SemBioNLQA

SemBioNLQA performs semantic biomedical question answering by classifying natural language queries and retrieving and extracting precise answers from PubMed using UMLS-based similarity and BM25 ranking.

Key Features:

• Question Classification: Employs handcrafted lexico-syntactic patterns and machine learning to classify questions into yes/no, factoid, list, and summary categories.
• Document Retrieval: Uses the PubMed search engine combined with UMLS similarity measures to retrieve relevant biomedical documents.
• Passage Retrieval: Applies the BM25 model with stemmed words and UMLS concepts to pinpoint relevant passages within retrieved documents.
• Answer Extraction: Integrates the UMLS Metathesaurus and BioPortal synonyms with sentiment analysis and term-frequency metrics to extract concise and paragraph-sized answers.
• Supported Question Types: Handles yes/no, factoid, list, and summary question types.
• Performance Evaluation: Evaluated on BioASQ datasets (2015–2017) demonstrating competitive performance in biomedical QA benchmarks.

Scientific Applications:

• Clinical decision support: Provides exact answers (e.g., yes/no and specific biomedical entities) to support clinical queries.
• Literature synthesis and review: Produces ideal, paragraph-sized summaries to assist comprehensive literature reviews.
• Biomedical QA benchmarking: Serves as a system for evaluation and comparison in challenges such as BioASQ.

Methodology:

The pipeline performs question classification using handcrafted lexico-syntactic patterns and machine learning, document retrieval via PubMed with UMLS similarity, passage retrieval using BM25 with stemming and UMLS concepts, and answer extraction using the UMLS Metathesaurus, BioPortal synonyms, sentiment analysis, and term-frequency metrics.