Improved biomedical term selection in pseudo relevance feedback.

Area of Science:

• Biomedical Informatics
• Information Retrieval
• Natural Language Processing

Background:

• Biomedical literature is rapidly expanding, making information retrieval complex.
• Users struggle to formulate precise queries for accurate information retrieval.
• Query expansion is crucial for enhancing the relevance of search results.

Purpose of the Study:

• To propose and evaluate a novel term selection algorithm for query expansion in biomedical information retrieval.
• To assess the impact of pre-processing techniques on retrieval performance in the biomedical domain.

Main Methods:

• Implementation of document frequency chi-square, a refined chi-square method, for term selection within pseudo-relevance feedback.
• Comparative analysis against state-of-the-art term selection algorithms.
• Evaluation of pre-processing steps, including stemming, on retrieval system performance.

Main Results:

• The proposed document frequency chi-square algorithm demonstrated an average performance improvement of 88% over existing methods.
• Stemming was found to decrease the overall performance of pseudo-relevance feedback in biomedical information retrieval.

Conclusions:

• Document frequency chi-square is a highly effective method for term selection in biomedical query expansion.
• Pre-processing, particularly stemming, can negatively impact the performance of pseudo-relevance feedback systems in this domain.