Add NER PII Masking module

autorag/nodes/passagefilter/__init__.py

@@ -1,3 +1,4 @@
 from .percentile_cutoff import similarity_percentile_cutoff
+from .ner_pii_masking import ner_pii_masking
 from .pass_passage_filter import pass_passage_filter
 from .threshold_cutoff import similarity_threshold_cutoff

autorag/nodes/passagefilter/base.py

@@ -1,10 +1,11 @@
 import functools
+import os
 from pathlib import Path
 from typing import Union, Tuple, List
 
 import pandas as pd
 
-from autorag.utils import result_to_dataframe, validate_qa_dataset
+from autorag.utils import result_to_dataframe, validate_qa_dataset, fetch_contents
 
 
 # same with passage filter from now
@@ -33,8 +34,18 @@ def wrapper(
         assert "retrieved_ids" in previous_result.columns, "previous_result must have retrieved_ids column."
         ids = previous_result["retrieved_ids"].tolist()
 
-        filtered_contents, filtered_ids, filtered_scores = func(queries=queries, contents_list=contents,
-                                                                scores_list=scores, ids_list=ids, *args, **kwargs)
+        if func.__name__ == 'recency_filter':
+            corpus_df = pd.read_parquet(os.path.join(project_dir, "data", "corpus.parquet"))
+            metadatas = fetch_contents(corpus_df, ids, column_name='metadata')
+            times = [[time['last_modified_datetime'] for time in time_list] for time_list in metadatas]
+            filtered_contents, filtered_ids, filtered_scores \
+                = func(contents_list=contents, scores_list=scores, ids_list=ids, time_list=times, *args, **kwargs)
+        elif func.__name__ == 'ner_pii_masking':
+            filtered_contents, filtered_ids, filtered_scores = func(contents_list=contents,
+                                                                    scores_list=scores, ids_list=ids, *args, **kwargs)
+        else:
+            filtered_contents, filtered_ids, filtered_scores = func(queries=queries, contents_list=contents,
+                                                                    scores_list=scores, ids_list=ids, *args, **kwargs)
 
         return filtered_contents, filtered_ids, filtered_scores
 

autorag/nodes/passagefilter/ner_pii_masking.py

@@ -0,0 +1,35 @@
+from typing import List, Tuple
+
+from transformers import pipeline
+
+from autorag.nodes.passagefilter.base import passage_filter_node
+
+
+@passage_filter_node
+def ner_pii_masking(contents_list: List[List[str]],
+                    scores_list: List[List[float]], ids_list: List[List[str]],
+                    ) -> Tuple[List[List[str]], List[List[str]], List[List[float]]]:
+    """
+    Mask PII in the contents using NER.
+    Use a Hugging Face NER model for PII Masking
+
+    :param contents_list: The list of lists of contents to filter
+    :param scores_list: The list of lists of scores retrieved
+    :param ids_list: The list of lists of ids retrieved
+    :return: Tuple of lists containing the filtered contents, ids, and scores
+    """
+    model = pipeline("ner", grouped_entities=True)
+
+    masked_contents_list = list(
+        map(lambda contents: list(map(lambda content: mask_pii(model, content), contents)), contents_list))
+
+    return masked_contents_list, ids_list, scores_list
+
+
+def mask_pii(model, text: str) -> str:
+    new_text = text
+    response = model(text)
+    for entry in response:
+        entity_group_tag = f"[{entry['entity_group']}_{entry['start']}]"
+        new_text = new_text.replace(entry["word"], entity_group_tag).strip()
+    return new_text

autorag/support.py

@@ -43,6 +43,7 @@ def get_support_modules(module_name: str) -> Callable:
         'pass_passage_filter': ('autorag.nodes.passagefilter', 'pass_passage_filter'),
         'similarity_threshold_cutoff': ('autorag.nodes.passagefilter', 'similarity_threshold_cutoff'),
         'similarity_percentile_cutoff': ('autorag.nodes.passagefilter', 'similarity_percentile_cutoff'),
+        'ner_pii_masking': ('autorag.nodes.passagefilter', 'ner_pii_masking'),
         # passage_compressor
         'tree_summarize': ('autorag.nodes.passagecompressor', 'tree_summarize'),
         'pass_compressor': ('autorag.nodes.passagecompressor', 'pass_compressor'),

docs/source/nodes/passage_filter/ner_pii_masking.md

@@ -0,0 +1,23 @@
+# NER PII Masking
+
+This module is inspired by
+LlamaIndex ['PII Masking'](https://docs.llamaindex.ai/en/stable/examples/node_postprocessor/PII/#option-1-use-ner-model-for-pii-masking).
+
+Use a Hugging Face NER model for PII Masking
+
+## What is PII Masking?
+
+PII(Personally Identifiable Information) Masking is a data protection method that obscures sensitive personal
+information to prevent unauthorized access while retaining data utility for analysis or development. Techniques include
+encryption, substitution, and scrambling, ensuring compliance and minimizing breach risks.
+
+## **Module Parameters**
+
+- **Not Applicable (N/A):** There are no direct module parameters specified for the `ner_pii_masking` module.
+
+## **Example config.yaml**
+
+```yaml
+modules:
+  - module_type: ner_pii_masking
+```

docs/source/nodes/passage_filter/passage_filter.md

@@ -51,4 +51,5 @@ maxdepth: 1
 ---
 similarity_threshold_cutoff.md
 similarity_percentile_cutoff.md
+ner_pii_masking.md
 ```

sample_config/full.yaml

@@ -76,6 +76,9 @@ node_lines:
         - module_type: pass_passage_filter
         - module_type: similarity_threshold_cutoff
           threshold: 0.85
+        - module_type: similarity_percentile_cutoff
+          threshold: 0.6
+        - module_type: ner_pii_masking
     - node_type: passage_compressor
       strategy:
         metrics: [retrieval_token_f1, retrieval_token_recall, retrieval_token_precision]

tests/autorag/nodes/passagefilter/test_ner_pii_masking.py

@@ -0,0 +1,15 @@
+from autorag.nodes.passagefilter import ner_pii_masking
+from tests.autorag.nodes.passagefilter.test_passage_filter_base import base_passage_filter_test, contents_example, \
+    ids_example, scores_example, project_dir, previous_result, base_passage_filter_node_test
+
+
+def test_ner_pii_masking():
+    original_ner = ner_pii_masking.__wrapped__
+    contents, ids, scores = original_ner(contents_example, scores_example, ids_example)
+    assert contents[1][3] == "[PER_0] is one of the members of [ORG_34]."
+    base_passage_filter_test(contents, ids, scores)
+
+
+def test_ner_pii_masking_node():
+    result_df = ner_pii_masking(project_dir=project_dir, previous_result=previous_result)
+    base_passage_filter_node_test(result_df)