102\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
communityfull_contentlevelranktitlerank_explanationsummaryfindingsfull_content_jsonid
0100# Natural Language Processing and Information ...28.5Natural Language Processing and Information Re...The rating is high due to the significant impa...The community of 'Natural Language Processing ...[{'explanation': 'Community summaries play a p...{\\n \"title\": \"Natural Language Processing a...8c20fd3b-1450-4841-a64a-18429d05d5d3
1101# Natural Language Processing and Information ...29.0Natural Language Processing and Information Re...The text is highly significant and impactful i...The community of Natural Language Processing (...[{'explanation': 'Recent advancements in trans...{\\n \"title\": \"Natural Language Processing a...cea459eb-b1cf-46cd-aabb-0ce4f3c184b7
271# Natural Language Processing and Information ...29.0Natural Language Processing and Information Re...The text is highly significant and impactful i...The community of 'Natural Language Processing ...[{'explanation': 'The 'NEWS ARTICLES' dataset ...{\\n \"title\": \"Natural Language Processing a...1458f6ef-6da1-486e-94d1-3070e4e280ff
372# Natural Language Processing and Information ...29.0Natural Language Processing and Information Re...The text is highly significant and impactful i...The community of 'Natural Language Processing ...[{'explanation': 'The 'PODCAST TRANSCRIPTS' da...{\\n \"title\": \"Natural Language Processing a...bcb2afbf-6b77-4e7c-ad5f-845d873cd10d
473# Graph RAG and Community Summarization in NLP...29.0Graph RAG and Community Summarization in NLP a...The rating is high due to the significant impa...The community revolves around the Graph RAG sy...[{'explanation': 'Graph RAG is a pivotal syste...{\\n \"title\": \"Graph RAG and Community Summa...6a14f684-f45d-4563-88f7-18a2cac9b5cb
\n", + "
" + ], + "text/plain": [ + " community full_content level rank \\\n", + "0 100 # Natural Language Processing and Information ... 2 8.5 \n", + "1 101 # Natural Language Processing and Information ... 2 9.0 \n", + "2 71 # Natural Language Processing and Information ... 2 9.0 \n", + "3 72 # Natural Language Processing and Information ... 2 9.0 \n", + "4 73 # Graph RAG and Community Summarization in NLP... 2 9.0 \n", + "\n", + " title \\\n", + "0 Natural Language Processing and Information Re... \n", + "1 Natural Language Processing and Information Re... \n", + "2 Natural Language Processing and Information Re... \n", + "3 Natural Language Processing and Information Re... \n", + "4 Graph RAG and Community Summarization in NLP a... \n", + "\n", + " rank_explanation \\\n", + "0 The rating is high due to the significant impa... \n", + "1 The text is highly significant and impactful i... \n", + "2 The text is highly significant and impactful i... \n", + "3 The text is highly significant and impactful i... \n", + "4 The rating is high due to the significant impa... \n", + "\n", + " summary \\\n", + "0 The community of 'Natural Language Processing ... \n", + "1 The community of Natural Language Processing (... \n", + "2 The community of 'Natural Language Processing ... \n", + "3 The community of 'Natural Language Processing ... \n", + "4 The community revolves around the Graph RAG sy... \n", + "\n", + " findings \\\n", + "0 [{'explanation': 'Community summaries play a p... \n", + "1 [{'explanation': 'Recent advancements in trans... \n", + "2 [{'explanation': 'The 'NEWS ARTICLES' dataset ... \n", + "3 [{'explanation': 'The 'PODCAST TRANSCRIPTS' da... \n", + "4 [{'explanation': 'Graph RAG is a pivotal syste... \n", + "\n", + " full_content_json \\\n", + "0 {\\n \"title\": \"Natural Language Processing a... \n", + "1 {\\n \"title\": \"Natural Language Processing a... \n", + "2 {\\n \"title\": \"Natural Language Processing a... \n", + "3 {\\n \"title\": \"Natural Language Processing a... \n", + "4 {\\n \"title\": \"Graph RAG and Community Summa... \n", + "\n", + " id \n", + "0 8c20fd3b-1450-4841-a64a-18429d05d5d3 \n", + "1 cea459eb-b1cf-46cd-aabb-0ce4f3c184b7 \n", + "2 1458f6ef-6da1-486e-94d1-3070e4e280ff \n", + "3 bcb2afbf-6b77-4e7c-ad5f-845d873cd10d \n", + "4 6a14f684-f45d-4563-88f7-18a2cac9b5cb " + ] + }, + "execution_count": 6, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "report_df = pd.read_parquet(f\"{INPUT_DIR}/{COMMUNITY_REPORT_TABLE}.parquet\")\n", + "reports = read_indexer_reports(report_df, entity_df, COMMUNITY_LEVEL)\n", + "\n", + "print(f\"Report records: {len(report_df)}\")\n", + "report_df.head()" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "#### Read text units" + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Text unit records: 30\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
idtextn_tokensdocument_idsentity_idsrelationship_idscovariate_ids
0e8d83e6e7a7c0f57b218cef24976b745From Local to Global: A Graph RAG Approach to\\...512[0668cddc5f873265ba50da5a0a06edad][b45241d70f0e43fca764df95b2b81f77, 4119fd06010...[28b7457ca5dc4a38a488946a3f8e207e, 8029a14d154...[d267bd27-b88d-41e7-a314-c9a8e873da73, 5a6063e...
1f0306814bf64f5c9e79603fc6a52f4ea.\\n1 Introduction\\nHuman endeavors across a r...512[0668cddc5f873265ba50da5a0a06edad][d91a266f766b4737a06b0fda588ba40b, 1943f245ee4...[a7c2a64e06374091adce74adb36801ab, 107568a67ca...[a7270549-cf82-4e3e-a640-f740c5576df7, 2e4e078...
2fb3c48579608fa28be585ceb6cd2f0feet al., 2006b) by asking questions of a globa...512[0668cddc5f873265ba50da5a0a06edad][c9632a35146940c2a86167c7726d35e9, 3d0dcbc8971...[f770bc07cecf4aba8fe2d2c33fdc5542, 7cea9903153...[b41beeef-ef1a-4e75-aa5b-67120c0d2fbf]
321e52bc06a82796b1f4bcd73edda1f2aa new RAG approach specifically targeting glo...512[0668cddc5f873265ba50da5a0a06edad][de988724cfdf45cebfba3b13c43ceede, 96aad7cb4b7...[192a6d23595045f38b0d46a3d8e52fd6, a6ae1d99330...[701f3b3d-d6e1-47b6-99da-a52d61caac96]
4bc9e2c9e369c4108cf4f6dd5f60960f4intermediate- and low-level community summari...512[0668cddc5f873265ba50da5a0a06edad][96aad7cb4b7d40e9b7e13b94a67af206, c9632a35146...[5174cdabb6024de0975762d3a80b059f, e379fba9011...[da077ccf-d904-45e5-9948-00fd86435b7d]
\n", + "
" + ], + "text/plain": [ + " id \\\n", + "0 e8d83e6e7a7c0f57b218cef24976b745 \n", + "1 f0306814bf64f5c9e79603fc6a52f4ea \n", + "2 fb3c48579608fa28be585ceb6cd2f0fe \n", + "3 21e52bc06a82796b1f4bcd73edda1f2a \n", + "4 bc9e2c9e369c4108cf4f6dd5f60960f4 \n", + "\n", + " text n_tokens \\\n", + "0 From Local to Global: A Graph RAG Approach to\\... 512 \n", + "1 .\\n1 Introduction\\nHuman endeavors across a r... 512 \n", + "2 et al., 2006b) by asking questions of a globa... 512 \n", + "3 a new RAG approach specifically targeting glo... 512 \n", + "4 intermediate- and low-level community summari... 512 \n", + "\n", + " document_ids \\\n", + "0 [0668cddc5f873265ba50da5a0a06edad] \n", + "1 [0668cddc5f873265ba50da5a0a06edad] \n", + "2 [0668cddc5f873265ba50da5a0a06edad] \n", + "3 [0668cddc5f873265ba50da5a0a06edad] \n", + "4 [0668cddc5f873265ba50da5a0a06edad] \n", + "\n", + " entity_ids \\\n", + "0 [b45241d70f0e43fca764df95b2b81f77, 4119fd06010... \n", + "1 [d91a266f766b4737a06b0fda588ba40b, 1943f245ee4... \n", + "2 [c9632a35146940c2a86167c7726d35e9, 3d0dcbc8971... \n", + "3 [de988724cfdf45cebfba3b13c43ceede, 96aad7cb4b7... \n", + "4 [96aad7cb4b7d40e9b7e13b94a67af206, c9632a35146... \n", + "\n", + " relationship_ids \\\n", + "0 [28b7457ca5dc4a38a488946a3f8e207e, 8029a14d154... \n", + "1 [a7c2a64e06374091adce74adb36801ab, 107568a67ca... \n", + "2 [f770bc07cecf4aba8fe2d2c33fdc5542, 7cea9903153... \n", + "3 [192a6d23595045f38b0d46a3d8e52fd6, a6ae1d99330... \n", + "4 [5174cdabb6024de0975762d3a80b059f, e379fba9011... \n", + "\n", + " covariate_ids \n", + "0 [d267bd27-b88d-41e7-a314-c9a8e873da73, 5a6063e... \n", + "1 [a7270549-cf82-4e3e-a640-f740c5576df7, 2e4e078... \n", + "2 [b41beeef-ef1a-4e75-aa5b-67120c0d2fbf] \n", + "3 [701f3b3d-d6e1-47b6-99da-a52d61caac96] \n", + "4 [da077ccf-d904-45e5-9948-00fd86435b7d] " + ] + }, + "execution_count": 7, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "text_unit_df = pd.read_parquet(f\"{INPUT_DIR}/{TEXT_UNIT_TABLE}.parquet\")\n", + "text_units = read_indexer_text_units(text_unit_df)\n", + "\n", + "print(f\"Text unit records: {len(text_unit_df)}\")\n", + "text_unit_df.head()" + ] + }, + { + "cell_type": "code", + "execution_count": 8, + "metadata": {}, + "outputs": [], + "source": [ + "api_key = os.environ[\"GRAPHRAG_API_KEY\"]\n", + "llm_model = os.environ[\"GRAPHRAG_LLM_MODEL\"]\n", + "embedding_model = os.environ[\"GRAPHRAG_EMBEDDING_MODEL\"]\n", + "api_base = os.environ[\"GRAPHRAG_API_BASE\"]\n", + "api_version = os.environ[\"GRAPHRAG_API_VERSION\"]\n", + "\n", + "\n", + "llm = ChatOpenAI(\n", + " api_key=api_key,\n", + " api_base=api_base,\n", + " api_version=api_version,\n", + " model=llm_model,\n", + " api_type=OpenaiApiType.AzureOpenAI, # OpenaiApiType.OpenAI or OpenaiApiType.AzureOpenAI\n", + " max_retries=20,\n", + ")\n", + "\n", + "token_encoder = tiktoken.get_encoding(\"cl100k_base\")\n", + "\n", + "text_embedder = OpenAIEmbedding(\n", + " api_key=api_key,\n", + " api_base=api_base,\n", + " api_version=api_version,\n", + " api_type=OpenaiApiType.AzureOpenAI,\n", + " model=embedding_model,\n", + " deployment_name=embedding_model,\n", + " max_retries=20,\n", + ")" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### Create local search context builder" + ] + }, + { + "cell_type": "code", + "execution_count": 9, + "metadata": {}, + "outputs": [], + "source": [ + "context_builder = LocalSearchMixedContext(\n", + " community_reports=reports,\n", + " text_units=text_units,\n", + " entities=entities,\n", + " relationships=relationships,\n", + " # if you did not run covariates during indexing, set this to None\n", + " covariates=covariates,\n", + " entity_text_embeddings=description_embedding_store,\n", + " embedding_vectorstore_key=EntityVectorStoreKey.ID, # if the vectorstore uses entity title as ids, set this to EntityVectorStoreKey.TITLE\n", + " text_embedder=text_embedder,\n", + " token_encoder=token_encoder,\n", + ")" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### Create local search engine" + ] + }, + { + "cell_type": "code", + "execution_count": 11, + "metadata": {}, + "outputs": [], + "source": [ + "# text_unit_prop: proportion of context window dedicated to related text units\n", + "# community_prop: proportion of context window dedicated to community reports.\n", + "# The remaining proportion is dedicated to entities and relationships. Sum of text_unit_prop and community_prop should be <= 1\n", + "# conversation_history_max_turns: maximum number of turns to include in the conversation history.\n", + "# conversation_history_user_turns_only: if True, only include user queries in the conversation history.\n", + "# top_k_mapped_entities: number of related entities to retrieve from the entity description embedding store.\n", + "# top_k_relationships: control the number of out-of-network relationships to pull into the context window.\n", + "# include_entity_rank: if True, include the entity rank in the entity table in the context window. Default entity rank = node degree.\n", + "# include_relationship_weight: if True, include the relationship weight in the context window.\n", + "# include_community_rank: if True, include the community rank in the context window.\n", + "# return_candidate_context: if True, return a set of dataframes containing all candidate entity/relationship/covariate records that\n", + "# could be relevant. Note that not all of these records will be included in the context window. The \"in_context\" column in these\n", + "# dataframes indicates whether the record is included in the context window.\n", + "# max_tokens: maximum number of tokens to use for the context window.\n", + "\n", + "\n", + "local_context_params = {\n", + " \"text_unit_prop\": 0.5,\n", + " \"community_prop\": 0.1,\n", + " \"conversation_history_max_turns\": 5,\n", + " \"conversation_history_user_turns_only\": True,\n", + " \"top_k_mapped_entities\": 10,\n", + " \"top_k_relationships\": 10,\n", + " \"include_entity_rank\": True,\n", + " \"include_relationship_weight\": True,\n", + " \"include_community_rank\": False,\n", + " \"return_candidate_context\": False,\n", + " \"embedding_vectorstore_key\": EntityVectorStoreKey.ID, # set this to EntityVectorStoreKey.TITLE if the vectorstore uses entity title as ids\n", + " \"max_tokens\": 12_000, # change this based on the token limit you have on your model (if you are using a model with 8k limit, a good setting could be 5000)\n", + "}\n", + "\n", + "llm_params = {\n", + " \"max_tokens\": 2_000, # change this based on the token limit you have on your model (if you are using a model with 8k limit, a good setting could be 1000=1500)\n", + " \"temperature\": 0.0,\n", + "}" + ] + }, + { + "cell_type": "code", + "execution_count": 12, + "metadata": {}, + "outputs": [], + "source": [ + "search_engine = LocalSearch(\n", + " llm=llm,\n", + " context_builder=context_builder,\n", + " token_encoder=token_encoder,\n", + " llm_params=llm_params,\n", + " context_builder_params=local_context_params,\n", + " response_type=\"multiple paragraphs\", # free form text describing the response type and format, can be anything, e.g. prioritized list, single paragraph, multiple paragraphs, multiple-page report\n", + ")" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### Run local search on sample queries" + ] + }, + { + "cell_type": "code", + "execution_count": 14, + "metadata": {}, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "Warning: No community records added when building community context.\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "### Overview of GraphRAG\n", + "\n", + "GraphRAG, short for Graph Retrieval-Augmented Generation, is an advanced method that leverages the modularity of graphs to enhance text summarization and question-answering tasks. Developed by NebulaGraph, this approach integrates multiple stages and concepts, including knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS), to support comprehensive and structured overviews of text corpora [Data: Entities (13, 386); Relationships (37, 185)].\n", + "\n", + "### Key Features\n", + "\n", + "1. **Graph-Based Text Indexing**: GraphRAG uses a graph-based text index to organize and retrieve information efficiently. This index is self-generated from the source documents and includes various elements extracted using Large Language Model (LLM) prompts [Data: Entities (119); Relationships (161)].\n", + "\n", + "2. **Community Summaries**: The method employs community detection algorithms to partition graphs into modular communities. Summaries of these communities are used to generate partial responses, which are then combined to form a final comprehensive answer [Data: Entities (13); Relationships (99, 107)].\n", + "\n", + "3. **Evaluation Metrics**: GraphRAG has been evaluated for its comprehensiveness, diversity, and empowerment in generating answers. It has shown to outperform naive RAG in these metrics while requiring fewer tokens [Data: Entities (13); Relationships (104, 105, 110)].\n", + "\n", + "### Applications\n", + "\n", + "GraphRAG is particularly effective in providing structured overviews of public figures across various sectors of the entertainment industry and generating answers for questions in datasets like news articles and podcast transcripts [Data: Entities (13); Relationships (108, 109, 147)].\n", + "\n", + "For more detailed information, you can access the open-source implementation at [https://aka.ms/graphrag](https://aka.ms/graphrag) [Data: Entities (19); Relationships (102)].\n", + "\n", + "In summary, GraphRAG represents a significant advancement in the field of text summarization and information retrieval, combining the strengths of graph structures and LLMs to deliver efficient and comprehensive results.\n" + ] + } + ], + "source": [ + "result = await search_engine.asearch(\"What is GraphRAG be short\")\n", + "print(result.response)" + ] + }, + { + "cell_type": "code", + "execution_count": 15, + "metadata": {}, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "Warning: No community records added when building community context.\n" + ] + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "### Purpose of GraphRAG\n", + "\n", + "GraphRAG, or Graph Retrieval-Augmented Generation, is a sophisticated method designed to enhance the capabilities of text summarization and question answering over large datasets. Its primary purpose is to leverage the natural modularity of graphs to partition data, facilitating global summarization tasks. This approach integrates multiple stages and concepts, including knowledge graph generation, retrieval-augmented generation (RAG), and query-focused summarization (QFS), to support human sensemaking over text corpora [Data: Entities (13, 407); Relationships (37)].\n", + "\n", + "### Enhancing Text Summarization\n", + "\n", + "One of the key purposes of GraphRAG is to improve the comprehensiveness and diversity of generated answers compared to traditional methods. By using a graph-based text index, GraphRAG can efficiently manage and retrieve information, allowing for more detailed and varied responses. This is particularly useful in providing comprehensive and structured overviews of public figures across various sectors of the entertainment industry, as well as generating answers for questions in datasets like news articles and podcast transcripts [Data: Entities (13); Relationships (67, 108, 109)].\n", + "\n", + "### Optimizing Token Usage\n", + "\n", + "GraphRAG also aims to optimize token usage in text generation tasks. By employing community summaries at different hierarchical levels (C0, C1, C2, C3), it can generate partial responses that are then summarized into a final global answer. This multi-stage mechanism allows for the comparison of multiple conditions, enhancing the overall effectiveness of the summarization process while requiring fewer tokens compared to traditional source text summarization methods [Data: Entities (13); Relationships (133, 134, 135, 136, 137)].\n", + "\n", + "### Versatility and Adaptability\n", + "\n", + "Another significant purpose of GraphRAG is its versatility in handling various text analysis and summarization applications. It combines global summarization with graph-based text indexing to answer questions over private text corpora, making it a powerful tool for information retrieval and natural language processing tasks. The method's adaptability is further enhanced by its ability to incorporate concepts from other systems, such as self-memory and parallel generation of community answers, which facilitate future generation cycles and improve the overall performance of the system [Data: Entities (13); Relationships (154, 173)].\n", + "\n", + "### Conclusion\n", + "\n", + "In summary, the purpose of GraphRAG is to provide a more efficient, comprehensive, and diverse approach to text summarization and question answering. By leveraging the strengths of graph structures and integrating advanced techniques from various domains, GraphRAG stands out as a powerful tool for handling complex queries and large datasets, ultimately supporting better human sensemaking and decision-making processes [Data: Entities (13, 407); Relationships (37, 67, 108, 109)].\n" + ] + } + ], + "source": [ + "question = \"What is the purpose of GraphRAG?\"\n", + "result = await search_engine.asearch(question)\n", + "print(result.response)" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "#### Inspecting the context data used to generate the response" + ] + }, + { + "cell_type": "code", + "execution_count": 16, + "metadata": {}, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
identitydescriptionnumber of relationshipsin_context
013GRAPH RAGGraph RAG (Retrieval-Augmented Generation) is ...90True
142GRAPH RAG PIPELINEGraph RAG pipeline is a process using an LLM-d...7True
2407GLOBAL APPROACH TO GRAPH RAGA method that combines knowledge graph generat...2True
311RAGRAG (Retrieval-Augmented Generation) is a deve...31True
4387GRAPHRAGA method that can create and reason over knowl...1True
\n", + "
" + ], + "text/plain": [ + " id entity \\\n", + "0 13 GRAPH RAG \n", + "1 42 GRAPH RAG PIPELINE \n", + "2 407 GLOBAL APPROACH TO GRAPH RAG \n", + "3 11 RAG \n", + "4 387 GRAPHRAG \n", + "\n", + " description number of relationships \\\n", + "0 Graph RAG (Retrieval-Augmented Generation) is ... 90 \n", + "1 Graph RAG pipeline is a process using an LLM-d... 7 \n", + "2 A method that combines knowledge graph generat... 2 \n", + "3 RAG (Retrieval-Augmented Generation) is a deve... 31 \n", + "4 A method that can create and reason over knowl... 1 \n", + "\n", + " in_context \n", + "0 True \n", + "1 True \n", + "2 True \n", + "3 True \n", + "4 True " + ] + }, + "execution_count": 16, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "result.context_data[\"entities\"].head()" + ] + }, + { + "cell_type": "code", + "execution_count": 17, + "metadata": {}, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
idsourcetargetdescriptionweightranklinksin_context
037RAGGRAPH RAGGraph RAG is a specific implementation of RAG ...7.01211True
1161GRAPH RAGGRAPH INDEXGraph RAG utilizes a self-generated graph inde...3.01082True
2183GRAPH RAGLOCAL GRAPH RAG APPROACHESGraph RAG includes local graph RAG approaches1.0911True
359RAGGRAPH INDEXA graph index is a data structure used in RAG ...2.0492True
467LLMGRAPH RAGGraph RAG utilizes Large Language Models (LLMs...4.01223True
\n", + "
" + ], + "text/plain": [ + " id source target \\\n", + "0 37 RAG GRAPH RAG \n", + "1 161 GRAPH RAG GRAPH INDEX \n", + "2 183 GRAPH RAG LOCAL GRAPH RAG APPROACHES \n", + "3 59 RAG GRAPH INDEX \n", + "4 67 LLM GRAPH RAG \n", + "\n", + " description weight rank links \\\n", + "0 Graph RAG is a specific implementation of RAG ... 7.0 121 1 \n", + "1 Graph RAG utilizes a self-generated graph inde... 3.0 108 2 \n", + "2 Graph RAG includes local graph RAG approaches 1.0 91 1 \n", + "3 A graph index is a data structure used in RAG ... 2.0 49 2 \n", + "4 Graph RAG utilizes Large Language Models (LLMs... 4.0 122 3 \n", + "\n", + " in_context \n", + "0 True \n", + "1 True \n", + "2 True \n", + "3 True \n", + "4 True " + ] + }, + "execution_count": 17, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "result.context_data[\"relationships\"].head()" + ] + }, + { + "cell_type": "code", + "execution_count": 19, + "metadata": {}, + "outputs": [ + { + "ename": "KeyError", + "evalue": "'reports'", + "output_type": "error", + "traceback": [ + "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m", + "\u001b[0;31mKeyError\u001b[0m Traceback (most recent call last)", + "Cell \u001b[0;32mIn[19], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m \u001b[43mresult\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mcontext_data\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mreports\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m\u001b[38;5;241m.\u001b[39mhead()\n", + "\u001b[0;31mKeyError\u001b[0m: 'reports'" + ] + } + ], + "source": [ + "result.context_data[\"reports\"].head()" + ] + }, + { + "cell_type": "code", + "execution_count": 20, + "metadata": {}, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
idtext
03a new RAG approach specifically targeting glo...
118these chunks into a vector space in which sim...
220dataset sizes, as well as to validate our sens...
317C0) require dramatically fewer tokens per que...
40From Local to Global: A Graph RAG Approach to\\...
