Ontology Based Biomedical Information Retrieval

In recent decades, along with the increasing of biomedical scientific literature,for example600000new articles had been added into the database in2008. Retrieverelevant information in this huge database becomes more and more important, and alsochallenging, the efficiency biomedical information retrieval technology is urgentlyneeded.Biomedical information retrieval has got more and more attention ofresearchers.Biomedical text usually contains many professional terms and abbreviations,there are a large number of synonyms. This makes the traditional keyword-matchingbased information retrieval methods perform very poor, so it is necessary to studysemantic based biomedical text retrieval method.This thesis mainly focuses on biomedical information retrieval method based onsemantic similarity. Firstly, the state of the art of biomedical information retrieval isreviewed, then different traditional text retrieval models are compared on theirperformance of biomedical information retrieval. According to the characteristics of theretrieval task, the asymmetric semantic similarity is firstly proposed based on MeSH.Since the MeSH medical subject headings do not cover the contents of the article well,biomedical retrieval based on semantic similarity performs poor. Therefore a newretrieval method is proposed to combine the semantic similarity based retrieval with thetraditional text retrieval. Experimental results indicate that the combined method canachieve significant improvement.The main contributions of this thesis include:First, the differences of traditional text retrieval models on the performance of thebiomedical text information retrieval are analysised and compared.Second, the semantic similarity method based on MeSH biomedical ontology isstudyed on. According to the characteristics of the retrieval task, the asymmetricsemantic similarity method is firstly proposed and two kinds of methods in obtainingMeSH terms is explored too, one is the use of MeSHUp tools, the other is the use ofpseudo relevance feedback technology. The experimental results show that the asymmetric semantic similarity improves the performance of semantic similarity basedretrieval.Third, the two combined method of traditional text retrieval and semanticsimilarity based retrieval is studyed on, the weighted linear combination and re-rankingcombination. The experimental results show that the combined method of semanticsimilarity and traditional text retrieval achieve significantly performance increasing, andthe linear weighted method can achieve better retrieval performance.