BioDI: A new approach to improve biomedical documents indexing

Abstract

The partial match between biomedical documents and controlled vocabularies allows to find in the documents more terms variants than those existing in the dictionaries. However, it generates irrelevant information. We propose a new approach for indexing biomedical documents with the Medical Subject Headings (MeSH) thesaurus that aims to overcome the limitation of the partial match. In fact, our indexing approach proposes to restrict the stemming process in the step of pretreatment. The step of the descriptors extraction is based essentially on the vector space model and combines semantic and statistic methods to compute a score to estimate the relevance of a descriptor given a document. The knowledge provided by the Unified Medical Language System (UMLS) is used then for filtering. The filtering method aims to keep only relevant descriptors. The experiments of our approach that have been carried out on the OHSUMED collection, showed very encouraging results. © 2013 Springer-Verlag.