Using Co-Authoring and Cross-Referencing Information for MEDLINE Indexing

Area of Science:

• Biomedical Informatics
• Information Science
• Medical Literature Analysis

Background:

• The MEDLINE database continuously grows, necessitating efficient tools for automatic indexing.
• Current indexing methods require improvement to handle the increasing volume of scientific publications.

Purpose of the Study:

• To investigate the utility of referencing and authoring links for enhancing automatic indexing.
• To develop and evaluate ranking scores for predicting appropriate Medical Subject Headings (MeSH) terms for new articles.

Main Methods:

• Utilized a JAMA-based subset of MEDLINE for developing ranking scores.
• Designed algorithms leveraging past indexing information from citation and author networks.
• Evaluated performance on an independent dataset of 1000 documents.

Main Results:

• The proposed ranking scores demonstrated improved recall, F-measure, and mean average precision compared to existing methods.
• Cited articles and authors' past contributions were identified as significant features in the top ranking algorithms.
• The hypothesis that past indexing information aids in indexing new material was supported.

Conclusions:

• Leveraging citation and authoring history provides a robust method for improving automatic MeSH term assignment.
• This approach offers significant benefits for managing and organizing the ever-expanding biomedical literature.
• Future work can explore further extensions and optimizations of these indexing techniques.