Improve Biomedical Information Retrieval Using Modified Learning to Rank Methods.

Area of Science:

• Biomedical Informatics
• Information Retrieval
• Computational Biology

Background:

• The exponential increase in biomedical literature presents significant challenges for manual information extraction by biologists.
• Existing information retrieval technologies struggle with the specialized terminology prevalent in the biomedical domain.

Purpose of the Study:

• To propose a novel framework for enhancing biomedical information retrieval using learning to rank techniques.
• To address the challenge of domain-specific terminologies by developing ranking models that ensure both relevance and diversity in search results.

Main Methods:

• Developed a framework based on learning to rank, incorporating novel document labeling strategies.
• Constructed ranking models designed to retrieve relevant documents and diversify search results for improved completeness.
• Combined traditional retrieval models as features for model training and investigated various learning to rank approaches.

Main Results:

• Experimental results on TREC Genomics datasets demonstrated the effectiveness of the proposed framework.
• The framework successfully enhanced biomedical information retrieval by improving result relevance and diversity.
• The proposed document labeling strategies and ranking models proved beneficial for the task.

Conclusions:

• The developed learning to rank framework offers an effective solution for improving biomedical information retrieval.
• Diversifying search results alongside relevance is crucial for comprehensive information access in specialized domains.
• The approach shows promise for assisting biologists in navigating the vast biomedical literature more efficiently.