\n", + "
" + ], + "text/plain": [ + " id text\n", + "0 3 a new RAG approach specifically targeting glo...\n", + "1 18 these chunks into a vector space in which sim...\n", + "2 20 dataset sizes, as well as to validate our sens...\n", + "3 17 C0) require dramatically fewer tokens per que...\n", + "4 0 From Local to Global: A Graph RAG Approach to\\..." + ] + }, + "execution_count": 20, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "result.context_data[\"sources\"].head()" + ] + }, + { + "cell_type": "code", + "execution_count": 21, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + " id entity object_id status start_date end_date \\\n", + "0 48 RAG NONE TRUE NONE NONE \n", + "\n", + " description in_context \n", + "0 RAG is identified as a developing research are... True \n" + ] + } + ], + "source": [ + "if \"claims\" in result.context_data:\n", + " print(result.context_data[\"claims\"].head())" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### Question Generation" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "This function takes a list of user queries and generates the next candidate questions." + ] + }, + { + "cell_type": "code", + "execution_count": 22, + "metadata": {}, + "outputs": [], + "source": [ + "question_generator = LocalQuestionGen(\n", + " llm=llm,\n", + " context_builder=context_builder,\n", + " token_encoder=token_encoder,\n", + " llm_params=llm_params,\n", + " context_builder_params=local_context_params,\n", + ")" + ] + }, + { + "cell_type": "code", + "execution_count": 23, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "['- What are the key contributions of AndrĆ©s Morales Esquivel to the work acknowledged in the document?', '- How do Ranade and Joshi utilize large language models in intelligence analysis?', '- What are the main findings of Kuratov et al. (2024) and Liu et al. (2023) regarding recall degradation in longer context windows?', '- What advancements have been made in multimodal models like the Gemini series?', '- How do LLM prompts enhance the performance of large language models in information retrieval tasks?']\n" + ] + } + ], + "source": [ + "question_history = [\n", + " \"Tell me about Agent Mercer\",\n", + " \"What happens in Dulce military base?\",\n", + "]\n", + "candidate_questions = await question_generator.agenerate(\n", + " question_history=question_history, context_data=None, question_count=5\n", + ")\n", + "print(candidate_questions.response)" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.11.2" + } + }, + "nbformat": 4, + "nbformat_minor": 2 +} diff --git a/notebook/get-started-graphfleet.ipynb b/notebook/get-started-graphfleet.ipynb index 047e00b1c..a75846008 100644 --- a/notebook/get-started-graphfleet.ipynb +++ b/notebook/get-started-graphfleet.ipynb @@ -24,95 +24,73 @@ "cell_type": "markdown", "metadata": {}, "source": [ - "## Batch import your PDF right from this notebook ! \n", - "Run the script below, an \"Upload\" button should have appeared, click on it and add your pdfs, it will automaticly convert and add them in a .txt in the right folder." + "## Clone the project" ] }, { "cell_type": "code", "execution_count": null, - "metadata": {}, + "metadata": { + "vscode": { + "languageId": "bat" + } + }, "outputs": [], "source": [ - "# Import libraries\n", - "import os\n", - "import json\n", - "import csv\n", - "import ipywidgets as widgets\n", - "from IPython.display import display, clear_output\n", - "\n", - "# Define the directory to save the .txt files\n", - "txt_directory = '../graphfleet/input'\n", - "if not os.path.exists(txt_directory):\n", - " os.makedirs(txt_directory)\n", - "\n", - "json_file_path = '../graphfleet/input/json'\n", - "if not os.path.exists(json_file_path):\n", - " os.makedirs(json_file_path)\n", - "\n", - "\n", - "# Create upload button\n", - "uploader = widgets.FileUpload(\n", - " accept='.json', # Accept only PDF files\n", - " multiple=True # Allow uploading multiple files\n", - ")\n", - "\n", - "# Create output area\n", - "output = widgets.Output()\n", - "\n", - "\n", - "\n", - "def json_to_txt(json_file_path, txt_directory):\n", - " # Load JSON data\n", - " with open(json_file_path, 'r') as json_file:\n", - " data = json.load(json_file)\n", - "\n", - " # Write data to TXT file\n", - " with open(txt_directory, 'w', newline='') as txt_file:\n", - " writer = csv.DictWriter(txt_file, fieldnames=data[0].keys(), delimiter='\\t')\n", - " writer.writeheader()\n", - " for row in data:\n", - " writer.writerow(row)\n", - "\n", - "# Usage\n", - "json_to_txt('input.json', 'output.txt')\n", - "\n", - "\n", - "\n", - "# Observe changes in the upload widget\n", - "uploader.observe(json_to_txt, names='value')\n", - "\n", - "# Display the upload button and output area\n", - "display(uploader)\n", - "display(output)\n", - "\n", - "\n", - "\n" + "git clone https://github.com/Qredence/GraphFleet.git" ] }, { - "cell_type": "markdown", - "metadata": {}, + "cell_type": "code", + "execution_count": null, + "metadata": { + "vscode": { + "languageId": "bat" + } + }, + "outputs": [], "source": [ - "### Great! Now that your PDF is formatted correctly and in the right location (graphfleet/input), we can initialize your workspace. Just execute the following command to get started:" + "cd GraphFleet\n", + "poetry shell\n", + "poetry install" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ - "git clone https://github.com/Qredence/GraphFleet.git" + "## GraphRAG Environment Variables" ] }, { "cell_type": "code", "execution_count": null, - "metadata": {}, + "metadata": { + "vscode": { + "languageId": "bat" + } + }, "outputs": [], "source": [ - "cd GraphFleet\n", - "poetry shell\n", - "poetry install" + "# Your Azure OpenAI API key\n", + "export GRAPHRAG_API_KEY=\"\"\n", + "\n", + "# The base URL for your Azure OpenAI API endpoint\n", + "export GRAPHRAG_API_BASE=\"\"\n", + "\n", + "# The API version you're using (e.g., \"2024-02-15-preview\")\n", + "export GRAPHRAG_API_VERSION=\"\"\n", + "\n", + "# The name of the language model you're using (e.g., \"gpt-4\")\n", + "export GRAPHRAG_LLM_MODEL=\"\"\n", + "\n", + "# The deployment name for your language model in Azure\n", + "export GRAPHRAG_DEPLOYMENT_NAME=\"\"\n", + "\n", + "# The name of the embedding model you're using (e.g., \"text-embedding-ada-002\")\n", + "export GRAPHRAG_EMBEDDING_MODEL=\"\"\n", + "\n", + "# Note: Replace the empty strings with your actual values before using GraphRAG" ] }, { @@ -190,8 +168,16 @@ "cell_type": "markdown", "metadata": {}, "source": [ - "## Indexing Your Data:\n", - " Now, let's index your data to make it searchable. This is the final step!\n" + "## Auto generate prompts for your specific data index :\n", + "This command does the following:\n", + "- Runs the prompt_tune module of GraphRAG\n", + "- Uses the configuration file settings.yaml in the ./graphfleet directory\n", + "- Sets the root directory to ./graphfleet\n", + "- Disables entity type generation with the --no-entity-types flag\n", + "- Specifies the output directory for the generated prompts as ./graphfleet/prompts\n", + "\n", + "\n", + "### This step is important because it customizes the prompts based on your specific data index, which can improve the relevance and effectiveness of your queries later on.\n" ] }, { @@ -200,7 +186,10684 @@ "metadata": {}, "outputs": [], "source": [ - "! python -m graphrag.index --root ../graphfleet" + "! python -m graphrag.prompt_tune --config ./graphfleet/settings.yaml --root ./graphfleet --no-entity-types --output ./graphfleet/prompts" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Indexing Your Data:\n", + " Now, let's index your data to make it searchable. This is the final step!\n" + ] + }, + { + "cell_type": "code", + "execution_count": 10, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "\u001b[2KLogging enabled at r \n", + "..\u001b[35m/graphfleet/../graphfleet/output/20240822-052358/reports/\u001b[0m\u001b[95mindexing-engine.log\u001b[0m\n", + "\u001b[2KStarting pipeline run for: \u001b[1;36m20240822\u001b[0m-\u001b[1;36m052358\u001b[0m, \u001b[33mdryrun\u001b[0m=\u001b[3;91mFalse\u001b[0m\n", + "\u001b[2KUsing default configuration: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"llm\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"api_key\"\u001b[0m: \u001b[32m\"==== REDACTED ====\"\u001b[0m,\n", + " \u001b[32m\"type\"\u001b[0m: \u001b[32m\"azure_openai_chat\"\u001b[0m,\n", + " \u001b[32m\"model\"\u001b[0m: \u001b[32m\"gpt-4o\"\u001b[0m,\n", + " \u001b[32m\"max_tokens\"\u001b[0m: \u001b[1;36m4000\u001b[0m,\n", + " \u001b[32m\"temperature\"\u001b[0m: \u001b[1;36m0.0\u001b[0m,\n", + " \u001b[32m\"top_p\"\u001b[0m: \u001b[1;36m1.0\u001b[0m,\n", + " \u001b[32m\"n\"\u001b[0m: \u001b[1;36m1\u001b[0m,\n", + " \u001b[32m\"request_timeout\"\u001b[0m: \u001b[1;36m180.0\u001b[0m,\n", + " \u001b[32m\"api_base\"\u001b[0m: \u001b[32m\"https://fleet-openai.openai.azure.com\"\u001b[0m,\n", + " \u001b[32m\"api_version\"\u001b[0m: \u001b[32m\"2024-04-01-preview\"\u001b[0m,\n", + " \u001b[32m\"proxy\"\u001b[0m: null,\n", + " \u001b[32m\"cognitive_services_endpoint\"\u001b[0m: null,\n", + " \u001b[32m\"deployment_name\"\u001b[0m: \u001b[32m\"gpt-4o\"\u001b[0m,\n", + " \u001b[32m\"model_supports_json\"\u001b[0m: true,\n", + " \u001b[32m\"tokens_per_minute\"\u001b[0m: \u001b[1;36m150000\u001b[0m,\n", + " \u001b[32m\"requests_per_minute\"\u001b[0m: \u001b[1;36m10000\u001b[0m,\n", + " \u001b[32m\"max_retries\"\u001b[0m: \u001b[1;36m10\u001b[0m,\n", + " \u001b[32m\"max_retry_wait\"\u001b[0m: \u001b[1;36m10.0\u001b[0m,\n", + " \u001b[32m\"sleep_on_rate_limit_recommendation\"\u001b[0m: true,\n", + " \u001b[32m\"concurrent_requests\"\u001b[0m: \u001b[1;36m25\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"parallelization\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"stagger\"\u001b[0m: \u001b[1;36m0.3\u001b[0m,\n", + " \u001b[32m\"num_threads\"\u001b[0m: \u001b[1;36m50\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"async_mode\"\u001b[0m: \u001b[32m\"threaded\"\u001b[0m,\n", + " \u001b[32m\"root_dir\"\u001b[0m: \u001b[32m\"../graphfleet/../graphfleet\"\u001b[0m,\n", + " \u001b[32m\"reporting\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"type\"\u001b[0m: \u001b[32m\"file\"\u001b[0m,\n", + " \u001b[32m\"base_dir\"\u001b[0m: \u001b[32m\"output/$\u001b[0m\u001b[32m{\u001b[0m\u001b[32mtimestamp\u001b[0m\u001b[32m}\u001b[0m\u001b[32m/reports\"\u001b[0m,\n", + " \u001b[32m\"storage_account_blob_url\"\u001b[0m: null\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"storage\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"type\"\u001b[0m: \u001b[32m\"file\"\u001b[0m,\n", + " \u001b[32m\"base_dir\"\u001b[0m: \u001b[32m\"output/$\u001b[0m\u001b[32m{\u001b[0m\u001b[32mtimestamp\u001b[0m\u001b[32m}\u001b[0m\u001b[32m/artifacts\"\u001b[0m,\n", + " \u001b[32m\"storage_account_blob_url\"\u001b[0m: null\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"cache\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"type\"\u001b[0m: \u001b[32m\"file\"\u001b[0m,\n", + " \u001b[32m\"base_dir\"\u001b[0m: \u001b[32m\"cache\"\u001b[0m,\n", + " \u001b[32m\"storage_account_blob_url\"\u001b[0m: null\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"input\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"type\"\u001b[0m: \u001b[32m\"file\"\u001b[0m,\n", + " \u001b[32m\"file_type\"\u001b[0m: \u001b[32m\"text\"\u001b[0m,\n", + " \u001b[32m\"base_dir\"\u001b[0m: \u001b[32m\"input\"\u001b[0m,\n", + " \u001b[32m\"storage_account_blob_url\"\u001b[0m: null,\n", + " \u001b[32m\"encoding\"\u001b[0m: \u001b[32m\"utf-8\"\u001b[0m,\n", + " \u001b[32m\"file_pattern\"\u001b[0m: \u001b[32m\".*\\\\.txt$\"\u001b[0m,\n", + " \u001b[32m\"file_filter\"\u001b[0m: null,\n", + " \u001b[32m\"source_column\"\u001b[0m: null,\n", + " \u001b[32m\"timestamp_column\"\u001b[0m: null,\n", + " \u001b[32m\"timestamp_format\"\u001b[0m: null,\n", + " \u001b[32m\"text_column\"\u001b[0m: \u001b[32m\"text\"\u001b[0m,\n", + " \u001b[32m\"title_column\"\u001b[0m: null,\n", + " \u001b[32m\"document_attribute_columns\"\u001b[0m: \u001b[1m[\u001b[0m\u001b[1m]\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"embed_graph\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"enabled\"\u001b[0m: true,\n", + " \u001b[32m\"num_walks\"\u001b[0m: \u001b[1;36m10\u001b[0m,\n", + " \u001b[32m\"walk_length\"\u001b[0m: \u001b[1;36m40\u001b[0m,\n", + " \u001b[32m\"window_size\"\u001b[0m: \u001b[1;36m2\u001b[0m,\n", + " \u001b[32m\"iterations\"\u001b[0m: \u001b[1;36m3\u001b[0m,\n", + " \u001b[32m\"random_seed\"\u001b[0m: \u001b[1;36m597832\u001b[0m,\n", + " \u001b[32m\"strategy\"\u001b[0m: null\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"embeddings\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"llm\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"api_key\"\u001b[0m: \u001b[32m\"==== REDACTED ====\"\u001b[0m,\n", + " \u001b[32m\"type\"\u001b[0m: \u001b[32m\"azure_openai_embedding\"\u001b[0m,\n", + " \u001b[32m\"model\"\u001b[0m: \u001b[32m\"text-embedding-ada-002\"\u001b[0m,\n", + " \u001b[32m\"max_tokens\"\u001b[0m: \u001b[1;36m4000\u001b[0m,\n", + " \u001b[32m\"temperature\"\u001b[0m: \u001b[1;36m0\u001b[0m,\n", + " \u001b[32m\"top_p\"\u001b[0m: \u001b[1;36m1\u001b[0m,\n", + " \u001b[32m\"n\"\u001b[0m: \u001b[1;36m1\u001b[0m,\n", + " \u001b[32m\"request_timeout\"\u001b[0m: \u001b[1;36m180.0\u001b[0m,\n", + " \u001b[32m\"api_base\"\u001b[0m: \u001b[32m\"https://fleet-openai.openai.azure.com\"\u001b[0m,\n", + " \u001b[32m\"api_version\"\u001b[0m: \u001b[32m\"2024-04-01-preview\"\u001b[0m,\n", + " \u001b[32m\"proxy\"\u001b[0m: null,\n", + " \u001b[32m\"cognitive_services_endpoint\"\u001b[0m: null,\n", + " \u001b[32m\"deployment_name\"\u001b[0m: \u001b[32m\"text-embedding-ada-002\"\u001b[0m,\n", + " \u001b[32m\"model_supports_json\"\u001b[0m: null,\n", + " \u001b[32m\"tokens_per_minute\"\u001b[0m: \u001b[1;36m250000\u001b[0m,\n", + " \u001b[32m\"requests_per_minute\"\u001b[0m: \u001b[1;36m10000\u001b[0m,\n", + " \u001b[32m\"max_retries\"\u001b[0m: \u001b[1;36m10\u001b[0m,\n", + " \u001b[32m\"max_retry_wait\"\u001b[0m: \u001b[1;36m10.0\u001b[0m,\n", + " \u001b[32m\"sleep_on_rate_limit_recommendation\"\u001b[0m: true,\n", + " \u001b[32m\"concurrent_requests\"\u001b[0m: \u001b[1;36m25\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"parallelization\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"stagger\"\u001b[0m: \u001b[1;36m0.3\u001b[0m,\n", + " \u001b[32m\"num_threads\"\u001b[0m: \u001b[1;36m50\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"async_mode\"\u001b[0m: \u001b[32m\"threaded\"\u001b[0m,\n", + " \u001b[32m\"batch_size\"\u001b[0m: \u001b[1;36m16\u001b[0m,\n", + " \u001b[32m\"batch_max_tokens\"\u001b[0m: \u001b[1;36m8191\u001b[0m,\n", + " \u001b[32m\"target\"\u001b[0m: \u001b[32m\"required\"\u001b[0m,\n", + " \u001b[32m\"skip\"\u001b[0m: \u001b[1m[\u001b[0m\u001b[1m]\u001b[0m,\n", + " \u001b[32m\"vector_store\"\u001b[0m: null,\n", + " \u001b[32m\"strategy\"\u001b[0m: null\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"chunks\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"size\"\u001b[0m: \u001b[1;36m512\u001b[0m,\n", + " \u001b[32m\"overlap\"\u001b[0m: \u001b[1;36m64\u001b[0m,\n", + " \u001b[32m\"group_by_columns\"\u001b[0m: \u001b[1m[\u001b[0m\n", + " \u001b[32m\"id\"\u001b[0m\n", + " \u001b[1m]\u001b[0m,\n", + " \u001b[32m\"strategy\"\u001b[0m: null,\n", + " \u001b[32m\"encoding_model\"\u001b[0m: null\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"snapshots\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"graphml\"\u001b[0m: true,\n", + " \u001b[32m\"raw_entities\"\u001b[0m: true,\n", + " \u001b[32m\"top_level_nodes\"\u001b[0m: true\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"entity_extraction\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"llm\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"api_key\"\u001b[0m: \u001b[32m\"==== REDACTED ====\"\u001b[0m,\n", + " \u001b[32m\"type\"\u001b[0m: \u001b[32m\"azure_openai_chat\"\u001b[0m,\n", + " \u001b[32m\"model\"\u001b[0m: \u001b[32m\"gpt-4o\"\u001b[0m,\n", + " \u001b[32m\"max_tokens\"\u001b[0m: \u001b[1;36m4000\u001b[0m,\n", + " \u001b[32m\"temperature\"\u001b[0m: \u001b[1;36m0.0\u001b[0m,\n", + " \u001b[32m\"top_p\"\u001b[0m: \u001b[1;36m1.0\u001b[0m,\n", + " \u001b[32m\"n\"\u001b[0m: \u001b[1;36m1\u001b[0m,\n", + " \u001b[32m\"request_timeout\"\u001b[0m: \u001b[1;36m180.0\u001b[0m,\n", + " \u001b[32m\"api_base\"\u001b[0m: \u001b[32m\"https://fleet-openai.openai.azure.com\"\u001b[0m,\n", + " \u001b[32m\"api_version\"\u001b[0m: \u001b[32m\"2024-04-01-preview\"\u001b[0m,\n", + " \u001b[32m\"proxy\"\u001b[0m: null,\n", + " \u001b[32m\"cognitive_services_endpoint\"\u001b[0m: null,\n", + " \u001b[32m\"deployment_name\"\u001b[0m: \u001b[32m\"gpt-4o\"\u001b[0m,\n", + " \u001b[32m\"model_supports_json\"\u001b[0m: true,\n", + " \u001b[32m\"tokens_per_minute\"\u001b[0m: \u001b[1;36m150000\u001b[0m,\n", + " \u001b[32m\"requests_per_minute\"\u001b[0m: \u001b[1;36m10000\u001b[0m,\n", + " \u001b[32m\"max_retries\"\u001b[0m: \u001b[1;36m10\u001b[0m,\n", + " \u001b[32m\"max_retry_wait\"\u001b[0m: \u001b[1;36m10.0\u001b[0m,\n", + " \u001b[32m\"sleep_on_rate_limit_recommendation\"\u001b[0m: true,\n", + " \u001b[32m\"concurrent_requests\"\u001b[0m: \u001b[1;36m25\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"parallelization\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"stagger\"\u001b[0m: \u001b[1;36m0.3\u001b[0m,\n", + " \u001b[32m\"num_threads\"\u001b[0m: \u001b[1;36m50\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"async_mode\"\u001b[0m: \u001b[32m\"threaded\"\u001b[0m,\n", + " \u001b[32m\"prompt\"\u001b[0m: \u001b[32m\"prompts/entity_extraction.txt\"\u001b[0m,\n", + " \u001b[32m\"entity_types\"\u001b[0m: \u001b[1m[\u001b[0m\n", + " \u001b[32m\"organization\"\u001b[0m,\n", + " \u001b[32m\"person\"\u001b[0m,\n", + " \u001b[32m\"geo\"\u001b[0m,\n", + " \u001b[32m\"event\"\u001b[0m\n", + " \u001b[1m]\u001b[0m,\n", + " \u001b[32m\"max_gleanings\"\u001b[0m: \u001b[1;36m1\u001b[0m,\n", + " \u001b[32m\"strategy\"\u001b[0m: null,\n", + " \u001b[32m\"encoding_model\"\u001b[0m: null\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"summarize_descriptions\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"llm\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"api_key\"\u001b[0m: \u001b[32m\"==== REDACTED ====\"\u001b[0m,\n", + " \u001b[32m\"type\"\u001b[0m: \u001b[32m\"azure_openai_chat\"\u001b[0m,\n", + " \u001b[32m\"model\"\u001b[0m: \u001b[32m\"gpt-4o\"\u001b[0m,\n", + " \u001b[32m\"max_tokens\"\u001b[0m: \u001b[1;36m4000\u001b[0m,\n", + " \u001b[32m\"temperature\"\u001b[0m: \u001b[1;36m0.0\u001b[0m,\n", + " \u001b[32m\"top_p\"\u001b[0m: \u001b[1;36m1.0\u001b[0m,\n", + " \u001b[32m\"n\"\u001b[0m: \u001b[1;36m1\u001b[0m,\n", + " \u001b[32m\"request_timeout\"\u001b[0m: \u001b[1;36m180.0\u001b[0m,\n", + " \u001b[32m\"api_base\"\u001b[0m: \u001b[32m\"https://fleet-openai.openai.azure.com\"\u001b[0m,\n", + " \u001b[32m\"api_version\"\u001b[0m: \u001b[32m\"2024-04-01-preview\"\u001b[0m,\n", + " \u001b[32m\"proxy\"\u001b[0m: null,\n", + " \u001b[32m\"cognitive_services_endpoint\"\u001b[0m: null,\n", + " \u001b[32m\"deployment_name\"\u001b[0m: \u001b[32m\"gpt-4o\"\u001b[0m,\n", + " \u001b[32m\"model_supports_json\"\u001b[0m: true,\n", + " \u001b[32m\"tokens_per_minute\"\u001b[0m: \u001b[1;36m150000\u001b[0m,\n", + " \u001b[32m\"requests_per_minute\"\u001b[0m: \u001b[1;36m10000\u001b[0m,\n", + " \u001b[32m\"max_retries\"\u001b[0m: \u001b[1;36m10\u001b[0m,\n", + " \u001b[32m\"max_retry_wait\"\u001b[0m: \u001b[1;36m10.0\u001b[0m,\n", + " \u001b[32m\"sleep_on_rate_limit_recommendation\"\u001b[0m: true,\n", + " \u001b[32m\"concurrent_requests\"\u001b[0m: \u001b[1;36m25\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"parallelization\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"stagger\"\u001b[0m: \u001b[1;36m0.3\u001b[0m,\n", + " \u001b[32m\"num_threads\"\u001b[0m: \u001b[1;36m50\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"async_mode\"\u001b[0m: \u001b[32m\"threaded\"\u001b[0m,\n", + " \u001b[32m\"prompt\"\u001b[0m: \u001b[32m\"prompts/summarize_descriptions.txt\"\u001b[0m,\n", + " \u001b[32m\"max_length\"\u001b[0m: \u001b[1;36m500\u001b[0m,\n", + " \u001b[32m\"strategy\"\u001b[0m: null\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"community_reports\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"llm\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"api_key\"\u001b[0m: \u001b[32m\"==== REDACTED ====\"\u001b[0m,\n", + " \u001b[32m\"type\"\u001b[0m: \u001b[32m\"azure_openai_chat\"\u001b[0m,\n", + " \u001b[32m\"model\"\u001b[0m: \u001b[32m\"gpt-4o\"\u001b[0m,\n", + " \u001b[32m\"max_tokens\"\u001b[0m: \u001b[1;36m4000\u001b[0m,\n", + " \u001b[32m\"temperature\"\u001b[0m: \u001b[1;36m0.0\u001b[0m,\n", + " \u001b[32m\"top_p\"\u001b[0m: \u001b[1;36m1.0\u001b[0m,\n", + " \u001b[32m\"n\"\u001b[0m: \u001b[1;36m1\u001b[0m,\n", + " \u001b[32m\"request_timeout\"\u001b[0m: \u001b[1;36m180.0\u001b[0m,\n", + " \u001b[32m\"api_base\"\u001b[0m: \u001b[32m\"https://fleet-openai.openai.azure.com\"\u001b[0m,\n", + " \u001b[32m\"api_version\"\u001b[0m: \u001b[32m\"2024-04-01-preview\"\u001b[0m,\n", + " \u001b[32m\"proxy\"\u001b[0m: null,\n", + " \u001b[32m\"cognitive_services_endpoint\"\u001b[0m: null,\n", + " \u001b[32m\"deployment_name\"\u001b[0m: \u001b[32m\"gpt-4o\"\u001b[0m,\n", + " \u001b[32m\"model_supports_json\"\u001b[0m: true,\n", + " \u001b[32m\"tokens_per_minute\"\u001b[0m: \u001b[1;36m150000\u001b[0m,\n", + " \u001b[32m\"requests_per_minute\"\u001b[0m: \u001b[1;36m10000\u001b[0m,\n", + " \u001b[32m\"max_retries\"\u001b[0m: \u001b[1;36m10\u001b[0m,\n", + " \u001b[32m\"max_retry_wait\"\u001b[0m: \u001b[1;36m10.0\u001b[0m,\n", + " \u001b[32m\"sleep_on_rate_limit_recommendation\"\u001b[0m: true,\n", + " \u001b[32m\"concurrent_requests\"\u001b[0m: \u001b[1;36m25\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"parallelization\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"stagger\"\u001b[0m: \u001b[1;36m0.3\u001b[0m,\n", + " \u001b[32m\"num_threads\"\u001b[0m: \u001b[1;36m50\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"async_mode\"\u001b[0m: \u001b[32m\"threaded\"\u001b[0m,\n", + " \u001b[32m\"prompt\"\u001b[0m: \u001b[32m\"prompts/community_report.txt\"\u001b[0m,\n", + " \u001b[32m\"max_length\"\u001b[0m: \u001b[1;36m2000\u001b[0m,\n", + " \u001b[32m\"max_input_length\"\u001b[0m: \u001b[1;36m8000\u001b[0m,\n", + " \u001b[32m\"strategy\"\u001b[0m: null\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"claim_extraction\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"llm\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"api_key\"\u001b[0m: \u001b[32m\"==== REDACTED ====\"\u001b[0m,\n", + " \u001b[32m\"type\"\u001b[0m: \u001b[32m\"azure_openai_chat\"\u001b[0m,\n", + " \u001b[32m\"model\"\u001b[0m: \u001b[32m\"gpt-4o\"\u001b[0m,\n", + " \u001b[32m\"max_tokens\"\u001b[0m: \u001b[1;36m4000\u001b[0m,\n", + " \u001b[32m\"temperature\"\u001b[0m: \u001b[1;36m0.0\u001b[0m,\n", + " \u001b[32m\"top_p\"\u001b[0m: \u001b[1;36m1.0\u001b[0m,\n", + " \u001b[32m\"n\"\u001b[0m: \u001b[1;36m1\u001b[0m,\n", + " \u001b[32m\"request_timeout\"\u001b[0m: \u001b[1;36m180.0\u001b[0m,\n", + " \u001b[32m\"api_base\"\u001b[0m: \u001b[32m\"https://fleet-openai.openai.azure.com\"\u001b[0m,\n", + " \u001b[32m\"api_version\"\u001b[0m: \u001b[32m\"2024-04-01-preview\"\u001b[0m,\n", + " \u001b[32m\"proxy\"\u001b[0m: null,\n", + " \u001b[32m\"cognitive_services_endpoint\"\u001b[0m: null,\n", + " \u001b[32m\"deployment_name\"\u001b[0m: \u001b[32m\"gpt-4o\"\u001b[0m,\n", + " \u001b[32m\"model_supports_json\"\u001b[0m: true,\n", + " \u001b[32m\"tokens_per_minute\"\u001b[0m: \u001b[1;36m150000\u001b[0m,\n", + " \u001b[32m\"requests_per_minute\"\u001b[0m: \u001b[1;36m10000\u001b[0m,\n", + " \u001b[32m\"max_retries\"\u001b[0m: \u001b[1;36m10\u001b[0m,\n", + " \u001b[32m\"max_retry_wait\"\u001b[0m: \u001b[1;36m10.0\u001b[0m,\n", + " \u001b[32m\"sleep_on_rate_limit_recommendation\"\u001b[0m: true,\n", + " \u001b[32m\"concurrent_requests\"\u001b[0m: \u001b[1;36m25\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"parallelization\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"stagger\"\u001b[0m: \u001b[1;36m0.3\u001b[0m,\n", + " \u001b[32m\"num_threads\"\u001b[0m: \u001b[1;36m50\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"async_mode\"\u001b[0m: \u001b[32m\"threaded\"\u001b[0m,\n", + " \u001b[32m\"enabled\"\u001b[0m: true,\n", + " \u001b[32m\"prompt\"\u001b[0m: \u001b[32m\"prompts/claim_extraction.txt\"\u001b[0m,\n", + " \u001b[32m\"description\"\u001b[0m: \u001b[32m\"Any claims or facts that could be relevant to \u001b[0m\n", + "\u001b[32minformation discovery.\"\u001b[0m,\n", + " \u001b[32m\"max_gleanings\"\u001b[0m: \u001b[1;36m1\u001b[0m,\n", + " \u001b[32m\"strategy\"\u001b[0m: null,\n", + " \u001b[32m\"encoding_model\"\u001b[0m: null\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"cluster_graph\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"max_cluster_size\"\u001b[0m: \u001b[1;36m10\u001b[0m,\n", + " \u001b[32m\"strategy\"\u001b[0m: null\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"umap\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"enabled\"\u001b[0m: true\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"local_search\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"text_unit_prop\"\u001b[0m: \u001b[1;36m0.5\u001b[0m,\n", + " \u001b[32m\"community_prop\"\u001b[0m: \u001b[1;36m0.1\u001b[0m,\n", + " \u001b[32m\"conversation_history_max_turns\"\u001b[0m: \u001b[1;36m5\u001b[0m,\n", + " \u001b[32m\"top_k_entities\"\u001b[0m: \u001b[1;36m10\u001b[0m,\n", + " \u001b[32m\"top_k_relationships\"\u001b[0m: \u001b[1;36m10\u001b[0m,\n", + " \u001b[32m\"temperature\"\u001b[0m: \u001b[1;36m0.0\u001b[0m,\n", + " \u001b[32m\"top_p\"\u001b[0m: \u001b[1;36m1.0\u001b[0m,\n", + " \u001b[32m\"n\"\u001b[0m: \u001b[1;36m1\u001b[0m,\n", + " \u001b[32m\"max_tokens\"\u001b[0m: \u001b[1;36m12000\u001b[0m,\n", + " \u001b[32m\"llm_max_tokens\"\u001b[0m: \u001b[1;36m2000\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"global_search\"\u001b[0m: \u001b[1m{\u001b[0m\n", + " \u001b[32m\"temperature\"\u001b[0m: \u001b[1;36m0.0\u001b[0m,\n", + " \u001b[32m\"top_p\"\u001b[0m: \u001b[1;36m1.0\u001b[0m,\n", + " \u001b[32m\"n\"\u001b[0m: \u001b[1;36m1\u001b[0m,\n", + " \u001b[32m\"max_tokens\"\u001b[0m: \u001b[1;36m12000\u001b[0m,\n", + " \u001b[32m\"data_max_tokens\"\u001b[0m: \u001b[1;36m12000\u001b[0m,\n", + " \u001b[32m\"map_max_tokens\"\u001b[0m: \u001b[1;36m1000\u001b[0m,\n", + " \u001b[32m\"reduce_max_tokens\"\u001b[0m: \u001b[1;36m2000\u001b[0m,\n", + " \u001b[32m\"concurrency\"\u001b[0m: \u001b[1;36m32\u001b[0m\n", + " \u001b[1m}\u001b[0m,\n", + " \u001b[32m\"encoding_model\"\u001b[0m: \u001b[32m\"cl100k_base\"\u001b[0m,\n", + " \u001b[32m\"skip_workflows\"\u001b[0m: \u001b[1m[\u001b[0m\u001b[1m]\u001b[0m\n", + "\u001b[1m}\u001b[0m\n", + "\u001b[2Kā ¹ GraphRAG Indexer \n", + "\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer e.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K/workspaces/GraphFleet/.venv/lib/python3.11/site-packages/numpy/core/fromnumeric\n", + ".py:59: FutureWarning: 'DataFrame.swapaxes' is deprecated and will be removed in\n", + "a future version. Please use 'DataFrame.transpose' instead.\n", + " return bound(*args, **kwds)\n", + "ā ¦ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā””ā”€ā”€ create_base_text_units\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2KšŸš€ \u001b[32mcreate_base_text_units\u001b[0mā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:00\u001b[0m\n", + "ā ¦ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K id \u001b[33m...\u001b[0m n_tokens\n", + "\u001b[1;36m0\u001b[0m e8d83e6e7a7c0f57b218cef24976b745 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m1\u001b[0m f0306814bf64f5c9e79603fc6a52f4ea \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m2\u001b[0m fb3c48579608fa28be585ceb6cd2f0fe \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m3\u001b[0m 21e52bc06a82796b1f4bcd73edda1f2a \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m4\u001b[0m bc9e2c9e369c4108cf4f6dd5f60960f4 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m5\u001b[0m 2c6ed90897310eea2f28e33fff1c32b0 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m6\u001b[0m 7fb7d9ce2da9c940a32afdd87d1d9e56 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m7\u001b[0m 843fc5421e086120ffa1c75856ecf6cd \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m8\u001b[0m 1d07b4248c2655081c7af0e373bd70c9 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m9\u001b[0m 922778ce1cb2fdd6dbab1746c8795620 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m10\u001b[0m 973164fa90bf2b4ee267f4fd795916bf \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m11\u001b[0m 322e02986c8724eedbcf3ebfa20b989c \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m12\u001b[0m e8c8f911135faf3ff35f24107eb3f99c \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m13\u001b[0m 718017a4871c909420f84b85b8ba969d \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m14\u001b[0m ebf5249c888e07fedce6572a4c03f88c \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m15\u001b[0m 4c855404ee3d3c94aa2136f1513c666f \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m16\u001b[0m 36db32c37e1987e2c5863898ad882190 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m17\u001b[0m 6f33a085ff3304e5994f7fbb86c881a4 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m18\u001b[0m f35de4d9fb65f1d5a392064b20545c19 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m19\u001b[0m 92e93fc6449756c0a60200636b297f65 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m20\u001b[0m e4d9b12cf2b4c691c74019eefff4fb39 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m21\u001b[0m 086021a89900a39bcb62036981737bfa \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m22\u001b[0m 58ae80c41cfe46db39da26b6a83584e5 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m23\u001b[0m 00e8e4e881bd0862022f4dfc913b900b \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m24\u001b[0m 71f6daf11e64e5273a3847d46bf228e1 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m25\u001b[0m 6cd82819982879bd164547d2773ba5c7 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m26\u001b[0m 833e7d67dcd30790b26b71c9b5306f6b \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m27\u001b[0m 8d87efac8c50cf20cdf26bf61e5e2035 \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m28\u001b[0m fc4b27d64f055b7fc30176ba110dd02e \u001b[33m...