A passage retrieval method based on probabilistic information retrieval model and UMLS concepts in biomedical...

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Information Retrieval

Background:

• Passage retrieval is vital for biomedical question answering (QA) systems.
• Open-domain QA faces challenges, particularly in the biomedical field.
• Existing methods require further refinement for effective biomedical QA.

Purpose of the Study:

• To introduce an advanced biomedical passage retrieval method.
• To enhance the accuracy and efficiency of biomedical QA systems.

Main Methods:

• Utilized Stanford CoreNLP for sentence splitting and passage length analysis.
• Employed a probabilistic information retrieval (IR) model (BM25) with stemmed words and Unified Medical Language System (UMLS) concepts.
• Integrated PubMed search and UMLS similarity for initial document retrieval.

Main Results:

• The proposed method demonstrated strong performance on BioASQ datasets.
• Achieved a significant improvement of 6.84% in mean average precision (MAP) over state-of-the-art methods.
• Outperformed existing approaches in identifying relevant passages.

Conclusions:

• Developed an efficient and effective passage retrieval technique for biomedical QA.
• The method offers high mean average precision for retrieving relevant biomedical information.
• This approach contributes to advancing the capabilities of biomedical question answering systems.