Indexing biomedical documents with Bayesian networks and terminologies

Abstract

We proposed a new approach denoted SDIBN (Semantic Documents Indexing using Bayesian Networks) for indexing biomedical documents with terminologies. The main contribution of SDIBN is to use Bayesian Networks (BN) and the probability inference to perform a partial match between documents and biomedical concepts. The biomedical terminologies exploited are MeSH (Medical Subject Headings) thesaurus and SNOMED CT (Systematized Nomenclature of Medicine-Clinical Terms). Our approach exploits also UMLS (Unified Medical Language System) to filter the extracted concepts which allows to keep only relevant concepts. Our contribution also is to use DCG(Discount Cumulative Gain) measure for the first time to evaluate the indexing approaches. The experiments of SDIBN which are performed on subsets of OHSUMED and Cismef collections showed encouraging results.