\u001b[0m \u001b[1;36m512\u001b[0m\n", + "\u001b[1;36m29\u001b[0m b1bbda43309e8e0e2175ea034aa88e13 \u001b[33m...\u001b[0m \u001b[1;36m95\u001b[0m\n", + "\n", + "\u001b[1m[\u001b[0m\u001b[1;36m30\u001b[0m rows x \u001b[1;36m5\u001b[0m columns\u001b[1m]\u001b[0m\n", + "ā ¦ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:00\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:01\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:02\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:03\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:04\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:05\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:06\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:07\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:08\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:09\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:10\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:11\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:12\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:13\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:14\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:15\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:16\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:17\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:18\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:19\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:20\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:21\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:22\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:23\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:24\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:25\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:26\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:27\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:28\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:29\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:30\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:31\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:32\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:33\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:34\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:35\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:36\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:37\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:38\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:39\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:40\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:41\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:42\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:43\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:44\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:45\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:46\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:47\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:48\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:49\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:50\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:51\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:52\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:53\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:54\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:55\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:56\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:57\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:58\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:00:59\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:00\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:01\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:02\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:03\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:04\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 0%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:05\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 3%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:06\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 3%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:06\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 3%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:07\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 3%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:08\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 3%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:09\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 3%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:10\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 3%\u001b[0m \u001b[36m-:--:--\u001b[0m \u001b[33m0:01:11\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 7%\u001b[0m \u001b[36m0:02:54\u001b[0m \u001b[33m0:01:12\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 7%\u001b[0m \u001b[36m0:02:54\u001b[0m \u001b[33m0:01:12\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 10%\u001b[0m \u001b[36m0:01:31\u001b[0m \u001b[33m0:01:12\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 13%\u001b[0m \u001b[36m0:00:59\u001b[0m \u001b[33m0:01:13\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 13%\u001b[0m \u001b[36m0:00:59\u001b[0m \u001b[33m0:01:14\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 13%\u001b[0m \u001b[36m0:00:59\u001b[0m \u001b[33m0:01:15\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 13%\u001b[0m \u001b[36m0:00:59\u001b[0m \u001b[33m0:01:16\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer 0mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 17%\u001b[0m \u001b[36m0:01:11\u001b[0m \u001b[33m0:01:17\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer 0mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 17%\u001b[0m \u001b[36m0:01:11\u001b[0m \u001b[33m0:01:17\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer 90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 20%\u001b[0m \u001b[36m0:01:01\u001b[0m \u001b[33m0:01:18\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer 90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 20%\u001b[0m \u001b[36m0:01:01\u001b[0m \u001b[33m0:01:18\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer [90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 23%\u001b[0m \u001b[36m0:00:52\u001b[0m \u001b[33m0:01:19\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer [90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 23%\u001b[0m \u001b[36m0:00:52\u001b[0m \u001b[33m0:01:19\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer \u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 27%\u001b[0m \u001b[36m0:00:44\u001b[0m \u001b[33m0:01:20\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer \u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 27%\u001b[0m \u001b[36m0:00:44\u001b[0m \u001b[33m0:01:20\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer \u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 27%\u001b[0m \u001b[36m0:00:44\u001b[0m \u001b[33m0:01:21\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 30%\u001b[0m \u001b[36m0:00:44\u001b[0m \u001b[33m0:01:22\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 30%\u001b[0m \u001b[36m0:00:44\u001b[0m \u001b[33m0:01:22\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer 0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 33%\u001b[0m \u001b[36m0:00:38\u001b[0m \u001b[33m0:01:23\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer 0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 33%\u001b[0m \u001b[36m0:00:38\u001b[0m \u001b[33m0:01:24\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer [0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 37%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:01:24\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer [0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 37%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:01:25\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer \u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 40%\u001b[0m \u001b[36m0:00:32\u001b[0m \u001b[33m0:01:25\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer \u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 40%\u001b[0m \u001b[36m0:00:32\u001b[0m \u001b[33m0:01:26\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer \u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 40%\u001b[0m \u001b[36m0:00:32\u001b[0m \u001b[33m0:01:27\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā•ŗ\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 43%\u001b[0m \u001b[36m0:00:31\u001b[0m \u001b[33m0:01:27\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā•ŗ\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 43%\u001b[0m \u001b[36m0:00:31\u001b[0m \u001b[33m0:01:28\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā•ŗ\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 43%\u001b[0m \u001b[36m0:00:31\u001b[0m \u001b[33m0:01:29\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer mā•ŗ\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 47%\u001b[0m \u001b[36m0:00:29\u001b[0m \u001b[33m0:01:29\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer mā•ŗ\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 47%\u001b[0m \u001b[36m0:00:29\u001b[0m \u001b[33m0:01:30\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer 1mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 50%\u001b[0m \u001b[36m0:00:27\u001b[0m \u001b[33m0:01:30\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer 1mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 50%\u001b[0m \u001b[36m0:00:27\u001b[0m \u001b[33m0:01:31\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer 91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 53%\u001b[0m \u001b[36m0:00:24\u001b[0m \u001b[33m0:01:31\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer [91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 57%\u001b[0m \u001b[36m0:00:21\u001b[0m \u001b[33m0:01:31\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer [91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 57%\u001b[0m \u001b[36m0:00:21\u001b[0m \u001b[33m0:01:32\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer [91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 57%\u001b[0m \u001b[36m0:00:21\u001b[0m \u001b[33m0:01:33\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer \u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 60%\u001b[0m \u001b[36m0:00:20\u001b[0m \u001b[33m0:01:33\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer \u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 60%\u001b[0m \u001b[36m0:00:20\u001b[0m \u001b[33m0:01:34\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:34\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:35\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:36\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:37\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:38\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:39\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:40\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:41\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:42\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:43\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:44\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:45\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:46\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:47\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 63%\u001b[0m \u001b[36m0:00:18\u001b[0m \u001b[33m0:01:48\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer 0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 67%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:48\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer 0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 67%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:49\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer 0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 67%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:50\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer 0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 67%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:51\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:51\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:52\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:53\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:54\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:55\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:56\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:57\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:58\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:01:59\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:02:00\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:02:01\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:02:02\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:02:03\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer [0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 70%\u001b[0m \u001b[36m0:00:22\u001b[0m \u001b[33m0:02:04\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer \u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 73%\u001b[0m \u001b[36m0:01:06\u001b[0m \u001b[33m0:02:04\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 77%\u001b[0m \u001b[36m0:00:39\u001b[0m \u001b[33m0:02:05\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 77%\u001b[0m \u001b[36m0:00:39\u001b[0m \u001b[33m0:02:06\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 77%\u001b[0m \u001b[36m0:00:39\u001b[0m \u001b[33m0:02:07\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 77%\u001b[0m \u001b[36m0:00:39\u001b[0m \u001b[33m0:02:08\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 77%\u001b[0m \u001b[36m0:00:39\u001b[0m \u001b[33m0:02:09\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 77%\u001b[0m \u001b[36m0:00:39\u001b[0m \u001b[33m0:02:10\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 77%\u001b[0m \u001b[36m0:00:39\u001b[0m \u001b[33m0:02:11\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 80%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:02:11\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 80%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:02:12\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 80%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:02:13\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 80%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:02:14\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 80%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:02:15\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 80%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:02:16\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 80%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:02:17\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 80%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:02:18\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 80%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:02:19\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 80%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:02:20\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”ā”\u001b[0m \u001b[35m 80%\u001b[0m \u001b[36m0:00:35\u001b[0m \u001b[33m0:02:21\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”\u001b[0m \u001b[35m 83%\u001b[0m \u001b[36m0:00:30\u001b[0m \u001b[33m0:02:22\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”\u001b[0m \u001b[35m 83%\u001b[0m \u001b[36m0:00:30\u001b[0m \u001b[33m0:02:22\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”\u001b[0m \u001b[35m 83%\u001b[0m \u001b[36m0:00:30\u001b[0m \u001b[33m0:02:23\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”\u001b[0m \u001b[35m 83%\u001b[0m \u001b[36m0:00:30\u001b[0m \u001b[33m0:02:24\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”\u001b[0m \u001b[35m 83%\u001b[0m \u001b[36m0:00:30\u001b[0m \u001b[33m0:02:25\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”\u001b[0m \u001b[35m 83%\u001b[0m \u001b[36m0:00:30\u001b[0m \u001b[33m0:02:26\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”\u001b[0m \u001b[35m 83%\u001b[0m \u001b[36m0:00:30\u001b[0m \u001b[33m0:02:27\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”ā”\u001b[0m \u001b[35m 83%\u001b[0m \u001b[36m0:00:30\u001b[0m \u001b[33m0:02:28\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:28\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:29\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:30\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:31\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:32\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:33\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:34\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:35\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:36\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:37\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:38\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:39\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:40\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:41\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”ā”\u001b[0m \u001b[35m 87%\u001b[0m \u001b[36m0:00:25\u001b[0m \u001b[33m0:02:42\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”\u001b[0m \u001b[35m 90%\u001b[0m \u001b[36m0:00:33\u001b[0m \u001b[33m0:02:43\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”\u001b[0m \u001b[35m 90%\u001b[0m \u001b[36m0:00:33\u001b[0m \u001b[33m0:02:43\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”\u001b[0m \u001b[35m 90%\u001b[0m \u001b[36m0:00:33\u001b[0m \u001b[33m0:02:44\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”\u001b[0m \u001b[35m 90%\u001b[0m \u001b[36m0:00:33\u001b[0m \u001b[33m0:02:45\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”\u001b[0m \u001b[35m 90%\u001b[0m \u001b[36m0:00:33\u001b[0m \u001b[33m0:02:46\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”\u001b[0m \u001b[35m 90%\u001b[0m \u001b[36m0:00:33\u001b[0m \u001b[33m0:02:47\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”\u001b[0m \u001b[35m 90%\u001b[0m \u001b[36m0:00:33\u001b[0m \u001b[33m0:02:48\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”\u001b[0m \u001b[35m 90%\u001b[0m \u001b[36m0:00:33\u001b[0m \u001b[33m0:02:49\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”ā”\u001b[0m \u001b[35m 90%\u001b[0m \u001b[36m0:00:33\u001b[0m \u001b[33m0:02:50\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”\u001b[0m \u001b[35m 93%\u001b[0m \u001b[36m0:00:20\u001b[0m \u001b[33m0:02:51\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”\u001b[0m \u001b[35m 93%\u001b[0m \u001b[36m0:00:20\u001b[0m \u001b[33m0:02:51\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”\u001b[0m \u001b[35m 93%\u001b[0m \u001b[36m0:00:20\u001b[0m \u001b[33m0:02:52\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”\u001b[0m \u001b[35m 93%\u001b[0m \u001b[36m0:00:20\u001b[0m \u001b[33m0:02:53\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”\u001b[0m \u001b[35m 93%\u001b[0m \u001b[36m0:00:20\u001b[0m \u001b[33m0:02:54\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”\u001b[0m \u001b[35m 93%\u001b[0m \u001b[36m0:00:20\u001b[0m \u001b[33m0:02:55\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”\u001b[0m \u001b[35m 93%\u001b[0m \u001b[36m0:00:20\u001b[0m \u001b[33m0:02:56\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”\u001b[0m \u001b[35m 93%\u001b[0m \u001b[36m0:00:20\u001b[0m \u001b[33m0:02:57\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”ā”\u001b[0m \u001b[35m 93%\u001b[0m \u001b[36m0:00:20\u001b[0m \u001b[33m0:02:58\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:02:59\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:02:59\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:00\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:01\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:02\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:03\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¼ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:04\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:06\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:07\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:08\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:09\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā § GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:10\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:11\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:12\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā “ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:13\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:14\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‹ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:15\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ø GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:16\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:17\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā  GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:18\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ™ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:19\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "ā””ā”€ā”€ create_base_extracted_entities\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer ā”ā”ā”ā”ā”ā”ā”\u001b[0m\u001b[91mā•ø\u001b[0m\u001b[90mā”\u001b[0m \u001b[35m 97%\u001b[0m \u001b[36m0:00:08\u001b[0m \u001b[33m0:03:20\u001b[0m\n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ¦ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2Kā ‡ GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2KšŸš€ \u001b[32mcreate_base_extracted_entities\u001b[0m\n", + "ā  GraphRAG Indexer \n", + "ā”œā”€ā”€ Loading Input (InputFileType.text) - 1 files loaded (0 filtered) \u001b[90mā”\u001b[0m \u001b[35m100%\u001b[0m \u001b[36mā€¦\u001b[0m \u001b[33m0ā€¦\u001b[0m\n", + "ā”œā”€ā”€ create_base_text_units\n", + "\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K\u001b[1A\u001b[2K entity_graph\n", + "\u001b[1;36m0\u001b[0m =1.0.2", markers = "python_version < \"3.11\""} idna = ">=2.8" sniffio = ">=1.1" -typing-extensions = {version = ">=4.1", markers = "python_version < \"3.11\""} [package.extras] doc = ["Sphinx (>=7)", "packaging", "sphinx-autodoc-typehints (>=1.2.0)", "sphinx-rtd-theme"] @@ -241,9 +239,6 @@ files = [ {file = "async_lru-2.0.4-py3-none-any.whl", hash = "sha256:ff02944ce3c288c5be660c42dbcca0742b32c3b279d6dceda655190240b99224"}, ] -[package.dependencies] -typing-extensions = {version = ">=4.0.0", markers = "python_version < \"3.11\""} - [[package]] name = "at" version = "0.0.3" @@ -560,137 +555,6 @@ charset-normalizer = ["charset-normalizer"] html5lib = ["html5lib"] lxml = ["lxml"] -[[package]] -name = "bitarray" -version = "2.9.2" -description = "efficient arrays of booleans -- C extension" -optional = false -python-versions = "*" -files = [ - {file = "bitarray-2.9.2-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:917905de565d9576eb20f53c797c15ba88b9f4f19728acabec8d01eee1d3756a"}, - {file = "bitarray-2.9.2-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:b35bfcb08b7693ab4bf9059111a6e9f14e07d57ac93cd967c420db58ab9b71e1"}, - {file = "bitarray-2.9.2-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:ea1923d2e7880f9e1959e035da661767b5a2e16a45dfd57d6aa831e8b65ee1bf"}, - {file = "bitarray-2.9.2-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:1e0b63a565e8a311cc8348ff1262d5784df0f79d64031d546411afd5dd7ef67d"}, - {file = "bitarray-2.9.2-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:cf0620da2b81946d28c0b16f3e3704d38e9837d85ee4f0652816e2609aaa4fed"}, - {file = "bitarray-2.9.2-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:79a9b8b05f2876c7195a2b698c47528e86a73c61ea203394ff8e7a4434bda5c8"}, - {file = "bitarray-2.9.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:345c76b349ff145549652436235c5532e5bfe9db690db6f0a6ad301c62b9ef21"}, - {file = "bitarray-2.9.2-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:4e2936f090bf3f4d1771f44f9077ebccdbc0415d2b598d51a969afcb519df505"}, - {file = "bitarray-2.9.2-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:f9346e98fc2abcef90b942973087e2462af6d3e3710e82938078d3493f7fef52"}, - {file = "bitarray-2.9.2-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:e6ec283d4741befb86e8c3ea2e9ac1d17416c956d392107e45263e736954b1f7"}, - {file = "bitarray-2.9.2-cp310-cp310-musllinux_1_1_ppc64le.whl", hash = "sha256:962892646599529917ef26266091e4cb3077c88b93c3833a909d68dcc971c4e3"}, - {file = "bitarray-2.9.2-cp310-cp310-musllinux_1_1_s390x.whl", hash = "sha256:e8da5355d7d75a52df5b84750989e34e39919ec7e59fafc4c104cc1607ab2d31"}, - {file = "bitarray-2.9.2-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:603e7d640e54ad764d2b4da6b61e126259af84f253a20f512dd10689566e5478"}, - {file = "bitarray-2.9.2-cp310-cp310-win32.whl", hash = "sha256:f00079f8e69d75c2a417de7961a77612bb77ef46c09bc74607d86de4740771ef"}, - {file = "bitarray-2.9.2-cp310-cp310-win_amd64.whl", hash = "sha256:1bb33673e7f7190a65f0a940c1ef63266abdb391f4a3e544a47542d40a81f536"}, - {file = "bitarray-2.9.2-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:fe71fd4b76380c2772f96f1e53a524da7063645d647a4fcd3b651bdd80ca0f2e"}, - {file = "bitarray-2.9.2-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:d527172919cdea1e13994a66d9708a80c3d33dedcf2f0548e4925e600fef3a3a"}, - {file = "bitarray-2.9.2-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:052c5073bdcaa9dd10628d99d37a2f33ec09364b86dd1f6281e2d9f8d3db3060"}, - {file = "bitarray-2.9.2-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:e064caa55a6ed493aca1eda06f8b3f689778bc780a75e6ad7724642ba5dc62f7"}, - {file = "bitarray-2.9.2-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:508069a04f658210fdeee85a7a0ca84db4bcc110cbb1d21f692caa13210f24a7"}, - {file = "bitarray-2.9.2-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:4da73ebd537d75fa7bccfc2228fcaedea0803f21dd9d0bf0d3b67fef3c4af294"}, - {file = "bitarray-2.9.2-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:5cb378eaa65cd43098f11ff5d27e48ee3b956d2c00d2d6b5bfc2a09fe183be47"}, - {file = "bitarray-2.9.2-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d14c790b91f6cbcd9b718f88ed737c78939980c69ac8c7f03dd7e60040c12951"}, - {file = "bitarray-2.9.2-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:7eea9318293bc0ea6447e9ebfba600a62f3428bea7e9c6d42170ae4f481dbab3"}, - {file = "bitarray-2.9.2-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:b76ffec27c7450b8a334f967366a9ebadaea66ee43f5b530c12861b1a991f503"}, - {file = "bitarray-2.9.2-cp311-cp311-musllinux_1_1_ppc64le.whl", hash = "sha256:76b76a07d4ee611405045c6950a1e24c4362b6b44808d4ad6eea75e0dbc59af4"}, - {file = "bitarray-2.9.2-cp311-cp311-musllinux_1_1_s390x.whl", hash = "sha256:c7d16beeaaab15b075990cd26963d6b5b22e8c5becd131781514a00b8bdd04bd"}, - {file = "bitarray-2.9.2-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:60df43e868a615c7e15117a1e1c2e5e11f48f6457280eba6ddf8fbefbec7da99"}, - {file = "bitarray-2.9.2-cp311-cp311-win32.whl", hash = "sha256:e788608ed7767b7b3bbde6d49058bccdf94df0de9ca75d13aa99020cc7e68095"}, - {file = "bitarray-2.9.2-cp311-cp311-win_amd64.whl", hash = "sha256:a23397da092ef0a8cfe729571da64c2fc30ac18243caa82ac7c4f965087506ff"}, - {file = "bitarray-2.9.2-cp312-cp312-macosx_10_9_universal2.whl", hash = "sha256:90e3a281ffe3897991091b7c46fca38c2675bfd4399ffe79dfeded6c52715436"}, - {file = "bitarray-2.9.2-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:bed637b674db5e6c8a97a4a321e3e4d73e72d50b5c6b29950008a93069cc64cd"}, - {file = "bitarray-2.9.2-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:e49066d251dbbe4e6e3a5c3937d85b589e40e2669ad0eef41a00f82ec17d844b"}, - {file = "bitarray-2.9.2-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:3c4344e96642e2211fb3a50558feff682c31563a4c64529a931769d40832ca79"}, - {file = "bitarray-2.9.2-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:aeb60962ec4813c539a59fbd4f383509c7222b62c3fb1faa76b54943a613e33a"}, - {file = "bitarray-2.9.2-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:ed0f7982f10581bb16553719e5e8f933e003f5b22f7d25a68bdb30fac630a6ff"}, - {file = "bitarray-2.9.