A Relational Model of Semantic Similarity between Words using Automatically Extracted Lexical Pattern Clusters from the Web

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 803–812, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP A Relational Model of Semantic Similarity between Words using Automatically Extracted Lexical Pattern Clusters from the Web Danushka Bollegala ∗ danushka@mi.ci.i. u-tokyo.ac.jp Yutaka Matsuo matsuo@biz-model. t.u-tokyo.ac.jp The University of Tokyo 7-3-1, Hongo, Tokyo, 113-8656, Japan Mitsuru Ishizuka ishizuka@i. u-tokyo.ac.jp Abstract Semantic similarity is a central concept that extends across numerous fields such as artificial intelligence, natural language processing, cognitive science and psychol- ogy. Accurate measurement of semantic similarity between words is essential for various tasks such as, document cluster- ing, information retrieval, and synonym extraction. We propose a novel model of semantic similarity using the semantic relations that exist among words. Given two words, first, we represent the seman- tic relations that hold between those words using automatically extracted lexical pat- tern clusters. Next, the semantic similar- ity between the two words is computed using a Mahalanobis distance measure. We compare the proposed similarity mea- sure against previously proposed seman- tic similarity measures on Miller-Charles benchmark dataset and WordSimilarity- 353 collection. The proposed method out- performs all existing web-based seman- tic similarity measures, achieving a Pear- son correlation coefficient of 0.867 on the Millet-Charles dataset. 1 Introduction Similarity is a fundamental concept in theories of knowledge and behavior. Psychological ex- periments have shown that similarity acts as an organizing principle by which individuals clas- sify objects, and make generalizations (Goldstone, 1994). For example, a biologist would classify a newly found animal specimen based upon the properties that it shares with existing categories of animals. We can then make additional infer- ences on the new specimen using the properties ∗Research Fellow of the Japan Society for the Promotion of Science (JSPS) known for the existing category. As the simi- larity between two objects X and Y increases, so does the probability of correctly inferring that Y has the property T upon knowing that X has T (Tenenbaum, 1999). Accurate measurement of semantic similarity between lexical units such as words or phrases is important for numerous tasks in natural language processing such as word sense disambiguation (Resnik, 1995), synonym extrac- tion (Lin, 1998a), and automatic thesauri gener- ation (Curran, 2002). In information retrieval, similar or related words are used to expand user queries to improve recall (Sahami and Heilman, 2006). Semantic similarity is a context dependent and dynamic phenomenon. New words are constantly being created and existing words are assigned with new senses on the Web. To decide whether two words are semantically similar, it is important to know the semantic relations that hold between the words. For example, the words horse and cow can be considered semantically similar because both horses and cows are useful animals in agriculture. Similarly, a horse and a car can be considered se- mantically similar because cars, and historically horses, are used for transportation. Semantic re- lations such as X and Y are used in agriculture, or X and Y are used for transportation, …

Content truncated to 500 words