Improving accuracy for identifying related PubMed queries by an integrated approach

Area of Science:

• Biomedical Informatics
• Information Retrieval
• Computational Linguistics

Background:

• PubMed is a primary resource for biomedical literature searches.
• Users often refine searches through multiple related queries to meet information needs.
• Automatic segmentation of user query sessions is crucial for understanding search behavior.

Purpose of the Study:

• To develop and evaluate a novel approach for segmenting PubMed query sessions and identifying related queries.
• To compare the performance of the integrated approach with previous concept-mapping-based methods.
• To analyze PubMed user search strategies and the nature of information needs.

Main Methods:

• An integrated approach combining lexical and contextual analyses was developed.
• The method was tested on 1539 pairs of consecutive queries from 351 user sessions.
• Performance was compared against a previous method relying solely on concept mapping.

Main Results:

• The integrated approach achieved 90.7% accuracy in predicting related queries, significantly outperforming the previous method.
• Analysis of a one-day PubMed query log revealed that many information needs require multiple queries.
• Most consecutive queries for the same information need were found to be lexically related.

Conclusions:

• The novel approach effectively segments PubMed query sessions and identifies related queries with high accuracy.
• This method offers significant improvements over existing techniques for analyzing biomedical literature search behavior.
• The proposed PubMed distance metric accurately measures contextual similarity between biological terms, aiding in understanding search strategies.