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:c71d1cabdeee0cdda4669168618f0e46b7dace207b29da7b63aaa1adc2b54081"}, - {file = "bitarray-2.9.2-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:b0ef2d0a6f1502d38d911d25609b44c6cc27bee0a4363dd295df78b075041b60"}, - {file = "bitarray-2.9.2-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:6f71d92f533770fb027388b35b6e11988ab89242b883f48a6fe7202d238c61f8"}, - {file = "bitarray-2.9.2-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:ba0734aa300757c924f3faf8148e1b8c247176a0ac8e16aefdf9c1eb19e868f7"}, - {file = "bitarray-2.9.2-cp312-cp312-musllinux_1_1_ppc64le.whl", hash = "sha256:d91406f413ccbf4af6ab5ae7bc78f772a95609f9ddd14123db36ef8c37116d95"}, - {file = "bitarray-2.9.2-cp312-cp312-musllinux_1_1_s390x.whl", hash = "sha256:87abb7f80c0a042f3fe8e5264da1a2756267450bb602110d5327b8eaff7682e7"}, - {file = "bitarray-2.9.2-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:4b558ce85579b51a2e38703877d1e93b7728a7af664dd45a34e833534f0b755d"}, - {file = "bitarray-2.9.2-cp312-cp312-win32.whl", hash = "sha256:dac2399ee2889fbdd3472bfc2ede74c34cceb1ccf29a339964281a16eb1d3188"}, - {file = "bitarray-2.9.2-cp312-cp312-win_amd64.whl", hash = "sha256:48a30d718d1a6dfc22a49547450107abe8f4afdf2abdcbe76eb9ed88edc49498"}, - {file = "bitarray-2.9.2-cp36-cp36m-macosx_10_9_x86_64.whl", hash = "sha256:2c6be1b651fad8f3adb7a5aa12c65b612cd9b89530969af941844ae680f7d981"}, - {file = "bitarray-2.9.2-cp36-cp36m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:c5b399ae6ab975257ec359f03b48fc00b1c1cd109471e41903548469b8feae5c"}, - {file = "bitarray-2.9.2-cp36-cp36m-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:0b3543c8a1cb286ad105f11c25d8d0f712f41c5c55f90be39f0e5a1376c7d0b0"}, - {file = "bitarray-2.9.2-cp36-cp36m-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:03adaacb79e2fb8f483ab3a67665eec53bb3fd0cd5dbd7358741aef124688db3"}, - {file = "bitarray-2.9.2-cp36-cp36m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:9ae5b0657380d2581e13e46864d147a52c1e2bbac9f59b59c576e42fa7d10cf0"}, - {file = "bitarray-2.9.2-cp36-cp36m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:7c1f4bf6ea8eb9d7f30808c2e9894237a96650adfecbf5f3643862dc5982f89e"}, - {file = "bitarray-2.9.2-cp36-cp36m-musllinux_1_1_aarch64.whl", hash = "sha256:a8873089be2aa15494c0f81af1209f6e1237d762c5065bc4766c1b84321e1b50"}, - {file = "bitarray-2.9.2-cp36-cp36m-musllinux_1_1_i686.whl", hash = "sha256:677e67f50e2559efc677a4366707070933ad5418b8347a603a49a070890b19bc"}, - {file = "bitarray-2.9.2-cp36-cp36m-musllinux_1_1_ppc64le.whl", hash = "sha256:a620d8ce4ea2f1c73c6b6b1399e14cb68c6915e2be3fad5808c2998ed55b4acf"}, - {file = "bitarray-2.9.2-cp36-cp36m-musllinux_1_1_s390x.whl", hash = "sha256:64115ccabbdbe279c24c367b629c6b1d3da9ed36c7420129e27c338a3971bfee"}, - {file = "bitarray-2.9.2-cp36-cp36m-musllinux_1_1_x86_64.whl", hash = "sha256:5d6fb422772e75385b76ad1c52f45a68bd4efafd8be8d0061c11877be74c4d43"}, - {file = "bitarray-2.9.2-cp36-cp36m-win32.whl", hash = "sha256:852e202875dd6dfd6139ce7ec4e98dac2b17d8d25934dc99900831e81c3adaef"}, - {file = "bitarray-2.9.2-cp36-cp36m-win_amd64.whl", hash = "sha256:7dfefdcb0dc6a3ba9936063cec65a74595571b375beabe18742b3d91d087eefd"}, - {file = "bitarray-2.9.2-cp37-cp37m-macosx_10_9_x86_64.whl", hash = "sha256:b306c4cf66912511422060f7f5e1149c8bdb404f8e00e600561b0749fdd45659"}, - {file = "bitarray-2.9.2-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:a09c4f81635408e3387348f415521d4b94198c562c23330f560596a6aaa26eaf"}, - {file = "bitarray-2.9.2-cp37-cp37m-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:5361413fd2ecfdf44dc8f065177dc6aba97fa80a91b815586cb388763acf7f8d"}, - {file = "bitarray-2.9.2-cp37-cp37m-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:e8a9475d415ef1eaae7942df6f780fa4dcd48fce32825eda591a17abba869299"}, - {file = "bitarray-2.9.2-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:c9b87baa7bfff9a5878fcc1bffe49ecde6e647a72a64b39a69cd8a2992a43a34"}, - {file = "bitarray-2.9.2-cp37-cp37m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:bb6b86cfdfc503e92cb71c68766a24565359136961642504a7cc9faf936d9c88"}, - {file = "bitarray-2.9.2-cp37-cp37m-musllinux_1_1_aarch64.whl", hash = "sha256:cd56b8ae87ebc71bcacbd73615098e8a8de952ecbb5785b6b4e2b07da8a06e1f"}, - {file = "bitarray-2.9.2-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:3fa909cfd675004aed8b4cc9df352415933656e0155a6209d878b7cb615c787e"}, - {file = "bitarray-2.9.2-cp37-cp37m-musllinux_1_1_ppc64le.whl", hash = "sha256:b069ca9bf728e0c5c5b60e00a89df9af34cc170c695c3bfa3b372d8f40288efb"}, - {file = "bitarray-2.9.2-cp37-cp37m-musllinux_1_1_s390x.whl", hash = "sha256:6067f2f07a7121749858c7daa93c8774325c91590b3e81a299621e347740c2ae"}, - {file = "bitarray-2.9.2-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:321841cdad1dd0f58fe62e80e9c9c7531f8ebf8be93f047401e930dc47425b1e"}, - {file = "bitarray-2.9.2-cp37-cp37m-win32.whl", hash = "sha256:54e16e32e60973bb83c315de9975bc1bcfc9bd50bb13001c31da159bc49b0ca1"}, - {file = "bitarray-2.9.2-cp37-cp37m-win_amd64.whl", hash = "sha256:f4dcadb7b8034aa3491ee8f5a69b3d9ba9d7d1e55c3cc1fc45be313e708277f8"}, - {file = "bitarray-2.9.2-cp38-cp38-macosx_10_9_universal2.whl", hash = "sha256:c8919fdbd3bb596b104388b56ae4b266eb28da1f2f7dff2e1f9334a21840fe96"}, - {file = "bitarray-2.9.2-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:eb7a9d8a2e400a1026de341ad48e21670a6261a75b06df162c5c39b0d0e7c8f4"}, - {file = "bitarray-2.9.2-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:6ec84668dd7b937874a2b2c293cd14ba84f37be0d196dead852e0ada9815d807"}, - {file = "bitarray-2.9.2-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f2de9a31c34e543ae089fd2a5ced01292f725190e379921384f695e2d7184bd3"}, - {file = "bitarray-2.9.2-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:9521f49ae121a17c0a41e5112249e6fa7f6a571245b1118de81fb86e7c1bc1ce"}, - {file = "bitarray-2.9.2-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:a6cc6545d6d76542aee3d18c1c9485fb7b9812b8df4ebe52c4535ec42081b48f"}, - {file = "bitarray-2.9.2-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:856bbe1616425f71c0df5ef2e8755e878d9504d5a531acba58ab4273c52c117a"}, - {file = "bitarray-2.9.2-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d4bba8042ea6ab331ade91bc435d81ad72fddb098e49108610b0ce7780c14e68"}, - {file = "bitarray-2.9.2-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:a035da89c959d98afc813e3c62f052690d67cfd55a36592f25d734b70de7d4b0"}, - {file = "bitarray-2.9.2-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:6d70b1579da7fb71be5a841a1f965d19aca0ef27f629cfc07d06b09aafd0a333"}, - {file = "bitarray-2.9.2-cp38-cp38-musllinux_1_1_ppc64le.whl", hash = "sha256:405b83bed28efaae6d86b6ab287c75712ead0adbfab2a1075a1b7ab47dad4d62"}, - {file = "bitarray-2.9.2-cp38-cp38-musllinux_1_1_s390x.whl", hash = "sha256:7eb8be687c50da0b397d5e0ab7ca200b5ebb639e79a9f5e285851d1944c94be9"}, - {file = "bitarray-2.9.2-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:eceb551dfeaf19c609003a69a0cf8264b0efd7abc3791a11dfabf4788daf0d19"}, - {file = "bitarray-2.9.2-cp38-cp38-win32.whl", hash = "sha256:bb198c6ed1edbcdaf3d1fa3c9c9d1cdb7e179a5134ef5ee660b53cdec43b34e7"}, - {file = "bitarray-2.9.2-cp38-cp38-win_amd64.whl", hash = "sha256:648d2f2685590b0103c67a937c2fb9e09bcc8dfb166f0c7c77bd341902a6f5b3"}, - {file = "bitarray-2.9.2-cp39-cp39-macosx_10_9_universal2.whl", hash = "sha256:ea816dc8f8e65841a8bbdd30e921edffeeb6f76efe6a1eb0da147b60d539d1cf"}, - {file = "bitarray-2.9.2-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:4d0e32530f941c41eddfc77600ec89b65184cb909c549336463a738fab3ed285"}, - {file = "bitarray-2.9.2-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:4a22266fb416a3b6c258bf7f83c9fe531ba0b755a56986a81ad69dc0f3bcc070"}, - {file = "bitarray-2.9.2-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:fc6d3e80dd8239850f2604833ff3168b28909c8a9357abfed95632cccd17e3e7"}, - {file = "bitarray-2.9.2-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:f135e804986b12bf14f2cd1eb86674c47dea86c4c5f0fa13c88978876b97ebe6"}, - {file = "bitarray-2.9.2-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:87580c7f7d14f7ec401eda7adac1e2a25e95153e9c339872c8ae61b3208819a1"}, - {file = "bitarray-2.9.2-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:64b433e26993127732ac7b66a7821b2537c3044355798de7c5fcb0af34b8296f"}, - {file = "bitarray-2.9.2-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:1e497c535f2a9b68c69d36631bf2dba243e05eb343b00b9c7bbdc8c601c6802d"}, - {file = "bitarray-2.9.2-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:e40b3cb9fa1edb4e0175d7c06345c49c7925fe93e39ef55ecb0bc40c906b0c09"}, - {file = "bitarray-2.9.2-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:f2f8692f95c9e377eb19ca519d30d1f884b02feb7e115f798de47570a359e43f"}, - {file = "bitarray-2.9.2-cp39-cp39-musllinux_1_1_ppc64le.whl", hash = "sha256:f0b84fc50b6dbeced4fa390688c07c10a73222810fb0e08392bd1a1b8259de36"}, - {file = "bitarray-2.9.2-cp39-cp39-musllinux_1_1_s390x.whl", hash = "sha256:d656ad38c942e38a470ddbce26b5020e08e1a7ea86b8fd413bb9024b5189993a"}, - {file = "bitarray-2.9.2-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:6ab0f1dbfe5070db98771a56aa14797595acd45a1af9eadfb193851a270e7996"}, - {file = "bitarray-2.9.2-cp39-cp39-win32.whl", hash = "sha256:0a99b23ac845a9ea3157782c97465e6ae026fe0c7c4c1ed1d88f759fd6ea52d9"}, - {file = "bitarray-2.9.2-cp39-cp39-win_amd64.whl", hash = "sha256:9bbcfc7c279e8d74b076e514e669b683f77b4a2a328585b3f16d4c5259c91222"}, - {file = "bitarray-2.9.2-pp310-pypy310_pp73-macosx_10_9_x86_64.whl", hash = "sha256:43847799461d8ba71deb4d97b47250c2c2fb66d82cd3cb8b4caf52bb97c03034"}, - {file = "bitarray-2.9.2-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f4f44381b0a4bdf64416082f4f0e7140377ae962c0ced6f983c6d7bbfc034040"}, - {file = "bitarray-2.9.2-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a484061616fb4b158b80789bd3cb511f399d2116525a8b29b6334c68abc2310f"}, - {file = "bitarray-2.9.2-pp310-pypy310_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:1ff9e38356cc803e06134cf8ae9758e836ccd1b793135ef3db53c7c5d71e93bc"}, - {file = "bitarray-2.9.2-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:b44105792fbdcfbda3e26ee88786790fda409da4c71f6c2b73888108cf8f062f"}, - {file = "bitarray-2.9.2-pp37-pypy37_pp73-macosx_10_9_x86_64.whl", hash = "sha256:7e913098de169c7fc890638ce5e171387363eb812579e637c44261460ac00aa2"}, - {file = "bitarray-2.9.2-pp37-pypy37_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:d6fe315355cdfe3ed22ef355b8bdc81a805ca4d0949d921576560e5b227a1112"}, - {file = "bitarray-2.9.2-pp37-pypy37_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f708e91fdbe443f3bec2df394ed42328fb9b0446dff5cb4199023ac6499e09fd"}, - {file = "bitarray-2.9.2-pp37-pypy37_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:5b7b09489b71f9f1f64c0fa0977e250ec24500767dab7383ba9912495849cadf"}, - {file = "bitarray-2.9.2-pp37-pypy37_pp73-win_amd64.whl", hash = "sha256:128cc3488176145b9b137fdcf54c1c201809bbb8dd30b260ee40afe915843b43"}, - {file = "bitarray-2.9.2-pp38-pypy38_pp73-macosx_10_9_x86_64.whl", hash = "sha256:21f21e7f56206be346bdbda2a6bdb2165a5e6a11821f88fd4911c5a6bbbdc7e2"}, - {file = "bitarray-2.9.2-pp38-pypy38_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:5f4dd3af86dd8a617eb6464622fb64ca86e61ce99b59b5c35d8cd33f9c30603d"}, - {file = "bitarray-2.9.2-pp38-pypy38_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6465de861aff7a2559f226b37982007417eab8c3557543879987f58b453519bd"}, - {file = "bitarray-2.9.2-pp38-pypy38_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:dbaf2bb71d6027152d603f1d5f31e0dfd5e50173d06f877bec484e5396d4594b"}, - {file = "bitarray-2.9.2-pp38-pypy38_pp73-win_amd64.whl", hash = "sha256:2f32948c86e0d230a296686db28191b67ed229756f84728847daa0c7ab7406e3"}, - {file = "bitarray-2.9.2-pp39-pypy39_pp73-macosx_10_9_x86_64.whl", hash = "sha256:be94e5a685e60f9d24532af8fe5c268002e9016fa80272a94727f435de3d1003"}, - {file = "bitarray-2.9.2-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:a5cc9381fd54f3c23ae1039f977bfd6d041a5c3c1518104f616643c3a5a73b15"}, - {file = "bitarray-2.9.2-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:cd926e8ae4d1ed1ac4a8f37212a62886292f692bc1739fde98013bf210c2d175"}, - {file = "bitarray-2.9.2-pp39-pypy39_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:461a3dafb9d5fda0bb3385dc507d78b1984b49da3fe4c6d56c869a54373b7008"}, - {file = "bitarray-2.9.2-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:393cb27fd859af5fd9c16eb26b1c59b17b390ff66b3ae5d0dd258270191baf13"}, - {file = "bitarray-2.9.2.tar.gz", hash = "sha256:a8f286a51a32323715d77755ed959f94bef13972e9a2fe71b609e40e6d27957e"}, -] - [[package]] name = "bleach" version = "6.1.0" @@ -733,10 +597,8 @@ files = [ [package.dependencies] colorama = {version = "*", markers = "os_name == \"nt\""} -importlib-metadata = {version = ">=4.6", markers = "python_full_version < \"3.10.2\""} packaging = ">=19.1" pyproject_hooks = "*" -tomli = {version = ">=1.1.0", markers = "python_version < \"3.11\""} [package.extras] docs = ["furo (>=2023.08.17)", "sphinx (>=7.0,<8.0)", "sphinx-argparse-cli (>=1.5)", "sphinx-autodoc-typehints (>=1.10)", "sphinx-issues (>=3.0.0)"] @@ -747,13 +609,13 @@ virtualenv = ["virtualenv (>=20.0.35)"] [[package]] name = "cachetools" -version = "5.4.0" +version = "5.5.0" description = "Extensible memoizing collections and decorators" optional = false python-versions = ">=3.7" files = [ - {file = "cachetools-5.4.0-py3-none-any.whl", hash = "sha256:3ae3b49a3d5e28a77a0be2b37dbcb89005058959cb2323858c2657c4a8cab474"}, - {file = "cachetools-5.4.0.tar.gz", hash = "sha256:b8adc2e7c07f105ced7bc56dbb6dfbe7c4a00acce20e2227b3f355be89bc6827"}, + {file = "cachetools-5.5.0-py3-none-any.whl", hash = "sha256:02134e8439cdc2ffb62023ce1debca2944c3f289d66bb17ead3ab3dede74b292"}, + {file = "cachetools-5.5.0.tar.gz", hash = "sha256:2cc24fb4cbe39633fb7badd9db9ca6295d766d9c2995f245725a46715d050f2a"}, ] [[package]] @@ -1323,18 +1185,18 @@ tests = ["pytest", "pytest-cov", "pytest-xdist"] [[package]] name = "dask" -version = "2024.8.0" +version = "2024.8.1" description = "Parallel PyData with Task Scheduling" optional = false -python-versions = ">=3.9" +python-versions = ">=3.10" files = [ - {file = "dask-2024.8.0-py3-none-any.whl", hash = "sha256:250ea3df30d4a25958290eec4f252850091c6cfaed82d098179c3b25bba18309"}, - {file = "dask-2024.8.0.tar.gz", hash = "sha256:f1fec39373d2f101bc045529ad4e9b30e34e6eb33b7aa0fa7073aec7b1bf9eee"}, + {file = "dask-2024.8.1-py3-none-any.whl", hash = "sha256:b8b58cba91dc9c057c8676dcc80b8bc321602b4dfd21529d33b03b55d428e2c3"}, + {file = "dask-2024.8.1.tar.gz", hash = "sha256:4254e43ac8c3affad2b22952f126b00a00f52c87caae91c068d8e395a4ad1a72"}, ] [package.dependencies] click = ">=8.1" -cloudpickle = ">=1.5.0" +cloudpickle = ">=3.0.0" dask-expr = {version = ">=1.1,<1.2", optional = true, markers = "extra == \"dataframe\""} fsspec = ">=2021.09.0" importlib-metadata = {version = ">=4.13.0", markers = "python_version < \"3.12\""} @@ -1350,22 +1212,22 @@ array = ["numpy (>=1.21)"] complete = ["dask[array,dataframe,diagnostics,distributed]", "lz4 (>=4.3.2)", "pyarrow (>=7.0)", "pyarrow-hotfix"] dataframe = ["dask-expr (>=1.1,<1.2)", "dask[array]", "pandas (>=2.0)"] diagnostics = ["bokeh (>=2.4.2)", "jinja2 (>=2.10.3)"] -distributed = ["distributed (==2024.8.0)"] +distributed = ["distributed (==2024.8.1)"] test = ["pandas[test]", "pre-commit", "pytest", "pytest-cov", "pytest-rerunfailures", "pytest-timeout", "pytest-xdist"] [[package]] name = "dask-expr" -version = "1.1.10" +version = "1.1.11" description = "High Level Expressions for Dask" optional = false -python-versions = ">=3.9" +python-versions = ">=3.10" files = [ - {file = "dask_expr-1.1.10-py3-none-any.whl", hash = "sha256:c6365c6fa6d3e386c5ee79bd20d4c89e566c0cf78fb6c762f74b2f04028935c6"}, - {file = "dask_expr-1.1.10.tar.gz", hash = "sha256:3d9ac7231f41ce7a109faaf855a60d89bd4f90d304452894178a114470164014"}, + {file = "dask_expr-1.1.11-py3-none-any.whl", hash = "sha256:b9222b3d430152e3af4a1777f66bcee88651f510876cb57c720107d123d9ba63"}, + {file = "dask_expr-1.1.11.tar.gz", hash = "sha256:275689c269f9c30dbaf9d8d7e9d3b5ac5438ea6db73fdbf95b3f4bfb1381bc5a"}, ] [package.dependencies] -dask = "2024.8.0" +dask = "2024.8.1" pandas = ">=2" pyarrow = ">=7.0.0" @@ -1530,26 +1392,6 @@ files = [ {file = "distro-1.9.0.tar.gz", hash = "sha256:2fa77c6fd8940f116ee1d6b94a2f90b13b5ea8d019b98bc8bafdcabcdd9bdbed"}, ] -[[package]] -name = "dnspython" -version = "2.6.1" -description = "DNS toolkit" -optional = false -python-versions = ">=3.8" -files = [ - {file = "dnspython-2.6.1-py3-none-any.whl", hash = "sha256:5ef3b9680161f6fa89daf8ad451b5f1a33b18ae8a1c6778cdf4b43f08c0a6e50"}, - {file = "dnspython-2.6.1.tar.gz", hash = "sha256:e8f0f9c23a7b7cb99ded64e6c3a6f3e701d78f50c55e002b839dea7225cff7cc"}, -] - -[package.extras] -dev = ["black (>=23.1.0)", "coverage (>=7.0)", "flake8 (>=7)", "mypy (>=1.8)", "pylint (>=3)", "pytest (>=7.4)", "pytest-cov (>=4.1.0)", "sphinx (>=7.2.0)", "twine (>=4.0.0)", "wheel (>=0.42.0)"] -dnssec = ["cryptography (>=41)"] -doh = ["h2 (>=4.1.0)", "httpcore (>=1.0.0)", "httpx (>=0.26.0)"] -doq = ["aioquic (>=0.9.25)"] -idna = ["idna (>=3.6)"] -trio = ["trio (>=0.23)"] -wmi = ["wmi (>=1.5.1)"] - [[package]] name = "docstring-parser" version = "0.16" @@ -1561,21 +1403,6 @@ files = [ {file = "docstring_parser-0.16.tar.gz", hash = "sha256:538beabd0af1e2db0146b6bd3caa526c35a34d61af9fd2887f3a8a27a739aa6e"}, ] -[[package]] -name = "email-validator" -version = "2.2.0" -description = "A robust email address syntax and deliverability validation library." -optional = false -python-versions = ">=3.8" -files = [ - {file = "email_validator-2.2.0-py3-none-any.whl", hash = "sha256:561977c2d73ce3611850a06fa56b414621e0c8faa9d66f2611407d87465da631"}, - {file = "email_validator-2.2.0.tar.gz", hash = "sha256:cb690f344c617a714f22e66ae771445a1ceb46821152df8e165c5f9a364582b7"}, -] - -[package.dependencies] -dnspython = ">=2.0.0" -idna = ">=2.0.0" - [[package]] name = "environs" version = "11.0.0" @@ -1596,20 +1423,6 @@ dev = ["environs[tests]", "pre-commit (>=3.5,<4.0)", "tox"] django = ["dj-database-url", "dj-email-url", "django-cache-url"] tests = ["environs[django]", "pytest"] -[[package]] -name = "exceptiongroup" -version = "1.2.2" -description = "Backport of PEP 654 (exception groups)" -optional = false -python-versions = ">=3.7" -files = [ - {file = "exceptiongroup-1.2.2-py3-none-any.whl", hash = "sha256:3111b9d131c238bec2f8f516e123e14ba243563fb135d3fe885990585aa7795b"}, - {file = "exceptiongroup-1.2.2.tar.gz", hash = "sha256:47c2edf7c6738fafb49fd34290706d1a1a2f4d1c6df275526b62cbb4aa5393cc"}, -] - -[package.extras] -test = ["pytest (>=6)"] - [[package]] name = "executing" version = "2.0.1" @@ -1626,47 +1439,23 @@ tests = ["asttokens (>=2.1.0)", "coverage", "coverage-enable-subprocess", "ipyth [[package]] name = "fastapi" -version = "0.111.1" +version = "0.110.3" description = "FastAPI framework, high performance, easy to learn, fast to code, ready for production" optional = false python-versions = ">=3.8" files = [ - {file = "fastapi-0.111.1-py3-none-any.whl", hash = "sha256:4f51cfa25d72f9fbc3280832e84b32494cf186f50158d364a8765aabf22587bf"}, - {file = "fastapi-0.111.1.tar.gz", hash = "sha256:ddd1ac34cb1f76c2e2d7f8545a4bcb5463bce4834e81abf0b189e0c359ab2413"}, + {file = "fastapi-0.110.3-py3-none-any.whl", hash = "sha256:fd7600612f755e4050beb74001310b5a7e1796d149c2ee363124abdfa0289d32"}, + {file = "fastapi-0.110.3.tar.gz", hash = "sha256:555700b0159379e94fdbfc6bb66a0f1c43f4cf7060f25239af3d84b63a656626"}, ] [package.dependencies] -email_validator = ">=2.0.0" -fastapi-cli = ">=0.0.2" -httpx = ">=0.23.0" -jinja2 = ">=2.11.2" pydantic = ">=1.7.4,<1.8 || >1.8,<1.8.1 || >1.8.1,<2.0.0 || >2.0.0,<2.0.1 || >2.0.1,<2.1.0 || >2.1.0,<3.0.0" -python-multipart = ">=0.0.7" starlette = ">=0.37.2,<0.38.0" typing-extensions = ">=4.8.0" -uvicorn = {version = ">=0.12.0", extras = ["standard"]} [package.extras] all = ["email_validator (>=2.0.0)", "httpx (>=0.23.0)", "itsdangerous (>=1.1.0)", "jinja2 (>=2.11.2)", "orjson (>=3.2.1)", "pydantic-extra-types (>=2.0.0)", "pydantic-settings (>=2.0.0)", "python-multipart (>=0.0.7)", "pyyaml (>=5.3.1)", "ujson (>=4.0.1,!=4.0.2,!=4.1.0,!=4.2.0,!=4.3.0,!=5.0.0,!=5.1.0)", "uvicorn[standard] (>=0.12.0)"] -[[package]] -name = "fastapi-cli" -version = "0.0.5" -description = "Run and manage FastAPI apps from the command line with FastAPI CLI. šŸš€" -optional = false -python-versions = ">=3.8" -files = [ - {file = "fastapi_cli-0.0.5-py3-none-any.whl", hash = "sha256:e94d847524648c748a5350673546bbf9bcaeb086b33c24f2e82e021436866a46"}, - {file = "fastapi_cli-0.0.5.tar.gz", hash = "sha256:d30e1239c6f46fcb95e606f02cdda59a1e2fa778a54b64686b3ff27f6211ff9f"}, -] - -[package.dependencies] -typer = ">=0.12.3" -uvicorn = {version = ">=0.15.0", extras = ["standard"]} - -[package.extras] -standard = ["uvicorn[standard] (>=0.15.0)"] - [[package]] name = "fastjsonschema" version = "2.20.0" @@ -2077,13 +1866,13 @@ grpcio-gcp = ["grpcio-gcp (>=0.2.2,<1.0.dev0)"] [[package]] name = "google-auth" -version = "2.33.0" +version = "2.34.0" description = "Google Authentication Library" optional = false python-versions = ">=3.7" files = [ - {file = "google_auth-2.33.0-py2.py3-none-any.whl", hash = "sha256:8eff47d0d4a34ab6265c50a106a3362de6a9975bb08998700e389f857e4d39df"}, - {file = "google_auth-2.33.0.tar.gz", hash = "sha256:d6a52342160d7290e334b4d47ba390767e4438ad0d45b7630774533e82655b95"}, + {file = "google_auth-2.34.0-py2.py3-none-any.whl", hash = "sha256:72fd4733b80b6d777dcde515628a9eb4a577339437012874ea286bca7261ee65"}, + {file = "google_auth-2.34.0.tar.gz", hash = "sha256:8eb87396435c19b20d32abd2f984e31c191a15284af72eb922f10e5bde9c04cc"}, ] [package.dependencies] @@ -2093,7 +1882,7 @@ rsa = ">=3.1.4,<5" [package.extras] aiohttp = ["aiohttp (>=3.6.2,<4.0.0.dev0)", "requests (>=2.20.0,<3.0.0.dev0)"] -enterprise-cert = ["cryptography (==36.0.2)", "pyopenssl (==22.0.0)"] +enterprise-cert = ["cryptography", "pyopenssl"] pyopenssl = ["cryptography (>=38.0.3)", "pyopenssl (>=20.0.0)"] reauth = ["pyu2f (>=0.1.5)"] requests = ["requests (>=2.20.0,<3.0.0.dev0)"] @@ -2130,13 +1919,13 @@ grpc = ["grpcio (>=1.44.0,<2.0.0.dev0)"] [[package]] name = "graphrag" -version = "0.3.0" +version = "0.3.1" description = "" optional = false python-versions = "<3.13,>=3.10" files = [ - {file = "graphrag-0.3.0-py3-none-any.whl", hash = "sha256:115d376601c131f9433e9e25a99557da7120b55245402916c31c968c28845741"}, - {file = "graphrag-0.3.0.tar.gz", hash = "sha256:57c8338776e94d6c3c1a263bf41cf2f6317329c0f41a8a3a3b050f7d16499462"}, + {file = "graphrag-0.3.1-py3-none-any.whl", hash = "sha256:1d5f96060caff23febaa1f6d7ee1087770b7b6d23a964628353c3a5c9066f202"}, + {file = "graphrag-0.3.1.tar.gz", hash = "sha256:c475edd209d4f019f4db54f155fff35a121e6781db1e0610066f7b5778472aa3"}, ] [package.dependencies] @@ -2144,7 +1933,7 @@ aiofiles = ">=24.1.0,<25.0.0" aiolimiter = ">=1.1.0,<2.0.0" azure-identity = ">=1.17.1,<2.0.0" azure-search-documents = ">=11.4.0,<12.0.0" -azure-storage-blob = ">=12.19.0,<13.0.0" +azure-storage-blob = ">=12.22.0,<13.0.0" datashaper = ">=0.0.49,<0.0.50" devtools = ">=0.12.2,<0.13.0" environs = ">=11.0.0,<12.0.0" @@ -2154,22 +1943,22 @@ json-repair = ">=0.26.0,<0.27.0" lancedb = ">=0.11.0,<0.12.0" nest-asyncio = {version = ">=1.6.0,<2.0.0", markers = "platform_system == \"Windows\""} networkx = ">=3,<4" -nltk = "3.8.1" +nltk = "3.9.1" numba = "0.60.0" numpy = ">=1.25.2,<2.0.0" openai = ">=1.37.1,<2.0.0" pyaml-env = ">=1.2.1,<2.0.0" pydantic = ">=2,<3" python-dotenv = ">=1.0.0,<2.0.0" -pyyaml = ">=6.0.1,<7.0.0" +pyyaml = ">=6.0.2,<7.0.0" rich = ">=13.6.0,<14.0.0" scipy = "1.12.0" swifter = ">=1.4.0,<2.0.0" tenacity = ">=9.0.0,<10.0.0" -textual = ">=0.74.0,<0.75.0" +textual = ">=0.76.0,<0.77.0" tiktoken = ">=0.7.0,<0.8.0" typing-extensions = ">=4.12.2,<5.0.0" -uvloop = {version = ">=0.19.0,<0.20.0", markers = "platform_system != \"Windows\""} +uvloop = {version = ">=0.20.0,<0.21.0", markers = "platform_system != \"Windows\""} [[package]] name = "graspologic" @@ -2316,54 +2105,6 @@ http2 = ["h2 (>=3,<5)"] socks = ["socksio (==1.*)"] trio = ["trio (>=0.22.0,<0.26.0)"] -[[package]] -name = "httptools" -version = "0.6.1" -description = "A collection of framework independent HTTP protocol utils." -optional = false -python-versions = ">=3.8.0" -files = [ - {file = "httptools-0.6.1-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:d2f6c3c4cb1948d912538217838f6e9960bc4a521d7f9b323b3da579cd14532f"}, - {file = "httptools-0.6.1-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:00d5d4b68a717765b1fabfd9ca755bd12bf44105eeb806c03d1962acd9b8e563"}, - {file = "httptools-0.6.1-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:639dc4f381a870c9ec860ce5c45921db50205a37cc3334e756269736ff0aac58"}, - {file = "httptools-0.6.1-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e57997ac7fb7ee43140cc03664de5f268813a481dff6245e0075925adc6aa185"}, - {file = "httptools-0.6.1-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:0ac5a0ae3d9f4fe004318d64b8a854edd85ab76cffbf7ef5e32920faef62f142"}, - {file = "httptools-0.6.1-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:3f30d3ce413088a98b9db71c60a6ada2001a08945cb42dd65a9a9fe228627658"}, - {file = "httptools-0.6.1-cp310-cp310-win_amd64.whl", hash = "sha256:1ed99a373e327f0107cb513b61820102ee4f3675656a37a50083eda05dc9541b"}, - {file = "httptools-0.6.1-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:7a7ea483c1a4485c71cb5f38be9db078f8b0e8b4c4dc0210f531cdd2ddac1ef1"}, - {file = "httptools-0.6.1-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:85ed077c995e942b6f1b07583e4eb0a8d324d418954fc6af913d36db7c05a5a0"}, - {file = "httptools-0.6.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8b0bb634338334385351a1600a73e558ce619af390c2b38386206ac6a27fecfc"}, - {file = "httptools-0.6.1-cp311-cp311-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:7d9ceb2c957320def533671fc9c715a80c47025139c8d1f3797477decbc6edd2"}, - {file = "httptools-0.6.1-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:4f0f8271c0a4db459f9dc807acd0eadd4839934a4b9b892f6f160e94da309837"}, - {file = "httptools-0.6.1-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:6a4f5ccead6d18ec072ac0b84420e95d27c1cdf5c9f1bc8fbd8daf86bd94f43d"}, - {file = "httptools-0.6.1-cp311-cp311-win_amd64.whl", hash = "sha256:5cceac09f164bcba55c0500a18fe3c47df29b62353198e4f37bbcc5d591172c3"}, - {file = "httptools-0.6.1-cp312-cp312-macosx_10_9_universal2.whl", hash = "sha256:75c8022dca7935cba14741a42744eee13ba05db00b27a4b940f0d646bd4d56d0"}, - {file = "httptools-0.6.1-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:48ed8129cd9a0d62cf4d1575fcf90fb37e3ff7d5654d3a5814eb3d55f36478c2"}, - {file = "httptools-0.6.1-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:6f58e335a1402fb5a650e271e8c2d03cfa7cea46ae124649346d17bd30d59c90"}, - {file = "httptools-0.6.1-cp312-cp312-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:93ad80d7176aa5788902f207a4e79885f0576134695dfb0fefc15b7a4648d503"}, - {file = "httptools-0.6.1-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:9bb68d3a085c2174c2477eb3ffe84ae9fb4fde8792edb7bcd09a1d8467e30a84"}, - {file = "httptools-0.6.1-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:b512aa728bc02354e5ac086ce76c3ce635b62f5fbc32ab7082b5e582d27867bb"}, - {file = "httptools-0.6.1-cp312-cp312-win_amd64.whl", hash = "sha256:97662ce7fb196c785344d00d638fc9ad69e18ee4bfb4000b35a52efe5adcc949"}, - {file = "httptools-0.6.1-cp38-cp38-macosx_10_9_universal2.whl", hash = "sha256:8e216a038d2d52ea13fdd9b9c9c7459fb80d78302b257828285eca1c773b99b3"}, - {file = "httptools-0.6.1-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:3e802e0b2378ade99cd666b5bffb8b2a7cc8f3d28988685dc300469ea8dd86cb"}, - {file = "httptools-0.6.1-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4bd3e488b447046e386a30f07af05f9b38d3d368d1f7b4d8f7e10af85393db97"}, - {file = "httptools-0.6.1-cp38-cp38-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fe467eb086d80217b7584e61313ebadc8d187a4d95bb62031b7bab4b205c3ba3"}, - {file = "httptools-0.6.1-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:3c3b214ce057c54675b00108ac42bacf2ab8f85c58e3f324a4e963bbc46424f4"}, - {file = "httptools-0.6.1-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:8ae5b97f690badd2ca27cbf668494ee1b6d34cf1c464271ef7bfa9ca6b83ffaf"}, - {file = "httptools-0.6.1-cp38-cp38-win_amd64.whl", hash = "sha256:405784577ba6540fa7d6ff49e37daf104e04f4b4ff2d1ac0469eaa6a20fde084"}, - {file = "httptools-0.6.1-cp39-cp39-macosx_10_9_universal2.whl", hash = "sha256:95fb92dd3649f9cb139e9c56604cc2d7c7bf0fc2e7c8d7fbd58f96e35eddd2a3"}, - {file = "httptools-0.6.1-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:dcbab042cc3ef272adc11220517278519adf8f53fd3056d0e68f0a6f891ba94e"}, - {file = "httptools-0.6.1-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:0cf2372e98406efb42e93bfe10f2948e467edfd792b015f1b4ecd897903d3e8d"}, - {file = "httptools-0.6.1-cp39-cp39-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:678fcbae74477a17d103b7cae78b74800d795d702083867ce160fc202104d0da"}, - {file = "httptools-0.6.1-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:e0b281cf5a125c35f7f6722b65d8542d2e57331be573e9e88bc8b0115c4a7a81"}, - {file = "httptools-0.6.1-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:95658c342529bba4e1d3d2b1a874db16c7cca435e8827422154c9da76ac4e13a"}, - {file = "httptools-0.6.1-cp39-cp39-win_amd64.whl", hash = "sha256:7ebaec1bf683e4bf5e9fbb49b8cc36da482033596a415b3e4ebab5a4c0d7ec5e"}, - {file = "httptools-0.6.1.tar.gz", hash = "sha256:c6e26c30455600b95d94b1b836085138e82f177351454ee841c148f93a9bad5a"}, -] - -[package.extras] -test = ["Cython (>=0.29.24,<0.30.0)"] - [[package]] name = "httpx" version = "0.27.0" @@ -2451,18 +2192,22 @@ test = ["flufl.flake8", "importlib-resources (>=1.3)", "jaraco.test (>=5.4)", "p [[package]] name = "importlib-resources" -version = "6.4.2" +version = "6.4.4" description = "Read resources from Python packages" optional = false python-versions = ">=3.8" files = [ - {file = "importlib_resources-6.4.2-py3-none-any.whl", hash = "sha256:8bba8c54a8a3afaa1419910845fa26ebd706dc716dd208d9b158b4b6966f5c5c"}, - {file = "importlib_resources-6.4.2.tar.gz", hash = "sha256:6cbfbefc449cc6e2095dd184691b7a12a04f40bc75dd4c55d31c34f174cdf57a"}, + {file = "importlib_resources-6.4.4-py3-none-any.whl", hash = "sha256:dda242603d1c9cd836c3368b1174ed74cb4049ecd209e7a1a0104620c18c5c11"}, + {file = "importlib_resources-6.4.4.tar.gz", hash = "sha256:20600c8b7361938dc0bb2d5ec0297802e575df486f5a544fa414da65e13721f7"}, ] [package.extras] +check = ["pytest-checkdocs (>=2.4)", "pytest-ruff (>=0.2.1)"] +cover = ["pytest-cov"] doc = ["furo", "jaraco.packaging (>=9.3)", "jaraco.tidelift (>=1.4)", "rst.linker (>=1.9)", "sphinx (>=3.5)", "sphinx-lint"] -test = ["jaraco.test (>=5.4)", "pytest (>=6,!