saloni-1919/biosum-reliable
AI-powered biomedical text summarization using extractive NLP, biomedical entity recognition, and transformer-based abstractive summarization.
What's novel
AI-powered biomedical text summarization using extractive NLP, biomedical entity recognition, and transformer-based abstractive summarization.
Code Analysis
10 files read · 3 roundsA FastAPI web service that performs extractive summarization of biomedical text using custom TF-IDF scoring with section-aware weighting, regex-based keyword entity matching, and an optional (but minimally fine-tuned) BART abstractive model.
Strengths
The extractive summarizer is genuinely implemented with thoughtful heuristics (section weighting, biomedical term boosting, redundancy filtering via Jaccard similarity, numeric content detection). The API layer is clean, well-structured, and follows FastAPI best practices with proper validation and error codes.
Weaknesses
The 'biomedical entity recognition' is a hardcoded regex keyword list of ~40 terms, not a real NER system. The abstractive model is a 200-sample, 1-epoch fine-tune of BART-CNN that adds negligible value, and several listed dependencies (spaCy, scispacy, sentence-transformers, faiss) are never used in the code.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
11
Contributors
1
Files
30
Active weeks
2
Repository
Language
Python
Stars
3
Forks
0
License
MIT