How karthikeyagit28/midsem_assignment works
LegalLens is an interactive search system for Indian Supreme Court judgments. A user submits a legal question in the Streamlit app; the app selects a retrieval mode, and the hybrid ranker combines BM25 and dense semantic scores, returning ranked cases with score breakdowns and a relevant passage. Separate capabilities include TF-IDF baseline search, legal query expansion, positional phrase indexing, and benchmark evaluation. The README documents dataset preparation and index/model building; unsampled implementation details are not inferred beyond those documented capabilities.
Main entry points: streamlit_app.py Streamlit search app · evaluation.py IR evaluation · hybrid_ranker.py Hybrid ranking · bm25_retriever.py BM25 retrieval
Components 16
Corpus preparation
- IndicLegalQA data
data/raw/IndicLegalQA_Dataset_10K_Revised.json - Document preparation
src/preprocessing.py - Canonical judgments
data/processed/documents.json
Retrieval and ranking
- Positional index
src/indexing.py - BM25 retrieval
src/bm25_retriever.py - Dense retrieval
src/semantic_retriever.py - TF-IDF baseline
src/tfidf_retriever.py - Hybrid ranking
src/hybrid_ranker.py - Passage extraction
src/passage_retrieval.py - Legal query expansion
src/extensions.py - Score and result helpers
src/utils.py
Search experience
- Streamlit search app
app/streamlit_app.py
Evaluation
- IR evaluation
src/evaluation.py - Evaluation queries
data/processed/queries_test.json - Benchmark results
results/metrics.csv
Other components
- Legal searcher
Connections 25
- Legal searcher → Streamlit search app: submits query
README.md - IndicLegalQA data → Document preparation: prepares records
README.md - Document preparation → Canonical judgments: creates judgments
README.md - Canonical judgments → Positional index: builds postings
README.md - Positional index → BM25 retrieval: supplies postings
src/bm25_retriever.py - Canonical judgments → Dense retrieval: encodes judgments
README.md - Canonical judgments → TF-IDF baseline: fits document matrix
src/tfidf_retriever.py - Streamlit search app → Hybrid ranking: runs hybrid search
app/streamlit_app.py - Streamlit search app → BM25 retrieval: runs lexical search
app/streamlit_app.py - Streamlit search app → Dense retrieval: runs dense search
app/streamlit_app.py - Streamlit search app → TF-IDF baseline: runs baseline search
app/streamlit_app.py - Streamlit search app → Legal query expansion: uses expansion
app/streamlit_app.py - Hybrid ranking → BM25 retrieval: scores lexically
src/hybrid_ranker.py - Hybrid ranking → Dense retrieval: scores semantically
src/hybrid_ranker.py - Hybrid ranking → Passage extraction: extracts passage
src/hybrid_ranker.py - BM25 retrieval → Score and result helpers: formats results
src/bm25_retriever.py - Dense retrieval → Score and result helpers: formats results
src/semantic_retriever.py - TF-IDF baseline → Score and result helpers: formats results
src/tfidf_retriever.py - Streamlit search app → IR evaluation: shows evaluation
- Evaluation queries → IR evaluation: benchmarks rankings
README.md - IR evaluation → TF-IDF baseline: evaluates baseline
src/evaluation.py - IR evaluation → BM25 retrieval: evaluates retrieval
src/evaluation.py - IR evaluation → Dense retrieval: evaluates retrieval
src/evaluation.py - IR evaluation → Hybrid ranking: evaluates fusion
src/evaluation.py - IR evaluation → Benchmark results: exports metrics
README.md