=8.1.*)", "pytest-checkdocs (>=2.4)", "pytest-cov", "pytest-enabler (>=2.2)", "pytest-mypy", "pytest-ruff (>=0.2.1)", "zipp (>=3.17)"] +enabler = ["pytest-enabler (>=2.2)"] +test = ["jaraco.test (>=5.4)", "pytest (>=6,!=8.1.*)", "zipp (>=3.17)"] +type = ["pytest-mypy"] [[package]] name = "iniconfig" @@ -2522,7 +2267,6 @@ files = [ [package.dependencies] colorama = {version = "*", markers = "sys_platform == \"win32\""} decorator = "*" -exceptiongroup = {version = "*", markers = "python_version < \"3.11\""} jedi = ">=0.16" matplotlib-inline = "*" pexpect = {version = ">4.3", markers = "sys_platform != \"win32\" and sys_platform != \"emscripten\""} @@ -2780,84 +2524,106 @@ files = [ [[package]] name = "jq" -version = "1.7.0" +version = "1.8.0" description = "jq is a lightweight and flexible JSON processor." optional = false -python-versions = ">=3.5" +python-versions = ">=3.6" files = [ - {file = "jq-1.7.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:d8fae014fa8b2704322a5baa39c112176d9acb71e22ebdb8e21c1c864ecff654"}, - {file = "jq-1.7.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:40fe068d1fdf2c712885b69be90ddb3e61bca3e4346ab3994641a4fbbeb7be82"}, - {file = "jq-1.7.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8ec105a0057f2f922d195e1d75d4b0ae41c4b38655ead04d1a3a47988fcb1939"}, - {file = "jq-1.7.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:38e2041ca578275334eff9e1d913ae386210345e5ae71cd9c16e3f208dc81deb"}, - {file = "jq-1.7.0-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:ce1df1b6fffeeeb265d4ea3397e9875ab170ba5a7af6b7997c2fd755934df065"}, - {file = "jq-1.7.0-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:05ebdaa868f068967d9e7cbf76e59e61fbdafa565dbc3579c387fb1f248592bb"}, - {file = "jq-1.7.0-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:b3f916cb812fcd26bb1b006634d9c0eff240090196ca0ebb5d229b344f624e53"}, - {file = "jq-1.7.0-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:9ad7749a16a16bafd6cebafd5e40990b641b4b6b7b661326864677effc44a500"}, - {file = "jq-1.7.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:2e99ea17b708f55e8bed2f4f68c022119184b17eb15987b384db12e8b6702bd5"}, - {file = "jq-1.7.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:76735cd19de65c15964d330adbc2c84add8e55dea35ebfe17b9acf88a06a7d57"}, - {file = "jq-1.7.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:c6b841ddd9089429fc0621d07d1c34ff24f7d6a6245c10125b82806f61e36ae8"}, - {file = "jq-1.7.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:1d6b1fc2515b7be92195d50b68f82329cc0250c7fbca790b887d74902ba33870"}, - {file = "jq-1.7.0-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:eb6546a57a3ceeed41961be2f1417b4e7a5b3170cca7bb82f5974d2ba9acaab6"}, - {file = "jq-1.7.0-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:3427ad0f377f188953958e36b76167c8d11b8c8c61575c22deafa4aba58d601f"}, - {file = "jq-1.7.0-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:79b9603219fa5082df97d265d71c426613286bd0e5378a8739ce39056fa1e2dc"}, - {file = "jq-1.7.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:a2981a24765a747163e0daa23648372b72a006e727895b95d032632aa51094bd"}, - {file = "jq-1.7.0-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:a0cc15b2ed511a1a8784c7c7dc07781e28d84a65934062de52487578732e0514"}, - {file = "jq-1.7.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:90032c2c4e710157d333d166818ede8b9c8ef0f697e59c9427304edc47146f3d"}, - {file = "jq-1.7.0-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:e715d5f0bdfc0be0ff33cd0a3f6f51f8bc5ad464fab737e2048a1b46b45bb582"}, - {file = "jq-1.7.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:76cc5a1ca3a540a5753dbd592f701c1ec7c9cc256becba604490283c055f3f1c"}, - {file = "jq-1.7.0-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:293b6e8e4b652d96fdeae7dd5ffb1644199d8b6fc1f95d528c16451925c0482e"}, - {file = "jq-1.7.0-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:f103868b8902d4ee7f643248bdd7a2de9f9396e4b262f42745b9f624c834d07a"}, - {file = "jq-1.7.0-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:e9c5ccfa3cf65f92b60c5805ef725f7cd799f2dc16e8601c6e8f12f38a9f48f3"}, - {file = "jq-1.7.0-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:0ca25608d51fdbf8bd5c682b433e1cb9f497155a7c1ea5901524df099f1ceff3"}, - {file = "jq-1.7.0-cp36-cp36m-macosx_10_9_x86_64.whl", hash = "sha256:6a2d34d962ce2da5136dab2664fc7efad9f71024d0dc328702f2dc70b4e2735c"}, - {file = "jq-1.7.0-cp36-cp36m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:757e8c4cb0cb1175f0aaa227f0a26e4765ba5da04d0bc875b0bd933eff6bd0a0"}, - {file = "jq-1.7.0-cp36-cp36m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:2d097098a628171b87961fb0400117ac340b1eb40cbbee2e58208c4254c23c20"}, - {file = "jq-1.7.0-cp36-cp36m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:45bc842806d71bd5839c190a88fd071ac5a0a8a1dd601e83228494a19f14559c"}, - {file = "jq-1.7.0-cp36-cp36m-musllinux_1_1_aarch64.whl", hash = "sha256:f0629743417f8709305d1f77d3929493912efdc3fd1cce3a7fcc76b81bc6b82d"}, - {file = "jq-1.7.0-cp36-cp36m-musllinux_1_1_i686.whl", hash = "sha256:9b9a49e8b14d3a368011ed1412c8c3e193a7135d5eb4310d77ee643470112b47"}, - {file = "jq-1.7.0-cp36-cp36m-musllinux_1_1_x86_64.whl", hash = "sha256:a10e3f88b6d2bbb4c47b368f919ec7b648196bf9c60a5cc921d04239d68240c2"}, - {file = "jq-1.7.0-cp37-cp37m-macosx_10_9_x86_64.whl", hash = "sha256:aa85b47effb4152e1cf1120607f475a1c11395d072323ff23e8bb59ce6752713"}, - {file = "jq-1.7.0-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:9413f67ea28037e37ccf8951f9f0b380f31d79162f33e216faa6bd0d8eca0dc7"}, - {file = "jq-1.7.0-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:3daf3b3443c4e871c23ac1e698eb70d1225b46a4ac79c73968234adcd70f3ed8"}, - {file = "jq-1.7.0-cp37-cp37m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:dbe03f95ab02dc045691c3b5c7da8d8c2128e60450fb2124ea8b49034c74f158"}, - {file = "jq-1.7.0-cp37-cp37m-musllinux_1_1_aarch64.whl", hash = "sha256:a6b2e9f4e63644a30726c58c25d80015f9b83325b125615a46e10d4439b9dc99"}, - {file = "jq-1.7.0-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:9fffcffc8e56585223878edd7c5d719eb8547281d64af2bac43911f1bb9e7029"}, - {file = "jq-1.7.0-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:95d4bcd5a999ce0aaadaadcaca967989f0efc96c1097a81746b21b6126cf7aaf"}, - {file = "jq-1.7.0-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:0137445eb67c43eb0eb46933aff7e8afbbd6c5aaf8574efd5df536dc9d177d1d"}, - {file = "jq-1.7.0-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4ee0e9307b6d4fe89a8556a92c1db65e0d66218bcc13fdeb92a09645a55ff87a"}, - {file = "jq-1.7.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:4e0f95cecb690df66f23a8d76c746d2ed15671de3f6101140e3fe2b98b97e0a8"}, - {file = "jq-1.7.0-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:95e472aa54efe418d3627dcd2a369ac0b21e1a5e352550144fd5f0c40585a5b7"}, - {file = "jq-1.7.0-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:4be2a2b56fa139f3235cdb8422ea16eccdd48d62bf91d9fac10761cd55d26c84"}, - {file = "jq-1.7.0-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:7db8260ecb57827bb3fb6f44d4a6f0db0570ded990eee95a5fd3ac9ba14f60d7"}, - {file = "jq-1.7.0-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:fdbb7ff2dfce2cc0f421f498dcb64176997bd9d9e6cab474e59577e7bff3090d"}, - {file = "jq-1.7.0-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:396bef4b4c9c1ebe3e0e04e287bc79a861b991e12db45681c398d3906ee85468"}, - {file = "jq-1.7.0-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:18d8a81c6e241585a0bf748903082d65c4eaa6ba80248f507e5cebda36e05c6c"}, - {file = "jq-1.7.0-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:ade00a39990fdfe0acc7d2a900e3e5e6b11a71eb5289954ff0df31ac0afae25b"}, - {file = "jq-1.7.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:0c777e88f3cce496c17f5c3bdbc7d74ff12b5cbdaea30f3a374f3cc92e5bba8d"}, - {file = "jq-1.7.0-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:79957008c67d8f1d9134cd0e01044bff5d795f7e94db9532a9fe9212e1f88a77"}, - {file = "jq-1.7.0-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:2bc5cb77dd12e861296cfa69587aa6797ccfee4f5f3aa571b02f0273ab1efec1"}, - {file = "jq-1.7.0-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:8e10a5937aab9c383632ab151f73d43dc0c4be99f62221a7044988dc8ddd4bdc"}, - {file = "jq-1.7.0-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:1e6e13e0f8d3204aefe861159160116e822c90bae773a3ccdd4d9e79a06e086e"}, - {file = "jq-1.7.0-pp310-pypy310_pp73-macosx_10_13_x86_64.whl", hash = "sha256:0cdbd32463ef632b0b4ca6dab434e2387342bc5c895b411ec6b2a14bbf4b2c12"}, - {file = "jq-1.7.0-pp310-pypy310_pp73-macosx_11_0_arm64.whl", hash = "sha256:558a5c6b4430e05fa59c4b5631c0d3fc0f163100390c03edc1993663f59d8a9b"}, - {file = "jq-1.7.0-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4bbf77138cdd8d306bf335d998525a0477e4cb6f00eb6f361288f5b82274e84c"}, - {file = "jq-1.7.0-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a2e6919481ff43754ae9b17a98c877995d5e1346be114c71cd0dfd8ff7d0cd60"}, - {file = "jq-1.7.0-pp310-pypy310_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:0b0584ff33b2a9cc021edec325af4e0fa9fbd54cce80c1f7b8e0ba4cf2d75508"}, - {file = "jq-1.7.0-pp37-pypy37_pp73-macosx_10_9_x86_64.whl", hash = "sha256:a6e7259880ab7e75e845fb4d56c6d18922c68789d25d7cdbb6f433d9e714613a"}, - {file = "jq-1.7.0-pp37-pypy37_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2d472cdd0bcb3d47c87b00ff841edff41c79fe4422523c4a7c8bf913fb950f7f"}, - {file = "jq-1.7.0-pp37-pypy37_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:2a3430de179f8a7b0baf5675d5ee400f97344085d79f190a90fc0c7df990cbcc"}, - {file = "jq-1.7.0-pp37-pypy37_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:acbb375bdb2a44f1a643123b8ec57563bb5542673f0399799ab5662ce90bf4a5"}, - {file = "jq-1.7.0-pp38-pypy38_pp73-macosx_10_9_x86_64.whl", hash = "sha256:39a0c71ed2f1ec0462d54678333f1b14d9f25fd62a9f46df140d68552f79d204"}, - {file = "jq-1.7.0-pp38-pypy38_pp73-macosx_11_0_arm64.whl", hash = "sha256:306c1e3ba531d7dc3284e128689f0b75409a4e8e8a3bdac2c51cc26f2d3cca58"}, - {file = "jq-1.7.0-pp38-pypy38_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:88b8b0cc838c7387dc5e8c45b192c7504acd0510514658d2d5cd1716fcf15fe3"}, - {file = "jq-1.7.0-pp38-pypy38_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:9c75e16e542f4abaae25727b9fc4eeaf69cb07122be8a2a7672d02feb3a1cc9a"}, - {file = "jq-1.7.0-pp38-pypy38_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:b4828ac689a67fd9c021796bcacd95811bab806939dd6316eb0c2d3de016c584"}, - {file = "jq-1.7.0-pp39-pypy39_pp73-macosx_10_13_x86_64.whl", hash = "sha256:c94f95b27720d2db7f1039fdd371f70bc0cac8e204cbfd0626176d7b8a3053d6"}, - {file = "jq-1.7.0-pp39-pypy39_pp73-macosx_11_0_arm64.whl", hash = "sha256:d5ff445fc9b1eb4623a914e04bea9511e654e9143cde82b039383af4f7dc36f2"}, - {file = "jq-1.7.0-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:07e369ff021fad38a29d6a7a3fc24f7d313e9a239b15ce4eefaffee637466400"}, - {file = "jq-1.7.0-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:553dfbf674069cb20533d7d74cd8a9d7982bab8e4a5b473fde105d99278df09f"}, - {file = "jq-1.7.0-pp39-pypy39_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:e9fbc76f6fec66e5e58cc84f20a5de80addd3c64ad87a748f5c5f6b4ef01bc8c"}, - {file = "jq-1.7.0.tar.gz", hash = "sha256:f460d1f2c3791617e4fb339fa24efbdbebe672b02c861f057358553642047040"}, + {file = "jq-1.8.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:628848f92a0f24f5ca50c879d271555a63bf28746c1efd3571ee49e9a357b602"}, + {file = "jq-1.8.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:d375b0f372df24087fd0688ef85fef43a44a3e382a82afcc0cdfdfe59e59d313"}, + {file = "jq-1.8.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:cd0c30af5257ae0dccd27c5140726e24108a472e56dce8767b918905adfd9c99"}, + {file = "jq-1.8.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:59bda8b62453967a32f418562309d0ffe0da73227e8c5800334ee0b515c5d2e2"}, + {file = "jq-1.8.0-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:05e2c0a8944a3ff93de6353d60ed69fa85b155c08d6776ab20d4429197f50050"}, + {file = "jq-1.8.0-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:2526368e5658eaeb47984b551e7178a0216cc8c5fdd6dd343964574cae513c89"}, + {file = "jq-1.8.0-cp310-cp310-musllinux_1_2_i686.whl", hash = "sha256:881be44d8f804a97a1e37dc6360bf2deab43768d7fbb31cfb22ca8050dd6aed3"}, + {file = "jq-1.8.0-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:f057322a572fe2cf0cb9ea068dd4eec237bc15490e0944cd979aeb23b20db3ac"}, + {file = "jq-1.8.0-cp310-cp310-win32.whl", hash = "sha256:aaf6e17cd9bf26c076a9a6ff0b4bfac66fdaa37ed9e215683de58d657cc75f29"}, + {file = "jq-1.8.0-cp310-cp310-win_amd64.whl", hash = "sha256:53c87ef5491e484cdfb740303ccfc141af1d23275750569f539d4981524f4251"}, + {file = "jq-1.8.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:f8441fe181af789a05b742930d095ee61fc251fdd2b975c68e359ac7e85a4c2d"}, + {file = "jq-1.8.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:8e687ef4b360e7436c3b5f15ee25f2570bcbcadccb940ebbc80ebe4b05b91ee2"}, + {file = "jq-1.8.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:aaf862d1bc1d0095aef0efc76f8cef0da7ab996f2b9d34c5067e48427a069ea3"}, + {file = "jq-1.8.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:190fd2bf92b7abec3090a1f68db40cd001178e84c42754f75253ee1f9c17dfdf"}, + {file = "jq-1.8.0-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:3ecba9f181e7810a336a520f32df998e6ecc9fdebac80c6a636e402baa939e79"}, + {file = "jq-1.8.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:8b6322f647f9e1d7be7f6e8203106f4ff1b7c0e07c9023607c7414e1dc098b67"}, + {file = "jq-1.8.0-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:7bed3b9cc53d72383fc558cfe03345735e7532d1733a5ed3c2196f1eec1c26d7"}, + {file = "jq-1.8.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:1a01261e4df11d3a0fe42fece73bb458d2e4a33b481d67e5e817acec8b0e923d"}, + {file = "jq-1.8.0-cp311-cp311-win32.whl", hash = "sha256:52cac82de5608f9174d22a1a805d61ba47ea182b10a934135904648c618ebe34"}, + {file = "jq-1.8.0-cp311-cp311-win_amd64.whl", hash = "sha256:745d0f9786bd89eb9bff054ac08ce0e61877d28931857585e244e8674ac3727e"}, + {file = "jq-1.8.0-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:14f5988ae3604ebfdba2da398f9bd941bb3a72144a2831cfec2bc22bd23d5563"}, + {file = "jq-1.8.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:f8903b66fac9f46de72b3a2f69bfa3c638a7a8d52610d1894df87ef0a9e4d2d3"}, + {file = "jq-1.8.0-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:cccda466f5722fa9be789099ce253bfc177e49f9a981cb7f5b6369ea37041104"}, + {file = "jq-1.8.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:95f57649e84a09b334eeb80d22ecc96ff7b31701f3f818ef14cb8bb162c84863"}, + {file = "jq-1.8.0-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:7453731008eb7671725222781eb7bc5ed96e80fc9a652d177cb982276d3e08b4"}, + {file = "jq-1.8.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:917812663613fc0542117bbe7ec43c8733b0c6bb174db6be06a15fc612de3b70"}, + {file = "jq-1.8.0-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:ec9e4db978237470e9d65f747eb459f4ffee576c9c9f8ca92ab32d5687a46e4a"}, + {file = "jq-1.8.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:f9f2548c83473bbe88a32a0735cb949a5d01804f8d411efae5342b5d23be8a2f"}, + {file = "jq-1.8.0-cp312-cp312-win32.whl", hash = "sha256:e3da3538549d5bdc84e6282555be4ba5a50c3792db7d8d72d064cc6f48a2f722"}, + {file = "jq-1.8.0-cp312-cp312-win_amd64.whl", hash = "sha256:049ba2978e61e593299edc6dd57b9cefd680272740ad1d4703f8784f5fab644d"}, + {file = "jq-1.8.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:76aea6161c4d975230e85735c0214c386e66035e96cfc4fd69159e87f46c09d4"}, + {file = "jq-1.8.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:0c24a5f9e3807e277e19f305c8bcd0665b8b89251b053903f611969657680722"}, + {file = "jq-1.8.0-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:eb484525dd801583ebd695d02f9165445a4d1b2fb560b187e6fc654911f0600e"}, + {file = "jq-1.8.0-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ddd9abdf0c1b30be1bf853d8c52187c96a51b2cbc05f40c43a37bf6a9b956807"}, + {file = "jq-1.8.0-cp313-cp313-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:2c7464d9b88c74a7119b53f4bbf88028d07a9de9a1a279e45209b763b89d6582"}, + {file = "jq-1.8.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:b99761e8ec2cedb9906df4ceae33f467a377621019ef40a9a275689ac3577456"}, + {file = "jq-1.8.0-cp313-cp313-musllinux_1_2_i686.whl", hash = "sha256:1be1638f9d5f38c83440fb9626d8f78905ed5d70e926e3a664d3de1198e1ef79"}, + {file = "jq-1.8.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:2d7e82d58bf3afe373afb3a01f866e473bbd34f38377a2f216c6222ec028eeea"}, + {file = "jq-1.8.0-cp313-cp313-win32.whl", hash = "sha256:96cb0bb35d55b19b910b12aba3d72e333ad6348a703494c7738cc4664e4410f0"}, + {file = "jq-1.8.0-cp313-cp313-win_amd64.whl", hash = "sha256:53e60a87657efc365a5d9ccfea2b536cddc1ffab190e823f8645ad933b272d51"}, + {file = "jq-1.8.0-cp36-cp36m-macosx_10_9_x86_64.whl", hash = "sha256:a5c3a9e8fa0eedb600626719630ec3dc6018379075e10733d88899f147d26528"}, + {file = "jq-1.8.0-cp36-cp36m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8c86f0f9d496c6d51caa9597dae6bdb11b27c45cee820a3db3bb61303359d217"}, + {file = "jq-1.8.0-cp36-cp36m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:990116b7fcf3f37dd89cb12bbc5a09f85ca1fee368945501096470c71f1851de"}, + {file = "jq-1.8.0-cp36-cp36m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:6f3cefb4b9dde8edeb2af0b108c8df1942e9352e83406491959e7dc145ccf20a"}, + {file = "jq-1.8.0-cp36-cp36m-musllinux_1_2_aarch64.whl", hash = "sha256:aa70883dcbddb06bcb1510f5025f2709268d91ddbe23f31b297ffc73fec1ed3d"}, + {file = "jq-1.8.0-cp36-cp36m-musllinux_1_2_i686.whl", hash = "sha256:d302a987dabf2fbf7297bf32b1ed16e1232e85734d412c94abfa95bf7e4bf689"}, + {file = "jq-1.8.0-cp36-cp36m-musllinux_1_2_x86_64.whl", hash = "sha256:69ee5e888bb7e6549f51f1148e78ae31e584297f496a68e258af1baca81d8785"}, + {file = "jq-1.8.0-cp37-cp37m-macosx_10_9_x86_64.whl", hash = "sha256:a2528b279717339d3ca87fd263f1f38a66f79dabd3882fc8d73d68dd06db4260"}, + {file = "jq-1.8.0-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:88a0450b9e4b55f5e7c8fce00d6db7f5826334193f599daa27b8c44d6d5a3fd0"}, + {file = "jq-1.8.0-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f7262695b12039bdf66677b189bf0eb01c0d5b9b5ba905f1509984a1dbbc6505"}, + {file = "jq-1.8.0-cp37-cp37m-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:76d66b230ec285c5a211899d6f75bb7ac22fcf5c14f420df534d8d4544f9aa97"}, + {file = "jq-1.8.0-cp37-cp37m-musllinux_1_2_aarch64.whl", hash = "sha256:c757f4a2a08e98826875176bbc872da4913f5d64f8d3e27f3cf05fcf64cf6b92"}, + {file = "jq-1.8.0-cp37-cp37m-musllinux_1_2_i686.whl", hash = "sha256:88416952dc41192736e8884e1465e2121401a39e8d2fdaf4190d88d70102e4ad"}, + {file = "jq-1.8.0-cp37-cp37m-musllinux_1_2_x86_64.whl", hash = "sha256:c73ce1672863e84b506865da342cb014c3af795e7670d267d8d61d061d4b59f7"}, + {file = "jq-1.8.0-cp37-cp37m-win32.whl", hash = "sha256:5af7413dd18e7a448364a78a31739e0687d5fa00751e6d6acbbb5dde06e105b4"}, + {file = "jq-1.8.0-cp37-cp37m-win_amd64.whl", hash = "sha256:3d2186049890f9e08a90f351c4ac34ac449123f78e729994d501ceb02add9829"}, + {file = "jq-1.8.0-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:d24b8aa12ad7f465262ab0aeb0a7fa43df814ad3e50253ce454af40769da69d8"}, + {file = "jq-1.8.0-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:9d055dc15f76c8d7f5c917d2bc4540582e21f1783f12149758751e4b760888d7"}, + {file = "jq-1.8.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:58e092e54e1b543352b1dfd0fbfac233c46b999b2dfdba2b604536ad777566ae"}, + {file = "jq-1.8.0-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:415f8112a6f80a459d885e500f69ee510ca319fcc12e679ce5bf02c900f09118"}, + {file = "jq-1.8.0-cp38-cp38-musllinux_1_2_aarch64.whl", hash = "sha256:1ddb23e26d7606040ee4ec8b3845dc34eb56d4a905f9d0dcad398e269786135d"}, + {file = "jq-1.8.0-cp38-cp38-musllinux_1_2_i686.whl", hash = "sha256:12a78b891113193de32fcfd822f82e2065beeb5479c8b39dc5312c35cac77a6e"}, + {file = "jq-1.8.0-cp38-cp38-musllinux_1_2_x86_64.whl", hash = "sha256:c226c2b15c51efd0cbdd0470a2711dad3ead6a079052fbd4405e08f5e009449c"}, + {file = "jq-1.8.0-cp38-cp38-win32.whl", hash = "sha256:9fc84851be38bac073ab4a8dcd9966edef3f2a5bc69f6f85f7c5c1baf5d9bf6a"}, + {file = "jq-1.8.0-cp38-cp38-win_amd64.whl", hash = "sha256:4d86a99a72cef84ccd94905b1c10d8d4b9542e05cc94d2ae713a0f10ea1b52f6"}, + {file = "jq-1.8.0-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:1db31a68bf563a67579cc3c634f1676884ad29c9a43ce5d3858e989deafdc215"}, + {file = "jq-1.8.0-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:e3297cc2d2dfc9e7c92e119ba91ef30c7493b59d3528b9486b0c4dd819ff8d28"}, + {file = "jq-1.8.0-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:361c5089b912e558932202e4b10a8dd3d986ae8eb08ff39d5d419eb30db1df08"}, + {file = "jq-1.8.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:39b409e27cebc7d0323d966309ced5b46496a348475443f8ef38906f45bff7ff"}, + {file = "jq-1.8.0-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d9719abb172d40c01e4f42214db8b05cac4118ad6c6265f8b57ef14b86eedeaf"}, + {file = "jq-1.8.0-cp39-cp39-musllinux_1_2_aarch64.whl", hash = "sha256:2f59a71c51417e9fe10cad76be2557635da1febcef925ab948b66471b8d72232"}, + {file = "jq-1.8.0-cp39-cp39-musllinux_1_2_i686.whl", hash = "sha256:0627d3dd67e73a754d9950f57d964a711658b1258ddd135cf8c1e845c5efb49e"}, + {file = "jq-1.8.0-cp39-cp39-musllinux_1_2_x86_64.whl", hash = "sha256:cb90b618855e89c95396ab6aab09a1334df81fa4fbd1c0e939cab34a4993465f"}, + {file = "jq-1.8.0-cp39-cp39-win32.whl", hash = "sha256:0aca31819d07377f9036ebdeb57c1ccb73e10c502badb5c8601572ccb4fa96e2"}, + {file = "jq-1.8.0-cp39-cp39-win_amd64.whl", hash = "sha256:c30249ae524ac54ef73dabda6f2b5be077debb7847050e18f91d915f6b6f0208"}, + {file = "jq-1.8.0-pp310-pypy310_pp73-macosx_10_13_x86_64.whl", hash = "sha256:e14aa012606470d1a21fdc39835b8eef395f7ea143c720940a48156de94752e9"}, + {file = "jq-1.8.0-pp310-pypy310_pp73-macosx_11_0_arm64.whl", hash = "sha256:353db01bbb964eff9e39c8966e7c123cbdad1ff59cc3bee773a7a2034e2b843b"}, + {file = "jq-1.8.0-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:325480cba94f504b282f42912a16b32d94dd1e6347cf3a367ec3c97fe1dd1b3a"}, + {file = "jq-1.8.0-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:b4a79e94c83ebde789ff54e609f19b1923b2f57b2bd17ccb4953713577d4c3dc"}, + {file = "jq-1.8.0-pp310-pypy310_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:dc7ebcc1037c8a82db30aff9177f17379bcc91734def09548e939326717fd82d"}, + {file = "jq-1.8.0-pp37-pypy37_pp73-macosx_10_9_x86_64.whl", hash = "sha256:8776c33c0b69ae8de50cde9a338ef69cc0db4122ff6763a18c5532d6a5eb86f4"}, + {file = "jq-1.8.0-pp37-pypy37_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8ef1d313fec8820648879e7c167a3162ebbd711a5429a07427ac3f9c48ab8415"}, + {file = "jq-1.8.0-pp37-pypy37_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:21a0a58482e8e6be03d7b280365d40c3c4c1cf36d3ba58f98b1e351c42d6483d"}, + {file = "jq-1.8.0-pp37-pypy37_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:04b2f964c5ad6ac3013b052099bfc0cf8bd2cf80cedca670153687681c013641"}, + {file = "jq-1.8.0-pp38-pypy38_pp73-macosx_10_9_x86_64.whl", hash = "sha256:b8004ba63facc31eecd09e86e02cf5e1a5cf491cf8856f30d70c3fa96b8c74f9"}, + {file = "jq-1.8.0-pp38-pypy38_pp73-macosx_11_0_arm64.whl", hash = "sha256:01cc78dbf56e75debc9d46ba85ef61ac37472e8d629d01dbea79d4c09ef6dd51"}, + {file = "jq-1.8.0-pp38-pypy38_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:c049335a00f502b213376f67f651adc86cbe636468107190d08a4b1f77754fb5"}, + {file = "jq-1.8.0-pp38-pypy38_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e433560001d59cfa3551f276d7b6c6943fa6b6e05019b2071ccb41c9b2dc0c3c"}, + {file = "jq-1.8.0-pp38-pypy38_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:dafdae5ccc2e75df69b32518805c8d9d7aa97d0388cd6dc89b83d7bd516ea2eb"}, + {file = "jq-1.8.0-pp39-pypy39_pp73-macosx_10_13_x86_64.whl", hash = "sha256:4b32381ebdf1b5870e32a90737aa7d91824eaf5c78586973845de80802eb035a"}, + {file = "jq-1.8.0-pp39-pypy39_pp73-macosx_11_0_arm64.whl", hash = "sha256:c3c2ba32ea87d6f15a1e83af71d5af12c82814dac21809a3995fb8e5763968ff"}, + {file = "jq-1.8.0-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:850540641b8e0ecebc8763ff660811bcf5834468fd2572ee3ef8d79dea67050d"}, + {file = "jq-1.8.0-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:57e8bcdf2a744ee702542f3441fb2583db7f28602a6a2ff4a6d7009a11fafc86"}, + {file = "jq-1.8.0-pp39-pypy39_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:b6f61ea995c11dd1877f0452d12aad2b2a617b965e31033d2e62db4a530e87c0"}, + {file = "jq-1.8.0.tar.gz", hash = "sha256:53141eebca4bf8b4f2da5e44271a8a3694220dfd22d2b4b2cfb4816b2b6c9057"}, ] [[package]] @@ -3217,7 +2983,6 @@ jupyterlab-server = ">=2.27.1,<3" notebook-shim = ">=0.2" packaging = "*" setuptools = ">=40.1.0" -tomli = {version = ">=1.2.2", markers = "python_version < \"3.11\""} tornado = ">=6.2.0" traitlets = "*" @@ -3804,13 +3569,13 @@ files = [ [[package]] name = "marshmallow" -version = "3.21.3" +version = "3.22.0" description = "A lightweight library for converting complex datatypes to and from native Python datatypes." optional = false python-versions = ">=3.8" files = [ - {file = "marshmallow-3.21.3-py3-none-any.whl", hash = "sha256:86ce7fb914aa865001a4b2092c4c2872d13bc347f3d42673272cabfdbad386f1"}, - {file = "marshmallow-3.21.3.tar.gz", hash = "sha256:4f57c5e050a54d66361e826f94fba213eb10b67b2fdb02c3e0343ce207ba1662"}, + {file = "marshmallow-3.22.0-py3-none-any.whl", hash = "sha256:71a2dce49ef901c3f97ed296ae5051135fd3febd2bf43afe0ae9a82143a494d9"}, + {file = "marshmallow-3.22.0.tar.gz", hash = "sha256:4972f529104a220bb8637d595aa4c9762afbe7f7a77d82dc58c1615d70c5823e"}, ] [package.dependencies] @@ -3818,7 +3583,7 @@ packaging = ">=17.0" [package.extras] dev = ["marshmallow[tests]", "pre-commit (>=3.5,<4.0)", "tox"] -docs = ["alabaster (==0.7.16)", "autodocsumm (==0.2.12)", "sphinx (==7.3.7)", "sphinx-issues (==4.1.0)", "sphinx-version-warning (==1.1.2)"] +docs = ["alabaster (==1.0.0)", "autodocsumm (==0.2.13)", "sphinx (==8.0.2)", "sphinx-issues (==4.1.0)", "sphinx-version-warning (==1.1.2)"] tests = ["pytest", "pytz", "simplejson"] [[package]] @@ -4160,13 +3925,13 @@ test = ["pytest (>=7.2)", "pytest-cov (>=4.0)"] [[package]] name = "nltk" -version = "3.8.1" +version = "3.9.1" description = "Natural Language Toolkit" optional = false -python-versions = ">=3.7" +python-versions = ">=3.8" files = [ - {file = "nltk-3.8.1-py3-none-any.whl", hash = "sha256:fd5c9109f976fa86bcadba8f91e47f5e9293bd034474752e92a520f81c93dda5"}, - {file = "nltk-3.8.1.zip", hash = "sha256:1834da3d0682cba4f2cede2f9aad6b0fafb6461ba451db0efb6f9c39798d64d3"}, + {file = "nltk-3.9.1-py3-none-any.whl", hash = "sha256:4fa26829c5b00715afe3061398a8989dc643b92ce7dd93fb4585a70930d168a1"}, + {file = "nltk-3.9.1.tar.gz", hash = "sha256:87d127bd3de4bd89a4f81265e5fa59cb1b199b27440175370f7417d2bc7ae868"}, ] [package.dependencies] @@ -4331,13 +4096,13 @@ signedtoken = ["cryptography (>=3.0.0)", "pyjwt (>=2.0.0,<3)"] [[package]] name = "openai" -version = "1.40.6" +version = "1.42.0" description = "The official Python library for the openai API" optional = false python-versions = ">=3.7.1" files = [ - {file = "openai-1.40.6-py3-none-any.whl", hash = "sha256:b36372124a779381a420a34dd96f762baa748b6bdfaf83a6b9f2745f72ccc1c5"}, - {file = "openai-1.40.6.tar.gz", hash = "sha256:2239232bcb7f4bd4ce8e02544b5769618582411cf399816d96686d1b6c1e5c8d"}, + {file = "openai-1.42.0-py3-none-any.whl", hash = "sha256:dc91e0307033a4f94931e5d03cc3b29b9717014ad5e73f9f2051b6cb5eda4d80"}, + {file = "openai-1.42.0.tar.gz", hash = "sha256:c9d31853b4e0bc2dc8bd08003b462a006035655a701471695d0bfdc08529cde3"}, ] [package.dependencies] @@ -4505,13 +4270,13 @@ files = [ [[package]] name = "packaging" -version = "24.1" +version = "24.0" description = "Core utilities for Python packages" optional = false -python-versions = ">=3.8" +python-versions = ">=3.7" files = [ - {file = "packaging-24.1-py3-none-any.whl", hash = "sha256:5b8f2217dbdbd2f7f384c41c628544e6d52f2d0f53c6d0c3ea61aa5d1d7ff124"}, - {file = "packaging-24.1.tar.gz", hash = "sha256:026ed72c8ed3fcce5bf8950572258698927fd1dbda10a5e981cdf0ac37f4f002"}, + {file = "packaging-24.0-py3-none-any.whl", hash = "sha256:2ddfb553fdf02fb784c234c7ba6ccc288296ceabec964ad2eae3777778130bc5"}, + {file = "packaging-24.0.tar.gz", hash = "sha256:eb82c5e3e56209074766e6885bb04b8c38a0c015d0a30036ebe7ece34c9989e9"}, ] [[package]] @@ -4554,7 +4319,6 @@ files = [ [package.dependencies] numpy = [ - {version = ">=1.22.4", markers = "python_version < \"3.11\""}, {version = ">=1.23.2", markers = "python_version == \"3.11\""}, {version = ">=1.26.0", markers = "python_version >= \"3.12\""}, ] @@ -4681,23 +4445,6 @@ six = "*" [package.extras] test = ["pytest", "pytest-cov", "scipy"] -[[package]] -name = "pdfreader" -version = "0.1.15" -description = "Pythonic API for parsing PDF files" -optional = false -python-versions = ">=3.4" -files = [ - {file = "pdfreader-0.1.15-py3-none-any.whl", hash = "sha256:bfd0d29c0d70a81b767b42b6959dd588a8290086c8c72a828739c1e2bda07eba"}, - {file = "pdfreader-0.1.15.tar.gz", hash = "sha256:2ee1252cc5f21a2f8cadb458decd85c1313271abb5bac1e4363a3e0e17e2dd87"}, -] - -[package.dependencies] -bitarray = ">=1.1.0" -pillow = ">=7.1.0" -pycryptodome = ">=3.9.9" -python-dateutil = ">=2.8.1" - [[package]] name = "pexpect" version = "4.9.0" @@ -5314,47 +5061,6 @@ files = [ {file = "pycparser-2.22.tar.gz", hash = "sha256:491c8be9c040f5390f5bf44a5b07752bd07f56edf992381b05c701439eec10f6"}, ] -[[package]] -name = "pycryptodome" -version = "3.20.0" -description = "Cryptographic library for Python" -optional = false -python-versions = ">=2.7, !