Fine-grained indexing of the biomedical literature: MeSH subheading attachment for a MEDLINE indexing tool.

Objective: This paper reports on the latest results of an Indexing Initiative effort addressing the automatic attachment of subheadings to MeSH main headings recommended by the NLM's Medical Text Indexer.

Material And Methods: Several linguistic and statistical approaches are used to retrieve and attach the subheadings. Continuing collaboration with NLM indexers also provided insight on how automatic methods can better enhance indexing practice.

Results: The methods were evaluated on corpus of 50,000 MEDLINE citations. For main heading/subheading pair recommendations, the best precision is obtained with a post-processing rule method (58%) while the best recall is obtained by pooling all methods (64%). For stand-alone subheading recommendations, the best performance is obtained with the PubMed Related Citations algorithm.

Conclusion: Significant progress has been made in terms of subheading coverage. After further evaluation, some of this work may be integrated in the MEDLINE indexing workflow.