Best Match: New relevance search for PubMed

Area of Science:

• Biomedical Informatics
• Information Retrieval
• Machine Learning

Background:

• PubMed, a global biomedical literature search engine, faces challenges in retrieving relevant articles due to the rapid growth of scientific publications.
• Traditional search methods, like date sorting, are becoming less effective for users seeking specific information.

Purpose of the Study:

• To introduce and evaluate the "Best Match" algorithm, a novel relevance search algorithm for PubMed.
• To improve the retrieval of relevant biomedical literature for users by leveraging machine learning and user behavior.

Main Methods:

• Developed the "Best Match" algorithm, incorporating user search history and relevance-ranking signals such as article usage, publication date, and article type.
• Trained the algorithm using past user searches and evaluated its performance through benchmarking experiments and real-world user testing.

Main Results:

• The "Best Match" algorithm demonstrated state-of-the-art retrieval performance in benchmarking.
• Real-world testing showed an improved user experience, with a over 20% increase in user click-through rates.
• Post-deployment, searches using the relevance sort order increased by 60%, assisting millions of users weekly.

Conclusions:

• The "Best Match" algorithm significantly enhances PubMed's ability to deliver relevant biomedical literature.
• Increased user adoption and improved search effectiveness highlight the success of this machine learning-driven approach.
• Promoting awareness of this relevance sort option empowers users to find information more efficiently.