On Combining Text and MeSH Searches to Improve the Retrieval of MEDLINE documents

The MEDLINE database is the world largest repository of bio-medical abstracts. It is a central information entry point for most biologists despite the growing availability of full-text articles on the WWW. Each article is manually annotated by MeSH terms to allow easy access and in order to improve retrieval, the MeSH fields of MEDLINE records were successfully used in the past with pseudo-relevance feedback and MeSH query expansion. However, previous experiments often ignored the MeSH field structure information. This paper investigates the impact of the MEDLINE MeSH field structure on a method that combines text and MeSH searches on a large subset of the MEDLINE database. Robertson’s Offer Weight technique is used to generate MeSH queries. Our method is evaluated within the TREC 2005 Genomics Track on the ad hoc task collection and our results show that this approach does significantly improve retrieval performance. MOTS-CLÉS : information biomédicale, ontologie, « pseudo relevance feedback », combinaison de recherches.