Combining MeSH Thesaurus with UMLS in pseudo relevance feedback to improve biomedical information retrieval

With the rapid increase in the volume of biomedical publications, developing an efficient search strategy to retrieve relevant biomedical documents that match the user search intention is a tremendous challenge. This paper proposes a novel pseudo relevance feedback technique which combines MeSH terms and UMLS concepts to improve the performance of retrieving biomedical documents from MEDLINE. The MeSH terms that annotate the feedback documents and the UMLS concepts derived from the feedback documents are used to help retrieve publications relevant to the user query. Extensive performance studies using the OHSUMED collection show that the proposed pseudo relevance feedback scheme using both MeSH terms and UMLS concepts improves the retrieval performance by 43.3% over the approach based on unexpanded query in terms of mean average precision. We have integrated the proposed strategy into G-Bean, which is publicly accessible at http://bioinformatics.clemson.edu:8080/G-Bean/index.jsp.