NTNU: An Unsupervised Knowledge Approach for Taxonomy Extraction

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 938–943, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics NTNU: An Unsupervised Knowledge Approach for Taxonomy Extrac- tion Bamfa Ceesay Department of Computer Science and Information Engineering National Taiwan Normal University No. 88, Tingz Chou Road, Section 4, Taipei 116, Taiwan, R.O.C. bmfceesay@csie.ntnu.edu.tw Wen Juan Hou Department of Computer Science and Information Engineering National Taiwan Normal University No. 88, Tingz Chou Road, Section 4, Taipei 116, Taiwan, R.O.C. emilyhou@csie.ntnu.edu.tw Abstract Taxonomy structures are important tools in the science of classification of things or concepts, including the principles that un- derlie such classification. This paper pre- sents an approach to the problem of taxonomy construction from texts focusing on the hyponym-hypernym relation between two terms. Given a set of terms in a particu- lar domain, the approach in this study uses Wikipedia and WordNet as knowledge sources and applies the information extrac- tion methods to analyze and establish the hyponym-hypernym relationship between two terms. Our system is ranked fourth among the participating systems in SemEval-2015 task 17. 1 Introduction Taxonomies are essential tools for many Natural Language Processing (NLP) applications and the backbone of many structured knowledge re- sources. Taxonomies specific to a domain are becoming indispensable to a growing number of applications (Velardi et al., 2013). Several state- of-the-art approaches already exist to extract tax- onomies to characterize the domains of interest from the corpus using the information extraction techniques. Recently, attention has been devoted to inducing the taxonomy from a set of keyword phrases instead of from a text corpus (Liu et al., 2012). Such approaches enrich the set of key- word phrases by aggregating search results for each keyword phrase into a text corpus to over- come the lack of explicit relationships between keyword phrases from which the taxonomy can be induced. This approach faces a key challenge of ex- tracting explicit relationships among keyword phrases. However, semantic relatedness between concepts in a domain is an important clue to ex- tracting their taxonomy relationships. An im- portant contribution in relation to this is reported by Gabrilovich et al. (2007) that present an ex- plicit semantic analysis using the natural con- cepts and propose a uniform method of computing relatedness of both individual concept and arbitrarily long text fragments. Lexical data- bases such as WordNet (Miller, 1995) encode relations between words such as synonymy and hypernymy. Quite a few metrics have been de- fined that compute relatedness using various properties of the underlying graph structures of these resources. The obvious drawback of this approach is that the creation of lexical resources requires the lexicographic expertise as well as a lot of time and effort, and consequently such re- sources cover only a small fragment of the lan- guage lexicon. Specifically, such the resources contain few proper names, neologisms, slang, and domain-specific technical terms. Further- more, these resources have strong lexical orienta- tion and mainly contain information about individual words but little world knowledge in general. …

Content truncated to 500 words