=3.0.*, !=3.1.*, !=3.2.*, !=3.3.*, !=3.4.*" -files = [ - {file = "pycryptodome-3.20.0-cp27-cp27m-macosx_10_9_x86_64.whl", hash = "sha256:f0e6d631bae3f231d3634f91ae4da7a960f7ff87f2865b2d2b831af1dfb04e9a"}, - {file = "pycryptodome-3.20.0-cp27-cp27m-manylinux2010_i686.whl", hash = "sha256:baee115a9ba6c5d2709a1e88ffe62b73ecc044852a925dcb67713a288c4ec70f"}, - {file = "pycryptodome-3.20.0-cp27-cp27m-manylinux2010_x86_64.whl", hash = "sha256:417a276aaa9cb3be91f9014e9d18d10e840a7a9b9a9be64a42f553c5b50b4d1d"}, - {file = "pycryptodome-3.20.0-cp27-cp27m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2a1250b7ea809f752b68e3e6f3fd946b5939a52eaeea18c73bdab53e9ba3c2dd"}, - {file = "pycryptodome-3.20.0-cp27-cp27m-musllinux_1_1_aarch64.whl", hash = "sha256:d5954acfe9e00bc83ed9f5cb082ed22c592fbbef86dc48b907238be64ead5c33"}, - {file = "pycryptodome-3.20.0-cp27-cp27m-win32.whl", hash = "sha256:06d6de87c19f967f03b4cf9b34e538ef46e99a337e9a61a77dbe44b2cbcf0690"}, - {file = "pycryptodome-3.20.0-cp27-cp27m-win_amd64.whl", hash = "sha256:ec0bb1188c1d13426039af8ffcb4dbe3aad1d7680c35a62d8eaf2a529b5d3d4f"}, - {file = "pycryptodome-3.20.0-cp27-cp27mu-manylinux2010_i686.whl", hash = "sha256:5601c934c498cd267640b57569e73793cb9a83506f7c73a8ec57a516f5b0b091"}, - {file = "pycryptodome-3.20.0-cp27-cp27mu-manylinux2010_x86_64.whl", hash = "sha256:d29daa681517f4bc318cd8a23af87e1f2a7bad2fe361e8aa29c77d652a065de4"}, - {file = "pycryptodome-3.20.0-cp27-cp27mu-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:3427d9e5310af6680678f4cce149f54e0bb4af60101c7f2c16fdf878b39ccccc"}, - {file = "pycryptodome-3.20.0-cp27-cp27mu-musllinux_1_1_aarch64.whl", hash = "sha256:3cd3ef3aee1079ae44afaeee13393cf68b1058f70576b11439483e34f93cf818"}, - {file = "pycryptodome-3.20.0-cp35-abi3-macosx_10_9_universal2.whl", hash = "sha256:ac1c7c0624a862f2e53438a15c9259d1655325fc2ec4392e66dc46cdae24d044"}, - {file = "pycryptodome-3.20.0-cp35-abi3-macosx_10_9_x86_64.whl", hash = "sha256:76658f0d942051d12a9bd08ca1b6b34fd762a8ee4240984f7c06ddfb55eaf15a"}, - {file = "pycryptodome-3.20.0-cp35-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f35d6cee81fa145333137009d9c8ba90951d7d77b67c79cbe5f03c7eb74d8fe2"}, - {file = "pycryptodome-3.20.0-cp35-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:76cb39afede7055127e35a444c1c041d2e8d2f1f9c121ecef573757ba4cd2c3c"}, - {file = "pycryptodome-3.20.0-cp35-abi3-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:49a4c4dc60b78ec41d2afa392491d788c2e06edf48580fbfb0dd0f828af49d25"}, - {file = "pycryptodome-3.20.0-cp35-abi3-musllinux_1_1_aarch64.whl", hash = "sha256:fb3b87461fa35afa19c971b0a2b7456a7b1db7b4eba9a8424666104925b78128"}, - {file = "pycryptodome-3.20.0-cp35-abi3-musllinux_1_1_i686.whl", hash = "sha256:acc2614e2e5346a4a4eab6e199203034924313626f9620b7b4b38e9ad74b7e0c"}, - {file = "pycryptodome-3.20.0-cp35-abi3-musllinux_1_1_x86_64.whl", hash = "sha256:210ba1b647837bfc42dd5a813cdecb5b86193ae11a3f5d972b9a0ae2c7e9e4b4"}, - {file = "pycryptodome-3.20.0-cp35-abi3-win32.whl", hash = "sha256:8d6b98d0d83d21fb757a182d52940d028564efe8147baa9ce0f38d057104ae72"}, - {file = "pycryptodome-3.20.0-cp35-abi3-win_amd64.whl", hash = "sha256:9b3ae153c89a480a0ec402e23db8d8d84a3833b65fa4b15b81b83be9d637aab9"}, - {file = "pycryptodome-3.20.0-pp27-pypy_73-manylinux2010_x86_64.whl", hash = "sha256:4401564ebf37dfde45d096974c7a159b52eeabd9969135f0426907db367a652a"}, - {file = "pycryptodome-3.20.0-pp27-pypy_73-win32.whl", hash = "sha256:ec1f93feb3bb93380ab0ebf8b859e8e5678c0f010d2d78367cf6bc30bfeb148e"}, - {file = "pycryptodome-3.20.0-pp310-pypy310_pp73-macosx_10_9_x86_64.whl", hash = "sha256:acae12b9ede49f38eb0ef76fdec2df2e94aad85ae46ec85be3648a57f0a7db04"}, - {file = "pycryptodome-3.20.0-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f47888542a0633baff535a04726948e876bf1ed880fddb7c10a736fa99146ab3"}, - {file = "pycryptodome-3.20.0-pp310-pypy310_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:6e0e4a987d38cfc2e71b4a1b591bae4891eeabe5fa0f56154f576e26287bfdea"}, - {file = "pycryptodome-3.20.0-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:c18b381553638414b38705f07d1ef0a7cf301bc78a5f9bc17a957eb19446834b"}, - {file = "pycryptodome-3.20.0-pp39-pypy39_pp73-macosx_10_9_x86_64.whl", hash = "sha256:a60fedd2b37b4cb11ccb5d0399efe26db9e0dd149016c1cc6c8161974ceac2d6"}, - {file = "pycryptodome-3.20.0-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:405002eafad114a2f9a930f5db65feef7b53c4784495dd8758069b89baf68eab"}, - {file = "pycryptodome-3.20.0-pp39-pypy39_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:2ab6ab0cb755154ad14e507d1df72de9897e99fd2d4922851a276ccc14f4f1a5"}, - {file = "pycryptodome-3.20.0-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:acf6e43fa75aca2d33e93409f2dafe386fe051818ee79ee8a3e21de9caa2ac9e"}, - {file = "pycryptodome-3.20.0.tar.gz", hash = "sha256:09609209ed7de61c2b560cc5c8c4fbf892f8b15b1faf7e4cbffac97db1fffda7"}, -] - [[package]] name = "pydantic" version = "2.8.2" @@ -5675,22 +5381,22 @@ files = [ diagrams = ["jinja2", "railroad-diagrams"] [[package]] -name = "pypdf2" -version = "3.0.1" +name = "pypdf" +version = "4.3.1" description = "A pure-python PDF library capable of splitting, merging, cropping, and transforming PDF files" optional = false python-versions = ">=3.6" files = [ - {file = "PyPDF2-3.0.1.tar.gz", hash = "sha256:a74408f69ba6271f71b9352ef4ed03dc53a31aa404d29b5d31f53bfecfee1440"}, - {file = "pypdf2-3.0.1-py3-none-any.whl", hash = "sha256:d16e4205cfee272fbdc0568b68d82be796540b1537508cef59388f839c191928"}, + {file = "pypdf-4.3.1-py3-none-any.whl", hash = "sha256:64b31da97eda0771ef22edb1bfecd5deee4b72c3d1736b7df2689805076d6418"}, + {file = "pypdf-4.3.1.tar.gz", hash = "sha256:b2f37fe9a3030aa97ca86067a56ba3f9d3565f9a791b305c7355d8392c30d91b"}, ] [package.extras] -crypto = ["PyCryptodome"] -dev = ["black", "flit", "pip-tools", "pre-commit (<2.18.0)", "pytest-cov", "wheel"] +crypto = ["PyCryptodome", "cryptography"] +dev = ["black", "flit", "pip-tools", "pre-commit (<2.18.0)", "pytest-cov", "pytest-socket", "pytest-timeout", "pytest-xdist", "wheel"] docs = ["myst_parser", "sphinx", "sphinx_rtd_theme"] -full = ["Pillow", "PyCryptodome"] -image = ["Pillow"] +full = ["Pillow (>=8.0.0)", "PyCryptodome", "cryptography"] +image = ["Pillow (>=8.0.0)"] [[package]] name = "pyproject-hooks" @@ -5716,13 +5422,13 @@ files = [ [[package]] name = "pyright" -version = "1.1.376" +version = "1.1.377" description = "Command line wrapper for pyright" optional = false python-versions = ">=3.7" files = [ - {file = "pyright-1.1.376-py3-none-any.whl", hash = "sha256:0f2473b12c15c46b3207f0eec224c3cea2bdc07cd45dd4a037687cbbca0fbeff"}, - {file = "pyright-1.1.376.tar.gz", hash = "sha256:bffd63b197cd0810395bb3245c06b01f95a85ddf6bfa0e5644ed69c841e954dd"}, + {file = "pyright-1.1.377-py3-none-any.whl", hash = "sha256:af0dd2b6b636c383a6569a083f8c5a8748ae4dcde5df7914b3f3f267e14dd162"}, + {file = "pyright-1.1.377.tar.gz", hash = "sha256:aabc30fedce0ded34baa0c49b24f10e68f4bfc8f68ae7f3d175c4b0f256b4fcf"}, ] [package.dependencies] @@ -5757,11 +5463,9 @@ files = [ [package.dependencies] colorama = {version = "*", markers = "sys_platform == \"win32\""} -exceptiongroup = {version = ">=1.0.0rc8", markers = "python_version < \"3.11\""} iniconfig = "*" packaging = "*" pluggy = ">=1.5,<2" -tomli = {version = ">=1", markers = "python_version < \"3.11\""} [package.extras] dev = ["argcomplete", "attrs (>=19.2)", "hypothesis (>=3.56)", "mock", "pygments (>=2.7.2)", "requests", "setuptools", "xmlschema"] @@ -5837,6 +5541,17 @@ files = [ {file = "python_json_logger-2.0.7-py3-none-any.whl", hash = "sha256:f380b826a991ebbe3de4d897aeec42760035ac760345e57b812938dc8b35e2bd"}, ] +[[package]] +name = "python-magic" +version = "0.4.27" +description = "File type identification using libmagic" +optional = false +python-versions = ">=2.7, !=3.0.*, !=3.1.*, !=3.2.*, !=3.3.*, !=3.4.*" +files = [ + {file = "python-magic-0.4.27.tar.gz", hash = "sha256:c1ba14b08e4a5f5c31a302b7721239695b2f0f058d125bd5ce1ee36b9d9d3c3b"}, + {file = "python_magic-0.4.27-py2.py3-none-any.whl", hash = "sha256:c212960ad306f700aa0d01e5d7a325d20548ff97eb9920dcd29513174f0294d3"}, +] + [[package]] name = "python-multipart" version = "0.0.9" @@ -5975,120 +5690,120 @@ files = [ [[package]] name = "pyzmq" -version = "26.1.0" +version = "26.1.1" description = "Python bindings for 0MQ" optional = false python-versions = ">=3.7" files = [ - {file = "pyzmq-26.1.0-cp310-cp310-macosx_10_15_universal2.whl", hash = "sha256:263cf1e36862310bf5becfbc488e18d5d698941858860c5a8c079d1511b3b18e"}, - {file = "pyzmq-26.1.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:d5c8b17f6e8f29138678834cf8518049e740385eb2dbf736e8f07fc6587ec682"}, - {file = "pyzmq-26.1.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:75a95c2358fcfdef3374cb8baf57f1064d73246d55e41683aaffb6cfe6862917"}, - {file = "pyzmq-26.1.0-cp310-cp310-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:f99de52b8fbdb2a8f5301ae5fc0f9e6b3ba30d1d5fc0421956967edcc6914242"}, - {file = "pyzmq-26.1.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:7bcbfbab4e1895d58ab7da1b5ce9a327764f0366911ba5b95406c9104bceacb0"}, - {file = "pyzmq-26.1.0-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:77ce6a332c7e362cb59b63f5edf730e83590d0ab4e59c2aa5bd79419a42e3449"}, - {file = "pyzmq-26.1.0-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:ba0a31d00e8616149a5ab440d058ec2da621e05d744914774c4dde6837e1f545"}, - {file = "pyzmq-26.1.0-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:8b88641384e84a258b740801cd4dbc45c75f148ee674bec3149999adda4a8598"}, - {file = "pyzmq-26.1.0-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:2fa76ebcebe555cce90f16246edc3ad83ab65bb7b3d4ce408cf6bc67740c4f88"}, - {file = "pyzmq-26.1.0-cp310-cp310-win32.whl", hash = "sha256:fbf558551cf415586e91160d69ca6416f3fce0b86175b64e4293644a7416b81b"}, - {file = "pyzmq-26.1.0-cp310-cp310-win_amd64.whl", hash = "sha256:a7b8aab50e5a288c9724d260feae25eda69582be84e97c012c80e1a5e7e03fb2"}, - {file = "pyzmq-26.1.0-cp310-cp310-win_arm64.whl", hash = "sha256:08f74904cb066e1178c1ec706dfdb5c6c680cd7a8ed9efebeac923d84c1f13b1"}, - {file = "pyzmq-26.1.0-cp311-cp311-macosx_10_15_universal2.whl", hash = "sha256:46d6800b45015f96b9d92ece229d92f2aef137d82906577d55fadeb9cf5fcb71"}, - {file = "pyzmq-26.1.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:5bc2431167adc50ba42ea3e5e5f5cd70d93e18ab7b2f95e724dd8e1bd2c38120"}, - {file = "pyzmq-26.1.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b3bb34bebaa1b78e562931a1687ff663d298013f78f972a534f36c523311a84d"}, - {file = "pyzmq-26.1.0-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:bd3f6329340cef1c7ba9611bd038f2d523cea79f09f9c8f6b0553caba59ec562"}, - {file = "pyzmq-26.1.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:471880c4c14e5a056a96cd224f5e71211997d40b4bf5e9fdded55dafab1f98f2"}, - {file = "pyzmq-26.1.0-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:ce6f2b66799971cbae5d6547acefa7231458289e0ad481d0be0740535da38d8b"}, - {file = "pyzmq-26.1.0-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:0a1f6ea5b1d6cdbb8cfa0536f0d470f12b4b41ad83625012e575f0e3ecfe97f0"}, - {file = "pyzmq-26.1.0-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:b45e6445ac95ecb7d728604bae6538f40ccf4449b132b5428c09918523abc96d"}, - {file = "pyzmq-26.1.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:94c4262626424683feea0f3c34951d39d49d354722db2745c42aa6bb50ecd93b"}, - {file = "pyzmq-26.1.0-cp311-cp311-win32.whl", hash = "sha256:a0f0ab9df66eb34d58205913f4540e2ad17a175b05d81b0b7197bc57d000e829"}, - {file = "pyzmq-26.1.0-cp311-cp311-win_amd64.whl", hash = "sha256:8efb782f5a6c450589dbab4cb0f66f3a9026286333fe8f3a084399149af52f29"}, - {file = "pyzmq-26.1.0-cp311-cp311-win_arm64.whl", hash = "sha256:f133d05aaf623519f45e16ab77526e1e70d4e1308e084c2fb4cedb1a0c764bbb"}, - {file = "pyzmq-26.1.0-cp312-cp312-macosx_10_15_universal2.whl", hash = "sha256:3d3146b1c3dcc8a1539e7cc094700b2be1e605a76f7c8f0979b6d3bde5ad4072"}, - {file = "pyzmq-26.1.0-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:d9270fbf038bf34ffca4855bcda6e082e2c7f906b9eb8d9a8ce82691166060f7"}, - {file = "pyzmq-26.1.0-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:995301f6740a421afc863a713fe62c0aaf564708d4aa057dfdf0f0f56525294b"}, - {file = "pyzmq-26.1.0-cp312-cp312-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:e7eca8b89e56fb8c6c26dd3e09bd41b24789022acf1cf13358e96f1cafd8cae3"}, - {file = "pyzmq-26.1.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:90d4feb2e83dfe9ace6374a847e98ee9d1246ebadcc0cb765482e272c34e5820"}, - {file = "pyzmq-26.1.0-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:d4fafc2eb5d83f4647331267808c7e0c5722c25a729a614dc2b90479cafa78bd"}, - {file = "pyzmq-26.1.0-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:58c33dc0e185dd97a9ac0288b3188d1be12b756eda67490e6ed6a75cf9491d79"}, - {file = "pyzmq-26.1.0-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:68a0a1d83d33d8367ddddb3e6bb4afbb0f92bd1dac2c72cd5e5ddc86bdafd3eb"}, - {file = "pyzmq-26.1.0-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:2ae7c57e22ad881af78075e0cea10a4c778e67234adc65c404391b417a4dda83"}, - {file = "pyzmq-26.1.0-cp312-cp312-win32.whl", hash = "sha256:347e84fc88cc4cb646597f6d3a7ea0998f887ee8dc31c08587e9c3fd7b5ccef3"}, - {file = "pyzmq-26.1.0-cp312-cp312-win_amd64.whl", hash = "sha256:9f136a6e964830230912f75b5a116a21fe8e34128dcfd82285aa0ef07cb2c7bd"}, - {file = "pyzmq-26.1.0-cp312-cp312-win_arm64.whl", hash = "sha256:a4b7a989c8f5a72ab1b2bbfa58105578753ae77b71ba33e7383a31ff75a504c4"}, - {file = "pyzmq-26.1.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:d416f2088ac8f12daacffbc2e8918ef4d6be8568e9d7155c83b7cebed49d2322"}, - {file = "pyzmq-26.1.0-cp313-cp313-macosx_10_15_universal2.whl", hash = "sha256:ecb6c88d7946166d783a635efc89f9a1ff11c33d680a20df9657b6902a1d133b"}, - {file = "pyzmq-26.1.0-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:471312a7375571857a089342beccc1a63584315188560c7c0da7e0a23afd8a5c"}, - {file = "pyzmq-26.1.0-cp313-cp313-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:0e6cea102ffa16b737d11932c426f1dc14b5938cf7bc12e17269559c458ac334"}, - {file = "pyzmq-26.1.0-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ec7248673ffc7104b54e4957cee38b2f3075a13442348c8d651777bf41aa45ee"}, - {file = "pyzmq-26.1.0-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:0614aed6f87d550b5cecb03d795f4ddbb1544b78d02a4bd5eecf644ec98a39f6"}, - {file = "pyzmq-26.1.0-cp313-cp313-musllinux_1_1_aarch64.whl", hash = "sha256:e8746ce968be22a8a1801bf4a23e565f9687088580c3ed07af5846580dd97f76"}, - {file = "pyzmq-26.1.0-cp313-cp313-musllinux_1_1_i686.whl", hash = "sha256:7688653574392d2eaeef75ddcd0b2de5b232d8730af29af56c5adf1df9ef8d6f"}, - {file = "pyzmq-26.1.0-cp313-cp313-musllinux_1_1_x86_64.whl", hash = "sha256:8d4dac7d97f15c653a5fedcafa82626bd6cee1450ccdaf84ffed7ea14f2b07a4"}, - {file = "pyzmq-26.1.0-cp313-cp313-win32.whl", hash = "sha256:ccb42ca0a4a46232d716779421bbebbcad23c08d37c980f02cc3a6bd115ad277"}, - {file = "pyzmq-26.1.0-cp313-cp313-win_amd64.whl", hash = "sha256:e1e5d0a25aea8b691a00d6b54b28ac514c8cc0d8646d05f7ca6cb64b97358250"}, - {file = "pyzmq-26.1.0-cp313-cp313-win_arm64.whl", hash = "sha256:fc82269d24860cfa859b676d18850cbb8e312dcd7eada09e7d5b007e2f3d9eb1"}, - {file = "pyzmq-26.1.0-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:416ac51cabd54f587995c2b05421324700b22e98d3d0aa2cfaec985524d16f1d"}, - {file = "pyzmq-26.1.0-cp313-cp313t-macosx_10_15_universal2.whl", hash = "sha256:ff832cce719edd11266ca32bc74a626b814fff236824aa1aeaad399b69fe6eae"}, - {file = "pyzmq-26.1.0-cp313-cp313t-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:393daac1bcf81b2a23e696b7b638eedc965e9e3d2112961a072b6cd8179ad2eb"}, - {file = "pyzmq-26.1.0-cp313-cp313t-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:9869fa984c8670c8ab899a719eb7b516860a29bc26300a84d24d8c1b71eae3ec"}, - {file = "pyzmq-26.1.0-cp313-cp313t-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:3b3b8e36fd4c32c0825b4461372949ecd1585d326802b1321f8b6dc1d7e9318c"}, - {file = "pyzmq-26.1.0-cp313-cp313t-manylinux_2_28_x86_64.whl", hash = "sha256:3ee647d84b83509b7271457bb428cc347037f437ead4b0b6e43b5eba35fec0aa"}, - {file = "pyzmq-26.1.0-cp313-cp313t-musllinux_1_1_aarch64.whl", hash = "sha256:45cb1a70eb00405ce3893041099655265fabcd9c4e1e50c330026e82257892c1"}, - {file = "pyzmq-26.1.0-cp313-cp313t-musllinux_1_1_i686.whl", hash = "sha256:5cca7b4adb86d7470e0fc96037771981d740f0b4cb99776d5cb59cd0e6684a73"}, - {file = "pyzmq-26.1.0-cp313-cp313t-musllinux_1_1_x86_64.whl", hash = "sha256:91d1a20bdaf3b25f3173ff44e54b1cfbc05f94c9e8133314eb2962a89e05d6e3"}, - {file = "pyzmq-26.1.0-cp37-cp37m-macosx_10_9_x86_64.whl", hash = "sha256:c0665d85535192098420428c779361b8823d3d7ec4848c6af3abb93bc5c915bf"}, - {file = "pyzmq-26.1.0-cp37-cp37m-manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:96d7c1d35ee4a495df56c50c83df7af1c9688cce2e9e0edffdbf50889c167595"}, - {file = "pyzmq-26.1.0-cp37-cp37m-manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:b281b5ff5fcc9dcbfe941ac5c7fcd4b6c065adad12d850f95c9d6f23c2652384"}, - {file = "pyzmq-26.1.0-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:5384c527a9a004445c5074f1e20db83086c8ff1682a626676229aafd9cf9f7d1"}, - {file = "pyzmq-26.1.0-cp37-cp37m-musllinux_1_1_aarch64.whl", hash = "sha256:754c99a9840839375ee251b38ac5964c0f369306eddb56804a073b6efdc0cd88"}, - {file = "pyzmq-26.1.0-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:9bdfcb74b469b592972ed881bad57d22e2c0acc89f5e8c146782d0d90fb9f4bf"}, - {file = "pyzmq-26.1.0-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:bd13f0231f4788db619347b971ca5f319c5b7ebee151afc7c14632068c6261d3"}, - {file = "pyzmq-26.1.0-cp37-cp37m-win32.whl", hash = "sha256:c5668dac86a869349828db5fc928ee3f58d450dce2c85607067d581f745e4fb1"}, - {file = "pyzmq-26.1.0-cp37-cp37m-win_amd64.whl", hash = "sha256:ad875277844cfaeca7fe299ddf8c8d8bfe271c3dc1caf14d454faa5cdbf2fa7a"}, - {file = "pyzmq-26.1.0-cp38-cp38-macosx_10_15_universal2.whl", hash = "sha256:65c6e03cc0222eaf6aad57ff4ecc0a070451e23232bb48db4322cc45602cede0"}, - {file = "pyzmq-26.1.0-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:038ae4ffb63e3991f386e7fda85a9baab7d6617fe85b74a8f9cab190d73adb2b"}, - {file = "pyzmq-26.1.0-cp38-cp38-manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:bdeb2c61611293f64ac1073f4bf6723b67d291905308a7de9bb2ca87464e3273"}, - {file = "pyzmq-26.1.0-cp38-cp38-manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:61dfa5ee9d7df297c859ac82b1226d8fefaf9c5113dc25c2c00ecad6feeeb04f"}, - {file = "pyzmq-26.1.0-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f3292d384537b9918010769b82ab3e79fca8b23d74f56fc69a679106a3e2c2cf"}, - {file = "pyzmq-26.1.0-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:f9499c70c19ff0fbe1007043acb5ad15c1dec7d8e84ab429bca8c87138e8f85c"}, - {file = "pyzmq-26.1.0-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:d3dd5523ed258ad58fed7e364c92a9360d1af8a9371e0822bd0146bdf017ef4c"}, - {file = "pyzmq-26.1.0-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:baba2fd199b098c5544ef2536b2499d2e2155392973ad32687024bd8572a7d1c"}, - {file = "pyzmq-26.1.0-cp38-cp38-win32.whl", hash = "sha256:ddbb2b386128d8eca92bd9ca74e80f73fe263bcca7aa419f5b4cbc1661e19741"}, - {file = "pyzmq-26.1.0-cp38-cp38-win_amd64.whl", hash = "sha256:79e45a4096ec8388cdeb04a9fa5e9371583bcb826964d55b8b66cbffe7b33c86"}, - {file = "pyzmq-26.1.0-cp39-cp39-macosx_10_15_universal2.whl", hash = "sha256:add52c78a12196bc0fda2de087ba6c876ea677cbda2e3eba63546b26e8bf177b"}, - {file = "pyzmq-26.1.0-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:98c03bd7f3339ff47de7ea9ac94a2b34580a8d4df69b50128bb6669e1191a895"}, - {file = "pyzmq-26.1.0-cp39-cp39-manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:dcc37d9d708784726fafc9c5e1232de655a009dbf97946f117aefa38d5985a0f"}, - {file = "pyzmq-26.1.0-cp39-cp39-manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:5a6ed52f0b9bf8dcc64cc82cce0607a3dfed1dbb7e8c6f282adfccc7be9781de"}, - {file = "pyzmq-26.1.0-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:451e16ae8bea3d95649317b463c9f95cd9022641ec884e3d63fc67841ae86dfe"}, - {file = "pyzmq-26.1.0-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:906e532c814e1d579138177a00ae835cd6becbf104d45ed9093a3aaf658f6a6a"}, - {file = "pyzmq-26.1.0-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:05bacc4f94af468cc82808ae3293390278d5f3375bb20fef21e2034bb9a505b6"}, - {file = "pyzmq-26.1.0-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:57bb2acba798dc3740e913ffadd56b1fcef96f111e66f09e2a8db3050f1f12c8"}, - {file = "pyzmq-26.1.0-cp39-cp39-win32.whl", hash = "sha256:f774841bb0e8588505002962c02da420bcfb4c5056e87a139c6e45e745c0e2e2"}, - {file = "pyzmq-26.1.0-cp39-cp39-win_amd64.whl", hash = "sha256:359c533bedc62c56415a1f5fcfd8279bc93453afdb0803307375ecf81c962402"}, - {file = "pyzmq-26.1.0-cp39-cp39-win_arm64.whl", hash = "sha256:7907419d150b19962138ecec81a17d4892ea440c184949dc29b358bc730caf69"}, - {file = "pyzmq-26.1.0-pp310-pypy310_pp73-macosx_10_15_x86_64.whl", hash = "sha256:b24079a14c9596846bf7516fe75d1e2188d4a528364494859106a33d8b48be38"}, - {file = "pyzmq-26.1.0-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:59d0acd2976e1064f1b398a00e2c3e77ed0a157529779e23087d4c2fb8aaa416"}, - {file = "pyzmq-26.1.0-pp310-pypy310_pp73-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:911c43a4117915203c4cc8755e0f888e16c4676a82f61caee2f21b0c00e5b894"}, - {file = "pyzmq-26.1.0-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:b10163e586cc609f5f85c9b233195554d77b1e9a0801388907441aaeb22841c5"}, - {file = "pyzmq-26.1.0-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:28a8b2abb76042f5fd7bd720f7fea48c0fd3e82e9de0a1bf2c0de3812ce44a42"}, - {file = "pyzmq-26.1.0-pp37-pypy37_pp73-macosx_10_9_x86_64.whl", hash = "sha256:bef24d3e4ae2c985034439f449e3f9e06bf579974ce0e53d8a507a1577d5b2ab"}, - {file = "pyzmq-26.1.0-pp37-pypy37_pp73-manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:2cd0f4d314f4a2518e8970b6f299ae18cff7c44d4a1fc06fc713f791c3a9e3ea"}, - {file = "pyzmq-26.1.0-pp37-pypy37_pp73-manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:fa25a620eed2a419acc2cf10135b995f8f0ce78ad00534d729aa761e4adcef8a"}, - {file = "pyzmq-26.1.0-pp37-pypy37_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:ef3b048822dca6d231d8a8ba21069844ae38f5d83889b9b690bf17d2acc7d099"}, - {file = "pyzmq-26.1.0-pp37-pypy37_pp73-win_amd64.whl", hash = "sha256:9a6847c92d9851b59b9f33f968c68e9e441f9a0f8fc972c5580c5cd7cbc6ee24"}, - {file = "pyzmq-26.1.0-pp38-pypy38_pp73-macosx_10_9_x86_64.whl", hash = "sha256:c9b9305004d7e4e6a824f4f19b6d8f32b3578aad6f19fc1122aaf320cbe3dc83"}, - {file = "pyzmq-26.1.0-pp38-pypy38_pp73-manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:63c1d3a65acb2f9c92dce03c4e1758cc552f1ae5c78d79a44e3bb88d2fa71f3a"}, - {file = "pyzmq-26.1.0-pp38-pypy38_pp73-manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:d36b8fffe8b248a1b961c86fbdfa0129dfce878731d169ede7fa2631447331be"}, - {file = "pyzmq-26.1.0-pp38-pypy38_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:67976d12ebfd61a3bc7d77b71a9589b4d61d0422282596cf58c62c3866916544"}, - {file = "pyzmq-26.1.0-pp38-pypy38_pp73-win_amd64.whl", hash = "sha256:998444debc8816b5d8d15f966e42751032d0f4c55300c48cc337f2b3e4f17d03"}, - {file = "pyzmq-26.1.0-pp39-pypy39_pp73-macosx_10_15_x86_64.whl", hash = "sha256:e5c88b2f13bcf55fee78ea83567b9fe079ba1a4bef8b35c376043440040f7edb"}, - {file = "pyzmq-26.1.0-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8d906d43e1592be4b25a587b7d96527cb67277542a5611e8ea9e996182fae410"}, - {file = "pyzmq-26.1.0-pp39-pypy39_pp73-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:80b0c9942430d731c786545da6be96d824a41a51742e3e374fedd9018ea43106"}, - {file = "pyzmq-26.1.0-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:314d11564c00b77f6224d12eb3ddebe926c301e86b648a1835c5b28176c83eab"}, - {file = "pyzmq-26.1.0-pp39-pypy39_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:093a1a3cae2496233f14b57f4b485da01b4ff764582c854c0f42c6dd2be37f3d"}, - {file = "pyzmq-26.1.0-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:3c397b1b450f749a7e974d74c06d69bd22dd362142f370ef2bd32a684d6b480c"}, - {file = "pyzmq-26.1.0.tar.gz", hash = "sha256:6c5aeea71f018ebd3b9115c7cb13863dd850e98ca6b9258509de1246461a7e7f"}, + {file = "pyzmq-26.1.1-cp310-cp310-macosx_10_15_universal2.whl", hash = "sha256:b1bb952d1e407463c9333ea7e0c0600001e54e08ce836d4f0aff1fb3f902cf63"}, + {file = "pyzmq-26.1.1-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:65e2a18e845c6ea7ab849c70db932eaeadee5edede9e379eb21c0a44cf523b2e"}, + {file = "pyzmq-26.1.1-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:def7ae3006924b8a0c146a89ab4008310913fa903beedb95e25dea749642528e"}, + {file = "pyzmq-26.1.1-cp310-cp310-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:a8234571df7816f99dde89c3403cb396d70c6554120b795853a8ea56fcc26cd3"}, + {file = "pyzmq-26.1.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:18da8e84dbc30688fd2baefd41df7190607511f916be34f9a24b0e007551822e"}, + {file = "pyzmq-26.1.1-cp310-cp310-manylinux_2_28_x86_64.whl", hash = "sha256:c70dab93d98b2bf3f0ac1265edbf6e7f83acbf71dabcc4611889bb0dea45bed7"}, + {file = "pyzmq-26.1.1-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:fcb90592c5d5c562e1b1a1ceccf6f00036d73c51db0271bf4d352b8d6b31d468"}, + {file = "pyzmq-26.1.1-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:cf4be7460a0c1bc71e9b0e64ecdd75a86386ca6afaa36641686f5542d0314e9d"}, + {file = "pyzmq-26.1.1-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:a4cbecda4ddbfc1e309c3be04d333f9be3fc6178b8b6592b309676f929767a15"}, + {file = "pyzmq-26.1.1-cp310-cp310-win32.whl", hash = "sha256:583f73b113b8165713b6ce028d221402b1b69483055b5aa3f991937e34dd1ead"}, + {file = "pyzmq-26.1.1-cp310-cp310-win_amd64.whl", hash = "sha256:5e6f39ecb8eb7bfcb976c49262e8cf83ff76e082b77ca23ba90c9b6691a345be"}, + {file = "pyzmq-26.1.1-cp310-cp310-win_arm64.whl", hash = "sha256:8d042d6446cab3a1388b38596f5acabb9926b0b95c3894c519356b577a549458"}, + {file = "pyzmq-26.1.1-cp311-cp311-macosx_10_15_universal2.whl", hash = "sha256:362cac2423e36966d336d79d3ec3eafeabc153ee3e7a5cf580d7e74a34b3d912"}, + {file = "pyzmq-26.1.1-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:0841633446cb1539a832a19bb24c03a20c00887d0cedd1d891b495b07e5c5cb5"}, + {file = "pyzmq-26.1.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4e1fcdc333afbf9918d0a614a6e10858aede7da49a60f6705a77e343fe86a317"}, + {file = "pyzmq-26.1.1-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:cc8d655627d775475eafdcf0e49e74bcc1e5e90afd9ab813b4da98f092ed7b93"}, + {file = "pyzmq-26.1.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:32de51744820857a6f7c3077e620ab3f607d0e4388dfead885d5124ab9bcdc5e"}, + {file = "pyzmq-26.1.1-cp311-cp311-manylinux_2_28_x86_64.whl", hash = "sha256:a880240597010914ffb1d6edd04d3deb7ce6a2abf79a0012751438d13630a671"}, + {file = "pyzmq-26.1.1-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:26131b1cec02f941ed2d2b4b8cc051662b1c248b044eff5069df1f500bbced56"}, + {file = "pyzmq-26.1.1-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:ce05841322b58510607f9508a573138d995a46c7928887bc433de9cb760fd2ad"}, + {file = "pyzmq-26.1.1-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:32123ff0a6db521aadf2b95201e967a4e0d11fb89f73663a99d2f54881c07214"}, + {file = "pyzmq-26.1.1-cp311-cp311-win32.whl", hash = "sha256:e790602d7ea1d6c7d8713d571226d67de7ffe47b1e22ae2c043ebd537de1bccb"}, + {file = "pyzmq-26.1.1-cp311-cp311-win_amd64.whl", hash = "sha256:717960855f2d6fdc2dba9df49dff31c414187bb11c76af36343a57d1f7083d9a"}, + {file = "pyzmq-26.1.1-cp311-cp311-win_arm64.whl", hash = "sha256:08956c26dbcd4fd8835cb777a16e21958ed2412317630e19f0018d49dbeeb470"}, + {file = "pyzmq-26.1.1-cp312-cp312-macosx_10_15_universal2.whl", hash = "sha256:e80345900ae241c2c51bead7c9fa247bba6d4b2a83423e9791bae8b0a7f12c52"}, + {file = "pyzmq-26.1.1-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:ec8fe214fcc45dfb0c32e4a7ad1db20244ba2d2fecbf0cbf9d5242d81ca0a375"}, + {file = "pyzmq-26.1.1-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:cf4e283f97688d993cb7a8acbc22889effbbb7cbaa19ee9709751f44be928f5d"}, + {file = "pyzmq-26.1.1-cp312-cp312-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:2508bdc8ab246e5ed7c92023d4352aaad63020ca3b098a4e3f1822db202f703d"}, + {file = "pyzmq-26.1.