Python Conversational Search System

A conversational search system built in python.

Installation

Pull the repository from github, and install as a python package:

pip install -e .

Usage

If installed locally, henceforth the command py_css is available. Otherwise, the following entrypoint shall be called:

python py_css/main.py
# OR, if installed locally:
py_css

A detailed help page will be presented using:

py_css --help

CLI Mode

If installed as a python package, the following command is available:

py_css cli

Run Queries File

py_css run_file --log=INFO --queries=data/queries_train.csv --output=output/train.txt

Run Queries and Evaluate Performance

py_css eval --log=INFO --queries=data/queries_train.csv --qrels=data/qrels_train.txt

Create Kaggle Runfile Format

py_css kaggle --log=INFO --queries=data/queries_test.csv --output=output/kaggle-prf.csv

Retrieval Pipelines

As outlined in the paper, four retrieval pipelines were implemented:

Baseline

Can be selected by specifying the following parameters:

--method=baseline
--baseline-params=1000,1000,50

Indexing

For indexing, the document collection has to be placed into the data/ folder.
Further Instructions

Parameters

Position	ID	Description	Constraints
0	bm25_docs	The number of documents to be retrieved using BM25.
1	mono_t5_docs	The number of documents to be reranked by monoT5 after retrieval.	bm25_docs >= mono_t5_docs
2	duo_t5_docs	The number of documents to be reranked by duoT5 after monoT5 reranking.	mono_t5_docs <= duo_t5_docs

Baseline + RM3

Can be selected by specifying the following parameters:

--method=baseline-prf
--baseline-prf-params=1000,17,26,1000,50

Indexing

For indexing, the document collection has to be placed into the data/ folder.
Further Instructions

Parameters

Position	ID	Description	Constraints
0	bm25_docs	The number of documents to be retrieved using BM25.
1	rm3_fb_docs	The number of documents to be used for RM3 query expansion.
2	rm3_fb_terms	The number of terms to expand the query with using RM3.
3	mono_t5_docs	The number of documents to be reranked by monoT5 after retrieval.	bm25_docs >= mono_t5_docs
4	duo_t5_docs	The number of documents to be reranked by duoT5 after monoT5 reranking.	mono_t5_docs <= duo_t5_docs

doc2query

Can be selected by specifying the following parameters:

--method=doc2query
--doc2query-params=1000,1000,50

Indexing

For indexing, the document collection has to be placed into the data/ folder. Additionally, descriptive queries for each document have to be generated using this script.
Further Instructions

Parameters

Position	ID	Description	Constraints
0	bm25_docs	The number of documents to be retrieved using BM25.
1	mono_t5_docs	The number of documents to be reranked by monoT5 after retrieval.	bm25_docs >= mono_t5_docs
2	duo_t5_docs	The number of documents to be reranked by duoT5 after monoT5 reranking.	mono_t5_docs <= duo_t5_docs

doc2query + RM3

Can be selected by specifying the following parameters:

--method=doc2query-prf
--doc2query-prf-params=1000,17,26,1000,50

Indexing

For indexing, the document collection has to be placed into the data/ folder. Additionally, descriptive queries for each document have to be generated using this script.
Further Instructions

Parameters

Position	ID	Description	Constraints
0	bm25_docs	The number of documents to be retrieved using BM25.
1	rm3_fb_docs	The number of documents to be used for RM3 query expansion.
2	rm3_fb_terms	The number of terms to expand the query with using RM3.
3	mono_t5_docs	The number of documents to be reranked by monoT5 after retrieval.	bm25_docs >= mono_t5_docs
4	duo_t5_docs	The number of documents to be reranked by duoT5 after monoT5 reranking.	mono_t5_docs <= duo_t5_docs