Method of Extracting Is-A and Part-Of Relations Using Pattern Pairs in Mass Corpus

Method of Extracting Is-A and Part-Of Relations Using Pattern Pairs in Mass Corpus  Se-Jong Kim, Yong-Hun Lee, and Jong-Hyeok Lee Department of Computer Science and Engineering, Pohang University of Science and Technology (POSTECH), San 31, Hyoja-dong, Nam-gu, Pohang, 790-784, Republic of Korea {sejong, yhlee95, jhlee}@postech.ac.kr Abstract. This paper proposes a method that extracts term pairs satisfying is-a relations or part-of relations from a mass corpus using pairs of patterns sharing a term. We extracted reliable single patterns and pattern pairs using some term pairs that satisfy the target relation, and extracted reliable term pairs using these patterns. The extracted term pairs were used to extract new single patterns and pattern pairs, and we repeated these steps several times. The proposed method achieved 71.5% accuracy in detecting is-a relations and 88% accuracy in detecting part-of relations, and extracted 144 new is-a relations and 85 new part-of relations which could not be extracted using single patterns. These results are useful in constructing an ontology and a thesaurus because these language knowledge bases consist mainly of is-a relations and part-of relations. Keywords: Bootstrapping algorithm, shared term, reliability equation, subjective evaluation, morphological analysis.  The work reported in this paper was supported in part by the Korea Science and Engineering Foundation (KOSEF) grant funded by the Korean government (MEST No. 2009-0075211), and in part by the BK 21 Project in 2009. Copyright 2009 by Se-Jong Kim, Yong-Hun Lee, and Jong-Hyeok Lee 1 Introduction An ontology is a language knowledge base that organizes categories of concepts, and defines relations among the categories. The ontology is useful in semantic analysis of natural language processing (NLP) and information retrieval (IR). Methods that construct the ontology fall into two categories: those that combine existing ontologies and thesauruses (Maedche and Staab, 2002) and those that semi-automatically construct the ontology using a mass corpus (Kavalec and Svatek, 2005). The first method focuses on expanding the ontology using existing resources, and the second method focuses on constructing new ontologies. Particularly, constructing the ontology using a mass corpus begins with automatic extraction of semantic relations. Extraction of semantic relations is a process that extracts term pairs that satisfy a specified target relation. The most common approach in this process is a pattern-based approach for is-a and part-of relations (Berland and Charniak, 1999; Girju et al., 2003; Hearst, 1992; Pantel and Pennacchiotti, 2006; Ravichandran and Hovy, 2002). This approach regards words appearing between terms as patterns, and extracts term pairs using the patterns. Hearst (1992) extracts is-a relations using manual patterns, and proposes a bootstrapping algorithm using seed term pairs. To extract part-of relations, Berland and Charniak (1999) measure the reliabilities of terms using the frequencies of terms that have the concept of “whole” or “part,” and Girju et al. (2003) uses a lexical database (WordNet) and a decision tree. Ravichandran and Hovy (2002) extract semantic relations for various terms in a question answering (QA) system. Pantel and 260 23rd Pacific Asia Conference on Language, Information and Computation, pages 260–268 Pennacchiotti (2006) …

Content truncated to 500 words