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:741bdb4d96efe8192616abdc3671931d51a8bcd38c71da2d53fb3127149265d1"}, + {file = "pyzmq-26.1.1-cp312-cp312-manylinux_2_28_x86_64.whl", hash = "sha256:76154943e4c4054b2591792eb3484ef1dd23d59805759f9cebd2f010aa30ee8c"}, + {file = "pyzmq-26.1.1-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:9498ac427d20d0e0ef0e4bbd6200841e91640dfdf619f544ceec7f464cfb6070"}, + {file = "pyzmq-26.1.1-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:6f34453ef3496ca3462f30435bf85f535f9550392987341f9ccc92c102825a79"}, + {file = "pyzmq-26.1.1-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:50f0669324e27cc2091ef6ab76ca7112f364b6249691790b4cffce31e73fda28"}, + {file = "pyzmq-26.1.1-cp312-cp312-win32.whl", hash = "sha256:3ee5cbf2625b94de21c68d0cefd35327c8dfdbd6a98fcc41682b4e8bb00d841f"}, + {file = "pyzmq-26.1.1-cp312-cp312-win_amd64.whl", hash = "sha256:75bd448a28b1001b6928679015bc95dd5f172703ed30135bb9e34fc9cda0a3e7"}, + {file = "pyzmq-26.1.1-cp312-cp312-win_arm64.whl", hash = "sha256:4350233569b4bbef88595c5e77ee38995a6f1f1790fae148b578941bfffd1c24"}, + {file = "pyzmq-26.1.1-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:6c8087a3281c20b1d11042d372ed5a47734af05975d78e4d1d6e7bd1018535f3"}, + {file = "pyzmq-26.1.1-cp313-cp313-macosx_10_15_universal2.whl", hash = "sha256:ebef7d3fe11fe4c688f08bc0211a976c3318c097057f258428200737b9fff4da"}, + {file = "pyzmq-26.1.1-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:7a5342110510045a47de1e87f5f1dcc1d9d90109522316dc9830cfc6157c800f"}, + {file = "pyzmq-26.1.1-cp313-cp313-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:af690ea4be6ca92a67c2b44a779a023bf0838e92d48497a2268175dc4a505691"}, + {file = "pyzmq-26.1.1-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:bc994e220c1403ae087d7f0fa45129d583e46668a019e389060da811a5a9320e"}, + {file = "pyzmq-26.1.1-cp313-cp313-manylinux_2_28_x86_64.whl", hash = "sha256:b8e153f5dffb0310af71fc6fc9cd8174f4c8ea312c415adcb815d786fee78179"}, + {file = "pyzmq-26.1.1-cp313-cp313-musllinux_1_1_aarch64.whl", hash = "sha256:0065026e624052a51033857e5cd45a94b52946b44533f965f0bdf182460e965d"}, + {file = "pyzmq-26.1.1-cp313-cp313-musllinux_1_1_i686.whl", hash = "sha256:63351392f948b5d50b9f55161994bc4feedbfb3f3cfe393d2f503dea2c3ec445"}, + {file = "pyzmq-26.1.1-cp313-cp313-musllinux_1_1_x86_64.whl", hash = "sha256:ffecc43b3c18e36b62fcec995761829b6ac325d8dd74a4f2c5c1653afbb4495a"}, + {file = "pyzmq-26.1.1-cp313-cp313-win32.whl", hash = "sha256:6ff14c2fae6c0c2c1c02590c5c5d75aa1db35b859971b3ca2fcd28f983d9f2b6"}, + {file = "pyzmq-26.1.1-cp313-cp313-win_amd64.whl", hash = "sha256:85f2d2ee5ea9a8f1de86a300e1062fbab044f45b5ce34d20580c0198a8196db0"}, + {file = "pyzmq-26.1.1-cp313-cp313-win_arm64.whl", hash = "sha256:cc09b1de8b985ca5a0ca343dd7fb007267c6b329347a74e200f4654268084239"}, + {file = "pyzmq-26.1.1-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:bc904e86de98f8fc5bd41597da5d61232d2d6d60c4397f26efffabb961b2b245"}, + {file = "pyzmq-26.1.1-cp313-cp313t-macosx_10_15_universal2.whl", hash = "sha256:00f39c367bbd6aa8e4bc36af6510561944c619b58eb36199fa334b594a18f615"}, + {file = "pyzmq-26.1.1-cp313-cp313t-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:de6f384864a959866b782e6a3896538d1424d183f2d3c7ef079f71dcecde7284"}, + {file = "pyzmq-26.1.1-cp313-cp313t-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:3abb15df0c763339edb27a644c19381b2425ddd1aea3dbd77c1601a3b31867b8"}, + {file = "pyzmq-26.1.1-cp313-cp313t-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:40908ec2dd3b29bbadc0916a0d3c87f8dbeebbd8fead8e618539f09e0506dec4"}, + {file = "pyzmq-26.1.1-cp313-cp313t-manylinux_2_28_x86_64.whl", hash = "sha256:c11a95d3f6fc7e714ccd1066f68f9c1abd764a8b3596158be92f46dd49f41e03"}, + {file = "pyzmq-26.1.1-cp313-cp313t-musllinux_1_1_aarch64.whl", hash = "sha256:4437af9fee7a58302dbd511cc49f0cc2b35c112a33a1111fb123cf0be45205ca"}, + {file = "pyzmq-26.1.1-cp313-cp313t-musllinux_1_1_i686.whl", hash = "sha256:76390d3d66406cb01b9681c382874400e9dfd77f30ecdea4bd1bf5226dd4aff0"}, + {file = "pyzmq-26.1.1-cp313-cp313t-musllinux_1_1_x86_64.whl", hash = "sha256:4d4c7fe5e50e269f9c63a260638488fec194a73993008618a59b54c47ef6ae72"}, + {file = "pyzmq-26.1.1-cp37-cp37m-macosx_10_9_x86_64.whl", hash = "sha256:25d128524207f53f7aae7c5abdc2b63f8957a060b00521af5ffcd20986b5d8f4"}, + {file = "pyzmq-26.1.1-cp37-cp37m-manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:d74b925d997e4f92b042bdd7085cd0a309ee0fd7cb4dc376059bbff6b32ff34f"}, + {file = "pyzmq-26.1.1-cp37-cp37m-manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:732f957441e5b1c65a7509395e6b6cafee9e12df9aa5f4bf92ed266fe0ba70ee"}, + {file = "pyzmq-26.1.1-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f0a45102ad7ed9f9ddf2bd699cc5df37742cf7301111cba06001b927efecb120"}, + {file = "pyzmq-26.1.1-cp37-cp37m-musllinux_1_1_aarch64.whl", hash = "sha256:9f380d5333fc7cd17423f486125dcc073918676e33db70a6a8172b19fc78d23d"}, + {file = "pyzmq-26.1.1-cp37-cp37m-musllinux_1_1_i686.whl", hash = "sha256:8eaffcd6bf6a9d00b66a2052a33fa7e6a6575427e9644395f13c3d070f2918dc"}, + {file = "pyzmq-26.1.1-cp37-cp37m-musllinux_1_1_x86_64.whl", hash = "sha256:f1483d4975ae1b387b39bb8e23d1ff32fe5621aa9e4ed3055d05e9c5613fea53"}, + {file = "pyzmq-26.1.1-cp37-cp37m-win32.whl", hash = "sha256:a83653c6bbe5887caea55e49fbd2909c14b73acf43bcc051eb60b2d514bbd46e"}, + {file = "pyzmq-26.1.1-cp37-cp37m-win_amd64.whl", hash = "sha256:9763a8d3f5f74ef679989b373c37cc22e8d07e56d26439205cb83edb7722357f"}, + {file = "pyzmq-26.1.1-cp38-cp38-macosx_10_15_universal2.whl", hash = "sha256:2b045647caf620ce0ed6c8fd9fb6a73116f99aceed966b152a5ba1b416d25311"}, + {file = "pyzmq-26.1.1-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:f66dcb6625c002f209cdc12cae1a1fec926493cd2262efe37dc6b25a30cea863"}, + {file = "pyzmq-26.1.1-cp38-cp38-manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:0cf1d980c969fb9e538f52abd2227f09e015096bc5c3ef7aa26e0d64051c1db8"}, + {file = "pyzmq-26.1.1-cp38-cp38-manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:443ebf5e261a95ee9725693f2a5a71401f89b89df0e0ea58844b074067aac2f1"}, + {file = "pyzmq-26.1.1-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:29de77ba1b1877fe7defc1b9140e65cbd35f72a63bc501e56c2eae55bde5fff4"}, + {file = "pyzmq-26.1.1-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:2f6071ec95af145d7b659dae6786871cd85f0acc599286b6f8ba0c74592d83dd"}, + {file = "pyzmq-26.1.1-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:6f0512fc87629ad968889176bf2165d721cd817401a281504329e2a2ed0ca6a3"}, + {file = "pyzmq-26.1.1-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:5ccfcf13e80719f6a2d9c0a021d9e47d4550907a29253554be2c09582f6d7963"}, + {file = "pyzmq-26.1.1-cp38-cp38-win32.whl", hash = "sha256:809673947e95752e407aaaaf03f205ee86ebfff9ca51db6d4003dfd87b8428d1"}, + {file = "pyzmq-26.1.1-cp38-cp38-win_amd64.whl", hash = "sha256:62b5180e23e6f581600459cd983473cd723fdc64350f606d21407c99832aaf5f"}, + {file = "pyzmq-26.1.1-cp39-cp39-macosx_10_15_universal2.whl", hash = "sha256:fe73d7c89d6f803bed122135ff5783364e8cdb479cf6fe2d764a44b6349e7e0f"}, + {file = "pyzmq-26.1.1-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:db1b7e2b50ef21f398036786da4c153db63203a402396d9f21e08ea61f3f8dba"}, + {file = "pyzmq-26.1.1-cp39-cp39-manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:7c506a51cb01bb997a3f6440db0d121e5e7a32396e9948b1fdb6a7bfa67243f4"}, + {file = "pyzmq-26.1.1-cp39-cp39-manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:92eca4f80e8a748d880e55d3cf57ef487692e439f12d5c5a2e1cce84aaa7f6cb"}, + {file = "pyzmq-26.1.1-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:14bdbae02f72f4716b0ffe7500e9da303d719ddde1f3dcfb4c4f6cc1cf73bb02"}, + {file = "pyzmq-26.1.1-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:e03be7ed17836c9434cce0668ac1e2cc9143d7169f90f46a0167f6155e176e32"}, + {file = "pyzmq-26.1.1-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:bc5df31e36e4fddd4c8b5c42daee8d54d7b529e898ac984be97bf5517de166a7"}, + {file = "pyzmq-26.1.1-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:f218179c90a12d660906e04b25a340dd63e9743000ba16232ddaf46888f269da"}, + {file = "pyzmq-26.1.1-cp39-cp39-win32.whl", hash = "sha256:7dfabc180a4da422a4b349c63077347392463a75fa07aa3be96712ed6d42c547"}, + {file = "pyzmq-26.1.1-cp39-cp39-win_amd64.whl", hash = "sha256:c5248e6e0fcbbbc912982e99cdd51c342601f495b0fa5bd667f3bdbdbf3e170f"}, + {file = "pyzmq-26.1.1-cp39-cp39-win_arm64.whl", hash = "sha256:2ae7aa1408778dc74582a1226052b930f9083b54b64d7e6ef6ec0466cfdcdec2"}, + {file = "pyzmq-26.1.1-pp310-pypy310_pp73-macosx_10_15_x86_64.whl", hash = "sha256:be3fc2b11c0c384949cf1f01f9a48555039408b0f3e877863b1754225635953e"}, + {file = "pyzmq-26.1.1-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:48dee75c2a9fa4f4a583d4028d564a0453447ee1277a29b07acc3743c092e259"}, + {file = "pyzmq-26.1.1-pp310-pypy310_pp73-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:23f2fe4fb567e8098ebaa7204819658195b10ddd86958a97a6058eed2901eed3"}, + {file = "pyzmq-26.1.1-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:472cacd16f627c06d3c8b2d374345ab74446bae913584a6245e2aa935336d929"}, + {file = "pyzmq-26.1.1-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:8285b25aa20fcc46f1ca4afbc39fd3d5f2fe4c4bbf7f2c7f907a214e87a70024"}, + {file = "pyzmq-26.1.1-pp37-pypy37_pp73-macosx_10_9_x86_64.whl", hash = "sha256:2067e63fd9d5c13cfe12624dab0366053e523b37a7a01678ce4321f839398939"}, + {file = "pyzmq-26.1.1-pp37-pypy37_pp73-manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:cc109be2ee3638035d276e18eaf66a1e1f44201c0c4bea4ee0c692766bbd3570"}, + {file = "pyzmq-26.1.1-pp37-pypy37_pp73-manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:d0da97e65ee73261dba70469cc8f63d8da3a8a825337a2e3d246b9e95141cdd0"}, + {file = "pyzmq-26.1.1-pp37-pypy37_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:aa79c528706561306938b275f89bb2c6985ce08469c27e5de05bc680df5e826f"}, + {file = "pyzmq-26.1.1-pp37-pypy37_pp73-win_amd64.whl", hash = "sha256:3ddbd851a3a2651fdc5065a2804d50cf2f4b13b1bcd66de8e9e855d0217d4fcd"}, + {file = "pyzmq-26.1.1-pp38-pypy38_pp73-macosx_10_9_x86_64.whl", hash = "sha256:d3df226ab7464684ae6706e20a5cbab717c3735a7e409b3fa598b754d49f1946"}, + {file = "pyzmq-26.1.1-pp38-pypy38_pp73-manylinux_2_12_i686.manylinux2010_i686.whl", hash = "sha256:abad7b897e960d577eb4a0f3f789c1780bc3ffe2e7c27cf317e7c90ad26acf12"}, + {file = "pyzmq-26.1.1-pp38-pypy38_pp73-manylinux_2_12_x86_64.manylinux2010_x86_64.whl", hash = "sha256:c513d829a548c2d5c88983167be2b3aa537f6d1191edcdc6fcd8999e18bdd994"}, + {file = "pyzmq-26.1.1-pp38-pypy38_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:70af4c9c991714ef1c65957605a8de42ef0d0620dd5f125953c8e682281bdb80"}, + {file = "pyzmq-26.1.1-pp38-pypy38_pp73-win_amd64.whl", hash = "sha256:8d4234f335b0d0842f7d661d8cd50cbad0729be58f1c4deb85cd96b38fe95025"}, + {file = "pyzmq-26.1.1-pp39-pypy39_pp73-macosx_10_15_x86_64.whl", hash = "sha256:2c0fdb7b758e0e1605157e480b00b3a599073068a37091a1c75ec65bf7498645"}, + {file = "pyzmq-26.1.1-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:fc657577f057d60dd3642c9f95f28b432889b73143140061f7c1331d02f03df6"}, + {file = "pyzmq-26.1.1-pp39-pypy39_pp73-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:3e3b66fe6131b4f33d239f7d4c3bfb2f8532d8644bae3b3da4f3987073edac55"}, + {file = "pyzmq-26.1.1-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:59b57e912feef6951aec8bb03fe0faa5ad5f36962883c72a30a9c965e6d988fd"}, + {file = "pyzmq-26.1.1-pp39-pypy39_pp73-manylinux_2_28_x86_64.whl", hash = "sha256:146956aec7d947c5afc5e7da0841423d7a53f84fd160fff25e682361dcfb32cb"}, + {file = "pyzmq-26.1.1-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:9521b874fd489495865172f344e46e0159095d1f161858e3fc6e28e43ca15160"}, + {file = "pyzmq-26.1.1.tar.gz", hash = "sha256:a7db05d8b7cd1a8c6610e9e9aa55d525baae7a44a43e18bc3260eb3f92de96c6"}, ] [package.dependencies] @@ -6738,30 +6453,19 @@ win32 = ["pywin32"] [[package]] name = "setuptools" -version = "72.2.0" +version = "73.0.1" description = "Easily download, build, install, upgrade, and uninstall Python packages" optional = false python-versions = ">=3.8" files = [ - {file = "setuptools-72.2.0-py3-none-any.whl", hash = "sha256:f11dd94b7bae3a156a95ec151f24e4637fb4fa19c878e4d191bfb8b2d82728c4"}, - {file = "setuptools-72.2.0.tar.gz", hash = "sha256:80aacbf633704e9c8bfa1d99fa5dd4dc59573efcf9e4042c13d3bcef91ac2ef9"}, + {file = "setuptools-73.0.1-py3-none-any.whl", hash = "sha256:b208925fcb9f7af924ed2dc04708ea89791e24bde0d3020b27df0e116088b34e"}, + {file = "setuptools-73.0.1.tar.gz", hash = "sha256:d59a3e788ab7e012ab2c4baed1b376da6366883ee20d7a5fc426816e3d7b1193"}, ] [package.extras] -core = ["importlib-metadata (>=6)", "importlib-resources (>=5.10.2)", "jaraco.text (>=3.7)", "more-itertools (>=8.8)", "ordered-set (>=3.1.1)", "packaging (>=24)", "platformdirs (>=2.6.2)", "tomli (>=2.0.1)", "wheel (>=0.43.0)"] +core = ["importlib-metadata (>=6)", "importlib-resources (>=5.10.2)", "jaraco.text (>=3.7)", "more-itertools (>=8.8)", "packaging (>=24)", "platformdirs (>=2.6.2)", "tomli (>=2.0.1)", "wheel (>=0.43.0)"] doc = ["furo", "jaraco.packaging (>=9.3)", "jaraco.tidelift (>=1.4)", "pygments-github-lexers (==0.0.5)", "pyproject-hooks (!=1.1)", "rst.linker (>=1.9)", "sphinx (>=3.5)", "sphinx-favicon", "sphinx-inline-tabs", "sphinx-lint", "sphinx-notfound-page (>=1,<2)", "sphinx-reredirects", "sphinxcontrib-towncrier", "towncrier (<24.7)"] -test = ["build[virtualenv] (>=1.0.3)", "filelock (>=3.4.0)", "importlib-metadata", "ini2toml[lite] (>=0.14)", "jaraco.develop (>=7.21)", "jaraco.envs (>=2.2)", "jaraco.path (>=3.2.0)", "jaraco.test", "mypy (==1.11.*)", "packaging (>=23.2)", "pip (>=19.1)", "pyproject-hooks (!=1.1)", "pytest (>=6,!=8.1.*)", "pytest-checkdocs (>=2.4)", "pytest-cov", "pytest-enabler (>=2.2)", "pytest-home (>=0.5)", "pytest-mypy", "pytest-perf", "pytest-ruff (<0.4)", "pytest-ruff (>=0.2.1)", "pytest-ruff (>=0.3.2)", "pytest-subprocess", "pytest-timeout", "pytest-xdist (>=3)", "tomli", "tomli-w (>=1.0.0)", "virtualenv (>=13.0.0)", "wheel"] - -[[package]] -name = "shellingham" -version = "1.5.4" -description = "Tool to Detect Surrounding Shell" -optional = false -python-versions = ">=3.7" -files = [ - {file = "shellingham-1.5.4-py2.py3-none-any.whl", hash = "sha256:7ecfff8f2fd72616f7481040475a65b2bf8af90a56c89140852d1120324e8686"}, - {file = "shellingham-1.5.4.tar.gz", hash = "sha256:8dbca0739d487e5bd35ab3ca4b36e11c4078f3a234bfce294b0a0291363404de"}, -] +test = ["build[virtualenv] (>=1.0.3)", "filelock (>=3.4.0)", "importlib-metadata", "ini2toml[lite] (>=0.14)", "jaraco.develop (>=7.21)", "jaraco.envs (>=2.2)", "jaraco.path (>=3.2.0)", "jaraco.test", "mypy (==1.11.*)", "packaging (>=23.2)", "pip (>=19.1)", "pyproject-hooks (!=1.1)", "pytest (>=6,!=8.1.*)", "pytest-checkdocs (>=2.4)", "pytest-cov", "pytest-enabler (>=2.2)", "pytest-home (>=0.5)", "pytest-mypy", "pytest-perf", "pytest-ruff (<0.4)", "pytest-ruff (>=0.2.1)", "pytest-ruff (>=0.3.2)", "pytest-subprocess", "pytest-timeout", "pytest-xdist (>=3)", "tomli", "tomli-w (>=1.0.0)", "virtualenv (>=13.0.0)", "wheel (>=0.44.0)"] [[package]] name = "six" @@ -7086,13 +6790,13 @@ typing = ["mypy (>=1.6,<2.0)", "traitlets (>=5.11.1)"] [[package]] name = "textual" -version = "0.74.0" +version = "0.76.0" description = "Modern Text User Interface framework" optional = false python-versions = "<4.0.0,>=3.8.1" files = [ - {file = "textual-0.74.0-py3-none-any.whl", hash = "sha256:69b18904d11400e586274deb1270f5db7b4254a2ac00fe805f23b16c324345e4"}, - {file = "textual-0.74.0.tar.gz", hash = "sha256:808c4e8727283ef84a123620449b217b6033ffde49f3ed195b3d0f98bb042969"}, + {file = "textual-0.76.0-py3-none-any.whl", hash = "sha256:e2035609c889dba507d34a5d7b333f1c8c53a29fb170962cb92101507663517a"}, + {file = "textual-0.76.0.tar.gz", hash = "sha256:b12e8879d591090c0901b5cb8121d086e28e677353b368292d3865ec99b83b70"}, ] [package.dependencies] @@ -7272,32 +6976,15 @@ files = [ docs = ["myst-parser", "pydata-sphinx-theme", "sphinx"] test = ["argcomplete (>=3.0.3)", "mypy (>=1.7.0)", "pre-commit", "pytest (>=7.0,<8.2)", "pytest-mock", "pytest-mypy-testing"] -[[package]] -name = "typer" -version = "0.12.3" -description = "Typer, build great CLIs. Easy to code. Based on Python type hints." -optional = false -python-versions = ">=3.7" -files = [ - {file = "typer-0.12.3-py3-none-any.whl", hash = "sha256:070d7ca53f785acbccba8e7d28b08dcd88f79f1fbda035ade0aecec71ca5c914"}, - {file = "typer-0.12.3.tar.gz", hash = "sha256:49e73131481d804288ef62598d97a1ceef3058905aa536a1134f90891ba35482"}, -] - -[package.dependencies] -click = ">=8.0.0" -rich = ">=10.11.0" -shellingham = ">=1.3.0" -typing-extensions = ">=3.7.4.3" - [[package]] name = "types-python-dateutil" -version = "2.9.0.20240316" +version = "2.9.0.20240821" description = "Typing stubs for python-dateutil" optional = false python-versions = ">=3.8" files = [ - {file = "types-python-dateutil-2.9.0.20240316.tar.gz", hash = "sha256:5d2f2e240b86905e40944dd787db6da9263f0deabef1076ddaed797351ec0202"}, - {file = "types_python_dateutil-2.9.0.20240316-py3-none-any.whl", hash = "sha256:6b8cb66d960771ce5ff974e9dd45e38facb81718cc1e208b10b1baccbfdbee3b"}, + {file = "types-python-dateutil-2.9.0.20240821.tar.gz", hash = "sha256:9649d1dcb6fef1046fb18bebe9ea2aa0028b160918518c34589a46045f6ebd98"}, + {file = "types_python_dateutil-2.9.0.20240821-py3-none-any.whl", hash = "sha256:f5889fcb4e63ed4aaa379b44f93c32593d50b9a94c9a60a0c854d8cc3511cd57"}, ] [[package]] @@ -7407,68 +7094,60 @@ zstd = ["zstandard (>=0.18.0)"] [[package]] name = "uvicorn" -version = "0.30.6" +version = "0.25.0" description = "The lightning-fast ASGI server." optional = false python-versions = ">=3.8" files = [ - {file = "uvicorn-0.30.6-py3-none-any.whl", hash = "sha256:65fd46fe3fda5bdc1b03b94eb634923ff18cd35b2f084813ea79d1f103f711b5"}, - {file = "uvicorn-0.30.6.tar.gz", hash = "sha256:4b15decdda1e72be08209e860a1e10e92439ad5b97cf44cc945fcbee66fc5788"}, + {file = "uvicorn-0.25.0-py3-none-any.whl", hash = "sha256:ce107f5d9bd02b4636001a77a4e74aab5e1e2b146868ebbad565237145af444c"}, + {file = "uvicorn-0.25.0.tar.gz", hash = "sha256:6dddbad1d7ee0f5140aba5ec138ddc9612c5109399903828b4874c9937f009c2"}, ] [package.dependencies] click = ">=7.0" -colorama = {version = ">=0.4", optional = true, markers = "sys_platform == \"win32\" and extra == \"standard\""} h11 = ">=0.8" -httptools = {version = ">=0.5.0", optional = true, markers = "extra == \"standard\""} -python-dotenv = {version = ">=0.13", optional = true, markers = "extra == \"standard\""} -pyyaml = {version = ">=5.1", optional = true, markers = "extra == \"standard\""} -typing-extensions = {version = ">=4.0", markers = "python_version < \"3.11\""} -uvloop = {version = ">=0.14.0,<0.15.0 || >0.15.0,<0.15.1 || >0.15.1", optional = true, markers = "(sys_platform != \"win32\" and sys_platform != \"cygwin\") and platform_python_implementation != \"PyPy\" and extra == \"standard\""} -watchfiles = {version = ">=0.13", optional = true, markers = "extra == \"standard\""} -websockets = {version = ">=10.4", optional = true, markers = "extra == \"standard\""} [package.extras] standard = ["colorama (>=0.4)", "httptools (>=0.5.0)", "python-dotenv (>=0.13)", "pyyaml (>=5.1)", "uvloop (>=0.14.0,!=0.15.0,!=0.15.1)", "watchfiles (>=0.13)", "websockets (>=10.4)"] [[package]] name = "uvloop" -version = "0.19.0" +version = "0.20.0" description = "Fast implementation of asyncio event loop on top of libuv" optional = false python-versions = ">=3.8.0" files = [ - {file = "uvloop-0.19.0-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:de4313d7f575474c8f5a12e163f6d89c0a878bc49219641d49e6f1444369a90e"}, - {file = "uvloop-0.19.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:5588bd21cf1fcf06bded085f37e43ce0e00424197e7c10e77afd4bbefffef428"}, - {file = "uvloop-0.19.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:7b1fd71c3843327f3bbc3237bedcdb6504fd50368ab3e04d0410e52ec293f5b8"}, - {file = "uvloop-0.19.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:5a05128d315e2912791de6088c34136bfcdd0c7cbc1cf85fd6fd1bb321b7c849"}, - {file = "uvloop-0.19.0-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:cd81bdc2b8219cb4b2556eea39d2e36bfa375a2dd021404f90a62e44efaaf957"}, - {file = "uvloop-0.19.0-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:5f17766fb6da94135526273080f3455a112f82570b2ee5daa64d682387fe0dcd"}, - {file = "uvloop-0.19.0-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:4ce6b0af8f2729a02a5d1575feacb2a94fc7b2e983868b009d51c9a9d2149bef"}, - {file = "uvloop-0.19.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:31e672bb38b45abc4f26e273be83b72a0d28d074d5b370fc4dcf4c4eb15417d2"}, - {file = "uvloop-0.19.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:570fc0ed613883d8d30ee40397b79207eedd2624891692471808a95069a007c1"}, - {file = "uvloop-0.19.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:5138821e40b0c3e6c9478643b4660bd44372ae1e16a322b8fc07478f92684e24"}, - {file = "uvloop-0.19.0-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:91ab01c6cd00e39cde50173ba4ec68a1e578fee9279ba64f5221810a9e786533"}, - {file = "uvloop-0.19.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:47bf3e9312f63684efe283f7342afb414eea4d3011542155c7e625cd799c3b12"}, - {file = "uvloop-0.19.0-cp312-cp312-macosx_10_9_universal2.whl", hash = "sha256:da8435a3bd498419ee8c13c34b89b5005130a476bda1d6ca8cfdde3de35cd650"}, - {file = "uvloop-0.19.0-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:02506dc23a5d90e04d4f65c7791e65cf44bd91b37f24cfc3ef6cf2aff05dc7ec"}, - {file = "uvloop-0.19.0-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2693049be9d36fef81741fddb3f441673ba12a34a704e7b4361efb75cf30befc"}, - {file = "uvloop-0.19.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:7010271303961c6f0fe37731004335401eb9075a12680738731e9c92ddd96ad6"}, - {file = "uvloop-0.19.0-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:5daa304d2161d2918fa9a17d5635099a2f78ae5b5960e742b2fcfbb7aefaa593"}, - {file = "uvloop-0.19.0-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:7207272c9520203fea9b93843bb775d03e1cf88a80a936ce760f60bb5add92f3"}, - {file = "uvloop-0.19.0-cp38-cp38-macosx_10_9_universal2.whl", hash = "sha256:78ab247f0b5671cc887c31d33f9b3abfb88d2614b84e4303f1a63b46c046c8bd"}, - {file = "uvloop-0.19.0-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:472d61143059c84947aa8bb74eabbace30d577a03a1805b77933d6bd13ddebbd"}, - {file = "uvloop-0.19.0-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:45bf4c24c19fb8a50902ae37c5de50da81de4922af65baf760f7c0c42e1088be"}, - {file = "uvloop-0.19.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:271718e26b3e17906b28b67314c45d19106112067205119dddbd834c2b7ce797"}, - {file = "uvloop-0.19.0-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:34175c9fd2a4bc3adc1380e1261f60306344e3407c20a4d684fd5f3be010fa3d"}, - {file = "uvloop-0.19.0-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:e27f100e1ff17f6feeb1f33968bc185bf8ce41ca557deee9d9bbbffeb72030b7"}, - {file = "uvloop-0.19.0-cp39-cp39-macosx_10_9_universal2.whl", hash = "sha256:13dfdf492af0aa0a0edf66807d2b465607d11c4fa48f4a1fd41cbea5b18e8e8b"}, - {file = "uvloop-0.19.0-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:6e3d4e85ac060e2342ff85e90d0c04157acb210b9ce508e784a944f852a40e67"}, - {file = "uvloop-0.19.0-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8ca4956c9ab567d87d59d49fa3704cf29e37109ad348f2d5223c9bf761a332e7"}, - {file = "uvloop-0.19.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f467a5fd23b4fc43ed86342641f3936a68ded707f4627622fa3f82a120e18256"}, - {file = "uvloop-0.19.0-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:492e2c32c2af3f971473bc22f086513cedfc66a130756145a931a90c3958cb17"}, - {file = "uvloop-0.19.0-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:2df95fca285a9f5bfe730e51945ffe2fa71ccbfdde3b0da5772b4ee4f2e770d5"}, - {file = "uvloop-0.19.0.tar.gz", hash = "sha256:0246f4fd1bf2bf702e06b0d45ee91677ee5c31242f39aab4ea6fe0c51aedd0fd"}, + {file = "uvloop-0.20.0-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:9ebafa0b96c62881d5cafa02d9da2e44c23f9f0cd829f3a32a6aff771449c996"}, + {file = "uvloop-0.20.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:35968fc697b0527a06e134999eef859b4034b37aebca537daeb598b9d45a137b"}, + {file = "uvloop-0.20.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b16696f10e59d7580979b420eedf6650010a4a9c3bd8113f24a103dfdb770b10"}, + {file = "uvloop-0.20.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:9b04d96188d365151d1af41fa2d23257b674e7ead68cfd61c725a422764062ae"}, + {file = "uvloop-0.20.0-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:94707205efbe809dfa3a0d09c08bef1352f5d3d6612a506f10a319933757c006"}, + {file = "uvloop-0.20.0-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:89e8d33bb88d7263f74dc57d69f0063e06b5a5ce50bb9a6b32f5fcbe655f9e73"}, + {file = "uvloop-0.20.0-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:e50289c101495e0d1bb0bfcb4a60adde56e32f4449a67216a1ab2750aa84f037"}, + {file = "uvloop-0.20.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:e237f9c1e8a00e7d9ddaa288e535dc337a39bcbf679f290aee9d26df9e72bce9"}, + {file = "uvloop-0.20.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:746242cd703dc2b37f9d8b9f173749c15e9a918ddb021575a0205ec29a38d31e"}, + {file = "uvloop-0.20.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:82edbfd3df39fb3d108fc079ebc461330f7c2e33dbd002d146bf7c445ba6e756"}, + {file = "uvloop-0.20.0-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:80dc1b139516be2077b3e57ce1cb65bfed09149e1d175e0478e7a987863b68f0"}, + {file = "uvloop-0.20.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:4f44af67bf39af25db4c1ac27e82e9665717f9c26af2369c404be865c8818dcf"}, + {file = "uvloop-0.20.0-cp312-cp312-macosx_10_9_universal2.whl", hash = "sha256:4b75f2950ddb6feed85336412b9a0c310a2edbcf4cf931aa5cfe29034829676d"}, + {file = "uvloop-0.20.0-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:77fbc69c287596880ecec2d4c7a62346bef08b6209749bf6ce8c22bbaca0239e"}, + {file = "uvloop-0.20.0-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:6462c95f48e2d8d4c993a2950cd3d31ab061864d1c226bbf0ee2f1a8f36674b9"}, + {file = "uvloop-0.20.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:649c33034979273fa71aa25d0fe120ad1777c551d8c4cd2c0c9851d88fcb13ab"}, + {file = "uvloop-0.20.0-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:3a609780e942d43a275a617c0839d85f95c334bad29c4c0918252085113285b5"}, + {file = "uvloop-0.20.0-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:aea15c78e0d9ad6555ed201344ae36db5c63d428818b4b2a42842b3870127c00"}, + {file = "uvloop-0.20.0-cp38-cp38-macosx_10_9_universal2.whl", hash = "sha256:f0e94b221295b5e69de57a1bd4aeb0b3a29f61be6e1b478bb8a69a73377db7ba"}, + {file = "uvloop-0.20.0-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:fee6044b64c965c425b65a4e17719953b96e065c5b7e09b599ff332bb2744bdf"}, + {file = "uvloop-0.20.0-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:265a99a2ff41a0fd56c19c3838b29bf54d1d177964c300dad388b27e84fd7847"}, + {file = "uvloop-0.20.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:b10c2956efcecb981bf9cfb8184d27d5d64b9033f917115a960b83f11bfa0d6b"}, + {file = "uvloop-0.20.0-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:e7d61fe8e8d9335fac1bf8d5d82820b4808dd7a43020c149b63a1ada953d48a6"}, + {file = "uvloop-0.20.0-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:2beee18efd33fa6fdb0976e18475a4042cd31c7433c866e8a09ab604c7c22ff2"}, + {file = "uvloop-0.20.0-cp39-cp39-macosx_10_9_universal2.whl", hash = "sha256:d8c36fdf3e02cec92aed2d44f63565ad1522a499c654f07935c8f9d04db69e95"}, + {file = "uvloop-0.20.0-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:a0fac7be202596c7126146660725157d4813aa29a4cc990fe51346f75ff8fde7"}, + {file = "uvloop-0.20.0-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:9d0fba61846f294bce41eb44d60d58136090ea2b5b99efd21cbdf4e21927c56a"}, + {file = "uvloop-0.20.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:95720bae002ac357202e0d866128eb1ac82545bcf0b549b9abe91b5178d9b541"}, + {file = "uvloop-0.20.0-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:36c530d8fa03bfa7085af54a48f2ca16ab74df3ec7108a46ba82fd8b411a2315"}, + {file = "uvloop-0.20.0-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:e97152983442b499d7a71e44f29baa75b3b02e65d9c44ba53b10338e98dedb66"}, + {file = "uvloop-0.20.0.tar.gz", hash = "sha256:4603ca714a754fc8d9b197e325db25b2ea045385e8a3ad05d3463de725fdf469"}, ] [package.extras] @@ -7490,105 +7169,6 @@ files = [ docs = ["Sphinx (>=1.8.1)", "docutils", "pylons-sphinx-themes (>=1.0.9)"] testing = ["coverage (>=5.0)", "pytest", "pytest-cover"] -[[package]] -name = "watchfiles" -version = "0.23.0" -description = "Simple, modern and high performance file watching and code reload in python." -optional = false -python-versions = ">=3.8" -files = [ - {file = "watchfiles-0.23.0-cp310-cp310-macosx_10_12_x86_64.whl", hash = "sha256:bee8ce357a05c20db04f46c22be2d1a2c6a8ed365b325d08af94358e0688eeb4"}, - {file = "watchfiles-0.23.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:4ccd3011cc7ee2f789af9ebe04745436371d36afe610028921cab9f24bb2987b"}, - {file = "watchfiles-0.23.