PubMed Search System with Efficient Semantic Network Construction Algorithm

Abstract

Physicians and biomedical researchers consistently use PubMed for finding health information. To retrieve satisfactory answers, formulation of a wellorganize query is essential. However, it is a significant obstacle to the PubMed users in practice because it needs much time and effort. A semantic network is an excellent way to break down barriers to accessing the information from the database thus previous works exploit semantic network in information retrieval from the biomedical citations. However, a problem of existing semantic network methods is that they typically produce much unnecessary information, which frustrates users with exploring concepts and relations in the network arising from the search result. In this paper, we introduce an efficient online semantic construction method, COMPACT, to leave out unnecessary relations, Redundant relationship, in the semantic network. We find Related relationship in the semantic network and simplify its relationships. Also, we offer to visualization of the semantic network in which one can easily explore the concepts and relations among the concepts to expand their queries. We conducted experiments using MEDLINE, which is a database including bibliographic citations of life sciences and biomedical information. Our experiments showed that COMPACT was more efficient than the naive approach.We implemented and integrated our method into Refmed, which is a relevance-feedback based PubMed search engine at "http://dm.postech.ac.kr/refmed."