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:fb02d41c33be667e6135e6686f1bb76104c88a312a18faa0ef0262b5bf7f1a0f"}, - {file = "watchfiles-0.23.0-cp310-cp310-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:7cf12ac34c444362f3261fb3ff548f0037ddd4c5bb85f66c4be30d2936beb3c5"}, - {file = "watchfiles-0.23.0-cp310-cp310-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:a0b2c25040a3c0ce0e66c7779cc045fdfbbb8d59e5aabfe033000b42fe44b53e"}, - {file = "watchfiles-0.23.0-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:ecf2be4b9eece4f3da8ba5f244b9e51932ebc441c0867bd6af46a3d97eb068d6"}, - {file = "watchfiles-0.23.0-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:40cb8fa00028908211eb9f8d47744dca21a4be6766672e1ff3280bee320436f1"}, - {file = "watchfiles-0.23.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:8f48c917ffd36ff9a5212614c2d0d585fa8b064ca7e66206fb5c095015bc8207"}, - {file = "watchfiles-0.23.0-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:9d183e3888ada88185ab17064079c0db8c17e32023f5c278d7bf8014713b1b5b"}, - {file = "watchfiles-0.23.0-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:9837edf328b2805346f91209b7e660f65fb0e9ca18b7459d075d58db082bf981"}, - {file = "watchfiles-0.23.0-cp310-none-win32.whl", hash = "sha256:296e0b29ab0276ca59d82d2da22cbbdb39a23eed94cca69aed274595fb3dfe42"}, - {file = "watchfiles-0.23.0-cp310-none-win_amd64.whl", hash = "sha256:4ea756e425ab2dfc8ef2a0cb87af8aa7ef7dfc6fc46c6f89bcf382121d4fff75"}, - {file = "watchfiles-0.23.0-cp311-cp311-macosx_10_12_x86_64.whl", hash = "sha256:e397b64f7aaf26915bf2ad0f1190f75c855d11eb111cc00f12f97430153c2eab"}, - {file = "watchfiles-0.23.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:b4ac73b02ca1824ec0a7351588241fd3953748d3774694aa7ddb5e8e46aef3e3"}, - {file = "watchfiles-0.23.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:130a896d53b48a1cecccfa903f37a1d87dbb74295305f865a3e816452f6e49e4"}, - {file = "watchfiles-0.23.0-cp311-cp311-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:c5e7803a65eb2d563c73230e9d693c6539e3c975ccfe62526cadde69f3fda0cf"}, - {file = "watchfiles-0.23.0-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d1aa4cc85202956d1a65c88d18c7b687b8319dbe6b1aec8969784ef7a10e7d1a"}, - {file = "watchfiles-0.23.0-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:87f889f6e58849ddb7c5d2cb19e2e074917ed1c6e3ceca50405775166492cca8"}, - {file = "watchfiles-0.23.0-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:37fd826dac84c6441615aa3f04077adcc5cac7194a021c9f0d69af20fb9fa788"}, - {file = "watchfiles-0.23.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ee7db6e36e7a2c15923072e41ea24d9a0cf39658cb0637ecc9307b09d28827e1"}, - {file = "watchfiles-0.23.0-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:2368c5371c17fdcb5a2ea71c5c9d49f9b128821bfee69503cc38eae00feb3220"}, - {file = "watchfiles-0.23.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:857af85d445b9ba9178db95658c219dbd77b71b8264e66836a6eba4fbf49c320"}, - {file = "watchfiles-0.23.0-cp311-none-win32.whl", hash = "sha256:1d636c8aeb28cdd04a4aa89030c4b48f8b2954d8483e5f989774fa441c0ed57b"}, - {file = "watchfiles-0.23.0-cp311-none-win_amd64.whl", hash = "sha256:46f1d8069a95885ca529645cdbb05aea5837d799965676e1b2b1f95a4206313e"}, - {file = "watchfiles-0.23.0-cp311-none-win_arm64.whl", hash = "sha256:e495ed2a7943503766c5d1ff05ae9212dc2ce1c0e30a80d4f0d84889298fa304"}, - {file = "watchfiles-0.23.0-cp312-cp312-macosx_10_12_x86_64.whl", hash = "sha256:1db691bad0243aed27c8354b12d60e8e266b75216ae99d33e927ff5238d270b5"}, - {file = "watchfiles-0.23.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:62d2b18cb1edaba311fbbfe83fb5e53a858ba37cacb01e69bc20553bb70911b8"}, - {file = "watchfiles-0.23.0-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:e087e8fdf1270d000913c12e6eca44edd02aad3559b3e6b8ef00f0ce76e0636f"}, - {file = "watchfiles-0.23.0-cp312-cp312-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:dd41d5c72417b87c00b1b635738f3c283e737d75c5fa5c3e1c60cd03eac3af77"}, - {file = "watchfiles-0.23.0-cp312-cp312-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:1e5f3ca0ff47940ce0a389457b35d6df601c317c1e1a9615981c474452f98de1"}, - {file = "watchfiles-0.23.0-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:6991e3a78f642368b8b1b669327eb6751439f9f7eaaa625fae67dd6070ecfa0b"}, - {file = "watchfiles-0.23.0-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:7f7252f52a09f8fa5435dc82b6af79483118ce6bd51eb74e6269f05ee22a7b9f"}, - {file = "watchfiles-0.23.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:0e01bcb8d767c58865207a6c2f2792ad763a0fe1119fb0a430f444f5b02a5ea0"}, - {file = "watchfiles-0.23.0-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:8e56fbcdd27fce061854ddec99e015dd779cae186eb36b14471fc9ae713b118c"}, - {file = "watchfiles-0.23.0-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:bd3e2d64500a6cad28bcd710ee6269fbeb2e5320525acd0cfab5f269ade68581"}, - {file = "watchfiles-0.23.0-cp312-none-win32.whl", hash = "sha256:eb99c954291b2fad0eff98b490aa641e128fbc4a03b11c8a0086de8b7077fb75"}, - {file = "watchfiles-0.23.0-cp312-none-win_amd64.whl", hash = "sha256:dccc858372a56080332ea89b78cfb18efb945da858fabeb67f5a44fa0bcb4ebb"}, - {file = "watchfiles-0.23.0-cp312-none-win_arm64.whl", hash = "sha256:6c21a5467f35c61eafb4e394303720893066897fca937bade5b4f5877d350ff8"}, - {file = "watchfiles-0.23.0-cp313-cp313-macosx_10_12_x86_64.whl", hash = "sha256:ba31c32f6b4dceeb2be04f717811565159617e28d61a60bb616b6442027fd4b9"}, - {file = "watchfiles-0.23.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:85042ab91814fca99cec4678fc063fb46df4cbb57b4835a1cc2cb7a51e10250e"}, - {file = "watchfiles-0.23.0-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:24655e8c1c9c114005c3868a3d432c8aa595a786b8493500071e6a52f3d09217"}, - {file = "watchfiles-0.23.0-cp313-cp313-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:6b1a950ab299a4a78fd6369a97b8763732bfb154fdb433356ec55a5bce9515c1"}, - {file = "watchfiles-0.23.0-cp313-cp313-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:b8d3c5cd327dd6ce0edfc94374fb5883d254fe78a5e9d9dfc237a1897dc73cd1"}, - {file = "watchfiles-0.23.0-cp313-cp313-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:9ff785af8bacdf0be863ec0c428e3288b817e82f3d0c1d652cd9c6d509020dd0"}, - {file = "watchfiles-0.23.0-cp313-cp313-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:02b7ba9d4557149410747353e7325010d48edcfe9d609a85cb450f17fd50dc3d"}, - {file = "watchfiles-0.23.0-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:48a1b05c0afb2cd2f48c1ed2ae5487b116e34b93b13074ed3c22ad5c743109f0"}, - {file = "watchfiles-0.23.0-cp313-cp313-musllinux_1_1_aarch64.whl", hash = "sha256:109a61763e7318d9f821b878589e71229f97366fa6a5c7720687d367f3ab9eef"}, - {file = "watchfiles-0.23.0-cp313-cp313-musllinux_1_1_x86_64.whl", hash = "sha256:9f8e6bb5ac007d4a4027b25f09827ed78cbbd5b9700fd6c54429278dacce05d1"}, - {file = "watchfiles-0.23.0-cp313-none-win32.whl", hash = "sha256:f46c6f0aec8d02a52d97a583782d9af38c19a29900747eb048af358a9c1d8e5b"}, - {file = "watchfiles-0.23.0-cp313-none-win_amd64.whl", hash = "sha256:f449afbb971df5c6faeb0a27bca0427d7b600dd8f4a068492faec18023f0dcff"}, - {file = "watchfiles-0.23.0-cp38-cp38-macosx_10_12_x86_64.whl", hash = "sha256:2dddc2487d33e92f8b6222b5fb74ae2cfde5e8e6c44e0248d24ec23befdc5366"}, - {file = "watchfiles-0.23.0-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:e75695cc952e825fa3e0684a7f4a302f9128721f13eedd8dbd3af2ba450932b8"}, - {file = "watchfiles-0.23.0-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2537ef60596511df79b91613a5bb499b63f46f01a11a81b0a2b0dedf645d0a9c"}, - {file = "watchfiles-0.23.0-cp38-cp38-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:20b423b58f5fdde704a226b598a2d78165fe29eb5621358fe57ea63f16f165c4"}, - {file = "watchfiles-0.23.0-cp38-cp38-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:b98732ec893975455708d6fc9a6daab527fc8bbe65be354a3861f8c450a632a4"}, - {file = "watchfiles-0.23.0-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:ee1f5fcbf5bc33acc0be9dd31130bcba35d6d2302e4eceafafd7d9018c7755ab"}, - {file = "watchfiles-0.23.0-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:a8f195338a5a7b50a058522b39517c50238358d9ad8284fd92943643144c0c03"}, - {file = "watchfiles-0.23.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:524fcb8d59b0dbee2c9b32207084b67b2420f6431ed02c18bd191e6c575f5c48"}, - {file = "watchfiles-0.23.0-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:0eff099a4df36afaa0eea7a913aa64dcf2cbd4e7a4f319a73012210af4d23810"}, - {file = "watchfiles-0.23.0-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:a8323daae27ea290ba3350c70c836c0d2b0fb47897fa3b0ca6a5375b952b90d3"}, - {file = "watchfiles-0.23.0-cp38-none-win32.whl", hash = "sha256:aafea64a3ae698695975251f4254df2225e2624185a69534e7fe70581066bc1b"}, - {file = "watchfiles-0.23.0-cp38-none-win_amd64.whl", hash = "sha256:c846884b2e690ba62a51048a097acb6b5cd263d8bd91062cd6137e2880578472"}, - {file = "watchfiles-0.23.0-cp39-cp39-macosx_10_12_x86_64.whl", hash = "sha256:a753993635eccf1ecb185dedcc69d220dab41804272f45e4aef0a67e790c3eb3"}, - {file = "watchfiles-0.23.0-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:6bb91fa4d0b392f0f7e27c40981e46dda9eb0fbc84162c7fb478fe115944f491"}, - {file = "watchfiles-0.23.0-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b1f67312efa3902a8e8496bfa9824d3bec096ff83c4669ea555c6bdd213aa516"}, - {file = "watchfiles-0.23.0-cp39-cp39-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:7ca6b71dcc50d320c88fb2d88ecd63924934a8abc1673683a242a7ca7d39e781"}, - {file = "watchfiles-0.23.0-cp39-cp39-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:2aec5c29915caf08771d2507da3ac08e8de24a50f746eb1ed295584ba1820330"}, - {file = "watchfiles-0.23.0-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:1733b9bc2c8098c6bdb0ff7a3d7cb211753fecb7bd99bdd6df995621ee1a574b"}, - {file = "watchfiles-0.23.0-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:02ff5d7bd066c6a7673b17c8879cd8ee903078d184802a7ee851449c43521bdd"}, - {file = "watchfiles-0.23.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:18e2de19801b0eaa4c5292a223effb7cfb43904cb742c5317a0ac686ed604765"}, - {file = "watchfiles-0.23.0-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:8ada449e22198c31fb013ae7e9add887e8d2bd2335401abd3cbc55f8c5083647"}, - {file = "watchfiles-0.23.0-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:3af1b05361e1cc497bf1be654a664750ae61f5739e4bb094a2be86ec8c6db9b6"}, - {file = "watchfiles-0.23.0-cp39-none-win32.whl", hash = "sha256:486bda18be5d25ab5d932699ceed918f68eb91f45d018b0343e3502e52866e5e"}, - {file = "watchfiles-0.23.0-cp39-none-win_amd64.whl", hash = "sha256:d2d42254b189a346249424fb9bb39182a19289a2409051ee432fb2926bad966a"}, - {file = "watchfiles-0.23.0-pp310-pypy310_pp73-macosx_10_12_x86_64.whl", hash = "sha256:6a9265cf87a5b70147bfb2fec14770ed5b11a5bb83353f0eee1c25a81af5abfe"}, - {file = "watchfiles-0.23.0-pp310-pypy310_pp73-macosx_11_0_arm64.whl", hash = "sha256:9f02a259fcbbb5fcfe7a0805b1097ead5ba7a043e318eef1db59f93067f0b49b"}, - {file = "watchfiles-0.23.0-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:1ebaebb53b34690da0936c256c1cdb0914f24fb0e03da76d185806df9328abed"}, - {file = "watchfiles-0.23.0-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fd257f98cff9c6cb39eee1a83c7c3183970d8a8d23e8cf4f47d9a21329285cee"}, - {file = "watchfiles-0.23.0-pp38-pypy38_pp73-macosx_10_12_x86_64.whl", hash = "sha256:aba037c1310dd108411d27b3d5815998ef0e83573e47d4219f45753c710f969f"}, - {file = "watchfiles-0.23.0-pp38-pypy38_pp73-macosx_11_0_arm64.whl", hash = "sha256:a96ac14e184aa86dc43b8a22bb53854760a58b2966c2b41580de938e9bf26ed0"}, - {file = "watchfiles-0.23.0-pp38-pypy38_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:11698bb2ea5e991d10f1f4f83a39a02f91e44e4bd05f01b5c1ec04c9342bf63c"}, - {file = "watchfiles-0.23.0-pp38-pypy38_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:efadd40fca3a04063d40c4448c9303ce24dd6151dc162cfae4a2a060232ebdcb"}, - {file = "watchfiles-0.23.0-pp39-pypy39_pp73-macosx_10_12_x86_64.whl", hash = "sha256:556347b0abb4224c5ec688fc58214162e92a500323f50182f994f3ad33385dcb"}, - {file = "watchfiles-0.23.0-pp39-pypy39_pp73-macosx_11_0_arm64.whl", hash = "sha256:1cf7f486169986c4b9d34087f08ce56a35126600b6fef3028f19ca16d5889071"}, - {file = "watchfiles-0.23.0-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f18de0f82c62c4197bea5ecf4389288ac755896aac734bd2cc44004c56e4ac47"}, - {file = "watchfiles-0.23.0-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:532e1f2c491274d1333a814e4c5c2e8b92345d41b12dc806cf07aaff786beb66"}, - {file = "watchfiles-0.23.0.tar.gz", hash = "sha256:9338ade39ff24f8086bb005d16c29f8e9f19e55b18dcb04dfa26fcbc09da497b"}, -] - -[package.dependencies] -anyio = ">=3.0.0" - [[package]] name = "wcwidth" version = "0.2.13" @@ -7642,96 +7222,15 @@ docs = ["Sphinx (>=6.0)", "myst-parser (>=2.0.0)", "sphinx-rtd-theme (>=1.1.0)"] optional = ["python-socks", "wsaccel"] test = ["websockets"] -[[package]] -name = "websockets" -version = "12.0" -description = "An implementation of the WebSocket Protocol (RFC 6455 & 7692)" -optional = false -python-versions = ">=3.8" -files = [ - {file = "websockets-12.0-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:d554236b2a2006e0ce16315c16eaa0d628dab009c33b63ea03f41c6107958374"}, - {file = "websockets-12.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:2d225bb6886591b1746b17c0573e29804619c8f755b5598d875bb4235ea639be"}, - {file = "websockets-12.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:eb809e816916a3b210bed3c82fb88eaf16e8afcf9c115ebb2bacede1797d2547"}, - {file = "websockets-12.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:c588f6abc13f78a67044c6b1273a99e1cf31038ad51815b3b016ce699f0d75c2"}, - {file = "websockets-12.0-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:5aa9348186d79a5f232115ed3fa9020eab66d6c3437d72f9d2c8ac0c6858c558"}, - {file = "websockets-12.0-cp310-cp310-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6350b14a40c95ddd53e775dbdbbbc59b124a5c8ecd6fbb09c2e52029f7a9f480"}, - {file = "websockets-12.0-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:70ec754cc2a769bcd218ed8d7209055667b30860ffecb8633a834dde27d6307c"}, - {file = "websockets-12.0-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:6e96f5ed1b83a8ddb07909b45bd94833b0710f738115751cdaa9da1fb0cb66e8"}, - {file = "websockets-12.0-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:4d87be612cbef86f994178d5186add3d94e9f31cc3cb499a0482b866ec477603"}, - {file = "websockets-12.0-cp310-cp310-win32.whl", hash = "sha256:befe90632d66caaf72e8b2ed4d7f02b348913813c8b0a32fae1cc5fe3730902f"}, - {file = "websockets-12.0-cp310-cp310-win_amd64.whl", hash = "sha256:363f57ca8bc8576195d0540c648aa58ac18cf85b76ad5202b9f976918f4219cf"}, - {file = "websockets-12.0-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:5d873c7de42dea355d73f170be0f23788cf3fa9f7bed718fd2830eefedce01b4"}, - {file = "websockets-12.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:3f61726cae9f65b872502ff3c1496abc93ffbe31b278455c418492016e2afc8f"}, - {file = "websockets-12.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:ed2fcf7a07334c77fc8a230755c2209223a7cc44fc27597729b8ef5425aa61a3"}, - {file = "websockets-12.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8e332c210b14b57904869ca9f9bf4ca32f5427a03eeb625da9b616c85a3a506c"}, - {file = "websockets-12.0-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:5693ef74233122f8ebab026817b1b37fe25c411ecfca084b29bc7d6efc548f45"}, - {file = "websockets-12.0-cp311-cp311-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6e9e7db18b4539a29cc5ad8c8b252738a30e2b13f033c2d6e9d0549b45841c04"}, - {file = "websockets-12.0-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:6e2df67b8014767d0f785baa98393725739287684b9f8d8a1001eb2839031447"}, - {file = "websockets-12.0-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:bea88d71630c5900690fcb03161ab18f8f244805c59e2e0dc4ffadae0a7ee0ca"}, - {file = "websockets-12.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:dff6cdf35e31d1315790149fee351f9e52978130cef6c87c4b6c9b3baf78bc53"}, - {file = "websockets-12.0-cp311-cp311-win32.whl", hash = "sha256:3e3aa8c468af01d70332a382350ee95f6986db479ce7af14d5e81ec52aa2b402"}, - {file = "websockets-12.0-cp311-cp311-win_amd64.whl", hash = "sha256:25eb766c8ad27da0f79420b2af4b85d29914ba0edf69f547cc4f06ca6f1d403b"}, - {file = "websockets-12.0-cp312-cp312-macosx_10_9_universal2.whl", hash = "sha256:0e6e2711d5a8e6e482cacb927a49a3d432345dfe7dea8ace7b5790df5932e4df"}, - {file = "websockets-12.0-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:dbcf72a37f0b3316e993e13ecf32f10c0e1259c28ffd0a85cee26e8549595fbc"}, - {file = "websockets-12.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:12743ab88ab2af1d17dd4acb4645677cb7063ef4db93abffbf164218a5d54c6b"}, - {file = "websockets-12.0-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:7b645f491f3c48d3f8a00d1fce07445fab7347fec54a3e65f0725d730d5b99cb"}, - {file = "websockets-12.0-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:9893d1aa45a7f8b3bc4510f6ccf8db8c3b62120917af15e3de247f0780294b92"}, - {file = "websockets-12.0-cp312-cp312-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:1f38a7b376117ef7aff996e737583172bdf535932c9ca021746573bce40165ed"}, - {file = "websockets-12.0-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:f764ba54e33daf20e167915edc443b6f88956f37fb606449b4a5b10ba42235a5"}, - {file = "websockets-12.0-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:1e4b3f8ea6a9cfa8be8484c9221ec0257508e3a1ec43c36acdefb2a9c3b00aa2"}, - {file = "websockets-12.0-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:9fdf06fd06c32205a07e47328ab49c40fc1407cdec801d698a7c41167ea45113"}, - {file = "websockets-12.0-cp312-cp312-win32.whl", hash = "sha256:baa386875b70cbd81798fa9f71be689c1bf484f65fd6fb08d051a0ee4e79924d"}, - {file = "websockets-12.0-cp312-cp312-win_amd64.whl", hash = "sha256:ae0a5da8f35a5be197f328d4727dbcfafa53d1824fac3d96cdd3a642fe09394f"}, - {file = "websockets-12.0-cp38-cp38-macosx_10_9_universal2.whl", hash = "sha256:5f6ffe2c6598f7f7207eef9a1228b6f5c818f9f4d53ee920aacd35cec8110438"}, - {file = "websockets-12.0-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:9edf3fc590cc2ec20dc9d7a45108b5bbaf21c0d89f9fd3fd1685e223771dc0b2"}, - {file = "websockets-12.0-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:8572132c7be52632201a35f5e08348137f658e5ffd21f51f94572ca6c05ea81d"}, - {file = "websockets-12.0-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:604428d1b87edbf02b233e2c207d7d528460fa978f9e391bd8aaf9c8311de137"}, - {file = "websockets-12.0-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:1a9d160fd080c6285e202327aba140fc9a0d910b09e423afff4ae5cbbf1c7205"}, - {file = "websockets-12.0-cp38-cp38-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:87b4aafed34653e465eb77b7c93ef058516cb5acf3eb21e42f33928616172def"}, - {file = "websockets-12.0-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:b2ee7288b85959797970114deae81ab41b731f19ebcd3bd499ae9ca0e3f1d2c8"}, - {file = "websockets-12.0-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:7fa3d25e81bfe6a89718e9791128398a50dec6d57faf23770787ff441d851967"}, - {file = "websockets-12.0-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:a571f035a47212288e3b3519944f6bf4ac7bc7553243e41eac50dd48552b6df7"}, - {file = "websockets-12.0-cp38-cp38-win32.whl", hash = "sha256:3c6cc1360c10c17463aadd29dd3af332d4a1adaa8796f6b0e9f9df1fdb0bad62"}, - {file = "websockets-12.0-cp38-cp38-win_amd64.whl", hash = "sha256:1bf386089178ea69d720f8db6199a0504a406209a0fc23e603b27b300fdd6892"}, - {file = "websockets-12.0-cp39-cp39-macosx_10_9_universal2.whl", hash = "sha256:ab3d732ad50a4fbd04a4490ef08acd0517b6ae6b77eb967251f4c263011a990d"}, - {file = "websockets-12.0-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:a1d9697f3337a89691e3bd8dc56dea45a6f6d975f92e7d5f773bc715c15dde28"}, - {file = "websockets-12.0-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:1df2fbd2c8a98d38a66f5238484405b8d1d16f929bb7a33ed73e4801222a6f53"}, - {file = "websockets-12.0-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:23509452b3bc38e3a057382c2e941d5ac2e01e251acce7adc74011d7d8de434c"}, - {file = "websockets-12.0-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:2e5fc14ec6ea568200ea4ef46545073da81900a2b67b3e666f04adf53ad452ec"}, - {file = "websockets-12.0-cp39-cp39-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:46e71dbbd12850224243f5d2aeec90f0aaa0f2dde5aeeb8fc8df21e04d99eff9"}, - {file = "websockets-12.0-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:b81f90dcc6c85a9b7f29873beb56c94c85d6f0dac2ea8b60d995bd18bf3e2aae"}, - {file = "websockets-12.0-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:a02413bc474feda2849c59ed2dfb2cddb4cd3d2f03a2fedec51d6e959d9b608b"}, - {file = "websockets-12.0-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:bbe6013f9f791944ed31ca08b077e26249309639313fff132bfbf3ba105673b9"}, - {file = "websockets-12.0-cp39-cp39-win32.whl", hash = "sha256:cbe83a6bbdf207ff0541de01e11904827540aa069293696dd528a6640bd6a5f6"}, - {file = "websockets-12.0-cp39-cp39-win_amd64.whl", hash = "sha256:fc4e7fa5414512b481a2483775a8e8be7803a35b30ca805afa4998a84f9fd9e8"}, - {file = "websockets-12.0-pp310-pypy310_pp73-macosx_10_9_x86_64.whl", hash = "sha256:248d8e2446e13c1d4326e0a6a4e9629cb13a11195051a73acf414812700badbd"}, - {file = "websockets-12.0-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f44069528d45a933997a6fef143030d8ca8042f0dfaad753e2906398290e2870"}, - {file = "websockets-12.0-pp310-pypy310_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:c4e37d36f0d19f0a4413d3e18c0d03d0c268ada2061868c1e6f5ab1a6d575077"}, - {file = "websockets-12.0-pp310-pypy310_pp73-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:3d829f975fc2e527a3ef2f9c8f25e553eb7bc779c6665e8e1d52aa22800bb38b"}, - {file = "websockets-12.0-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:2c71bd45a777433dd9113847af751aae36e448bc6b8c361a566cb043eda6ec30"}, - {file = "websockets-12.0-pp38-pypy38_pp73-macosx_10_9_x86_64.whl", hash = "sha256:0bee75f400895aef54157b36ed6d3b308fcab62e5260703add87f44cee9c82a6"}, - {file = "websockets-12.0-pp38-pypy38_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:423fc1ed29f7512fceb727e2d2aecb952c46aa34895e9ed96071821309951123"}, - {file = "websockets-12.0-pp38-pypy38_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:27a5e9964ef509016759f2ef3f2c1e13f403725a5e6a1775555994966a66e931"}, - {file = "websockets-12.0-pp38-pypy38_pp73-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:c3181df4583c4d3994d31fb235dc681d2aaad744fbdbf94c4802485ececdecf2"}, - {file = "websockets-12.0-pp38-pypy38_pp73-win_amd64.whl", hash = "sha256:b067cb952ce8bf40115f6c19f478dc71c5e719b7fbaa511359795dfd9d1a6468"}, - {file = "websockets-12.0-pp39-pypy39_pp73-macosx_10_9_x86_64.whl", hash = "sha256:00700340c6c7ab788f176d118775202aadea7602c5cc6be6ae127761c16d6b0b"}, - {file = "websockets-12.0-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:e469d01137942849cff40517c97a30a93ae79917752b34029f0ec72df6b46399"}, - {file = "websockets-12.0-pp39-pypy39_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:ffefa1374cd508d633646d51a8e9277763a9b78ae71324183693959cf94635a7"}, - {file = "websockets-12.0-pp39-pypy39_pp73-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ba0cab91b3956dfa9f512147860783a1829a8d905ee218a9837c18f683239611"}, - {file = "websockets-12.0-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:2cb388a5bfb56df4d9a406783b7f9dbefb888c09b71629351cc6b036e9259370"}, - {file = "websockets-12.0-py3-none-any.whl", hash = "sha256:dc284bbc8d7c78a6c69e0c7325ab46ee5e40bb4d50e494d8131a07ef47500e9e"}, - {file = "websockets-12.0.tar.gz", hash = "sha256:81df9cbcbb6c260de1e007e58c011bfebe2dafc8435107b0537f393dd38c8b1b"}, -] - [[package]] name = "werkzeug" -version = "3.0.3" +version = "3.0.4" description = "The comprehensive WSGI web application library." optional = false python-versions = ">=3.8" files = [ - {file = "werkzeug-3.0.3-py3-none-any.whl", hash = "sha256:fc9645dc43e03e4d630d23143a04a7f947a9a3b5727cd535fdfe155a17cc48c8"}, - {file = "werkzeug-3.0.3.tar.gz", hash = "sha256:097e5bfda9f0aba8da6b8545146def481d06aa7d3266e7448e2cccf67dd8bd18"}, + {file = "werkzeug-3.0.4-py3-none-any.whl", hash = "sha256:02c9eb92b7d6c06f31a782811505d2157837cea66aaede3e217c7c27c039476c"}, + {file = "werkzeug-3.0.4.tar.gz", hash = "sha256:34f2371506b250df4d4f84bfe7b0921e4762525762bbd936614909fe25cd7306"}, ] [package.dependencies] @@ -7892,5 +7391,5 @@ test = ["big-O", "importlib-resources", "jaraco.functools", "jaraco.itertools", [metadata] lock-version = "2.0" -python-versions = ">=3.10,<3.13" -content-hash = "7f7ec7606a489263681ab3e193b9343e8f2f9db0224d79aa16daccef389081c5" +python-versions = ">=3.11.2,<3.13" +content-hash = "0cc354196e1fa32694e02324f27d0709aa943f1c3936d9044fdbaa0c4a33be4f" diff --git a/pyproject.toml b/pyproject.toml index 5bb87ad92..db47c4936 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -1,30 +1,30 @@ [tool.poetry] name = "graphfleet" -version = "0.4.15" +version = "0.5.11" description = "Build upon GraphRAG, provide a quick solution setup and is part of the overall Qredence System like AgenticFleet and FleetUI" authors = ["Zachary "] license = "Apache-2.0" readme = "README.md" -packages = [{ include = "graphfleet" }] +packages = [ + {include = "graphfleet"} +] [tool.poetry-dynamic-versioning] enable = true vcs = "git" style = "semver" + [tool.poetry.dependencies] -python = ">=3.10,<3.13" +python = ">=3.11.2,<3.13" environs = "^11.0.0" datashaper = "^0.0.49" - # Vector Stores -azure-search-documents = "^11.4.0" -lancedb = ">=0.11.0,<0.12.0" -azure-cosmos = "^4.7.0" +lancedb = "^0.11.0" # Event Loops -uvloop = { version = "^0.19.0", markers = "platform_system != 'Windows'" } -nest-asyncio = { version = "^1.6.0", markers = "platform_system == 'Windows'" } +uvloop = {version = ">=0.20.0,<0.21.0", markers = "platform_system != 'Windows'"} +nest-asyncio = {version = "^1.6.0", markers = "platform_system == 'Windows'"} # Async IO aiolimiter = "^1.1.0" @@ -32,17 +32,17 @@ aiofiles = "^24.1.0" # LLM openai = "^1.35.7" -nltk = "3.8.1" +nltk = "^3.9.1" # Changed from "3.8.1" to "^3.9.1" tiktoken = "^0.7.0" # Data-Sci -numba = "0.60.0" +numba = "^0.60.0" numpy = "^1.25.2" graspologic = "^3.4.1" +graspologic-native = "^1.2.1" networkx = "^3" fastparquet = "^2024.2.0" -# 1.13.0 was a footgun -scipy = "1.12.0" +scipy = "^1.12.0" # Configuration pyyaml = "^6.0.1" @@ -50,46 +50,43 @@ pyaml-env = "^1.2.1" python-dotenv = "^1.0.0" # Network -tenacity = ">=9.0.0,<10.0.0" +tenacity = ">=8.2.0,<10.0.0" swifter = "^1.4.0" pydantic = "^2" rich = "^13.6.0" -textual = ">=0.74.0,<0.75.0" +textual = ">=0.74.0" devtools = "^0.12.2" typing-extensions = "^4.12.2" -#Azure +# Azure dependencies +azure-search-documents = "^11.4.0" +azure-cosmos = "^4.7.0" azure-storage-blob = "^12.19.0" azure-identity = "^1.17.1" +azure-cli-core = "^2.62.0" +azure-core = "^1.30.2" at = "^0.0.3" yfiles-jupyter-graphs = "^1.7.3" jupyter-contrib-nbextensions = "^0.7.0" jupyter = "^1.0.0" -graphrag = "0.3.0" -packaging = "^24.1" -pdfreader = "^0.1.15" -pypdf2 = "^3.0.1" +packaging = ">=23.1,<24.1" - -azure-cli-core = "^2.62.0" yq = "^3.4.3" jq = "^1.7.0" -fastapi = "^0.111.1" - +fastapi = "^0.110.1" +uvicorn = "^0.25.0" promptflow = "^1.14.0" promptflow-tools = "^1.4.0" microsoft-bing-websearch = "^1.0.0" build = "^1.2.1" -uvicorn = "^0.30.6" requests = "^2.32.3" pandas = "^2.2.2" detect-secrets = "^1.5.0" -azure-core = "^1.30.2" kubernetes = "^30.1.0" httpx = "^0.27.0" opencensus-ext-azure = "^1.1.13" @@ -98,7 +95,12 @@ opencensus-context = "^0.1.3" opencensus = "^0.11.4" flake8 = "^7.1.1" ipython = "^8.26.0" - +python-magic = "^0.4.27" +tqdm = "^4.66.5" +python-multipart = "^0.0.9" +ipywidgets = "^8.1.3" +pypdf = "^4.3.1" +graphrag = "^0.3.1" [tool.poetry.group.dev.dependencies] coverage = "^7.6.0" @@ -114,6 +116,14 @@ ruff = "^0.5.2" semversioner = "^2.0.3" update-toml = "^0.2.1" +graphrag = "^0.3.1" + +[tool.ruff] +select = ["E", "F", "I"] +fix = true + +[tool.ruff.isort] +known-first-party = ["graphfleet"] [build-system] requires = ["poetry-core>=1.0.0", "poetry-dynamic-versioning>=1.0.0,<2.0.0"] @@ -122,4 +132,4 @@ build-backend = "poetry_dynamic_versioning.backend" [tool.poe.tasks] _sort_imports = "ruff check --select I --fix . --preview" _format_code = "ruff format . --preview" -_ruff_check = 'ruff check . --preview' +_ruff_check = 'ruff check . --preview' \ No newline at end of file diff --git a/scripts/copy_graphrag_source.py b/scripts/copy_graphrag_source.py new file mode 100644 index 000000000..558269ff3 --- /dev/null +++ b/scripts/copy_graphrag_source.py @@ -0,0 +1,36 @@ +## +# This script copies the GraphRAG source code to the graphfleet/libs/graphrag directory. +# Run this script from the root of the graphfleet project. +# python scripts/copy_graphrag_source.py + + +import os +import shutil +import graphrag + + +def copy_graphrag_source(): + # Get the location of the installed graphrag package in .venv + venv_path = os.path.join(".venv", "lib", "python3.10", "site-packages") + graphrag_path = os.path.join(venv_path, "graphrag") + + # Define the destination path in your project + dest_path = os.path.join("graphfleet", "libs", "graphrag") + + # Create the destination directory if it doesn't exist + os.makedirs(dest_path, exist_ok=True) + + # Copy the contents of the graphrag package to the destination + for item in os.listdir(graphrag_path): + s = os.path.join(graphrag_path, item) + d = os.path.join(dest_path, item) + if os.path.isdir(s): + shutil.copytree(s, d, dirs_exist_ok=True) + else: + shutil.copy2(s, d) + + print(f"GraphRAG source copied to {dest_path}") + + +if __name__ == "__main__": + copy_graphrag_source() \ No newline at end of file