Using a Weighted Semantic Network for Lexical Semantic Relatedness

Proceedings of Recent Advances in Natural Language Processing, pages 610–618, Hissar, Bulgaria, 7-13 September 2013. Using a Weighted Semantic Network for Lexical Semantic Relatedness Reda Siblini Concordia University 1400 de Maisonneuve Blvd. West Montreal, Quebec, Canada, H3G 1M8 r_sibl@encs.concordia.ca Leila Kosseim Concordia University 1400 de Maisonneuve Blvd. West Montreal, Quebec, Canada, H3G 1M8 kosseim@encs.concordia.ca Abstract The measurement of semantic relatedness between two words is an important met- ric for many natural language processing applications. In this paper, we present a novel approach for measuring semantic re- latedness that is based on a weighted se- mantic network. This approach explores the use of a lexicon, semantic relation types as weights, and word definitions as a basis to calculate semantic related- ness. Our results show that our approach outperforms many lexicon-based methods to semantic relatedness, especially on the TOEFL synonym test, achieving an accu- racy of 91.25%. 1 Introduction Lexical semantic relatedness is a measurement of how two words are related in meaning. Many natural language processing applications such as textual entailment, question answering, or infor- mation retrieval require a robust measurement of lexical semantic relatedness. Current approaches to address this problem can be categorized into three main categories: those that rely on a lexicon and its structure, those that use the distributional hypothesis on a large corpus, and hybrid ap- proaches. In this paper, we propose a new lexicon-based approach to measure semantic relatedness that is based on a weighted semantic network that in- cludes all 26 semantic relations found in WordNet in addition to information found in the glosses. 2 Related Work Approaches to computing semantic relatedness can be classified into three broad categories: lexicon-based, corpus-based, and hybrid ap- proaches. Lexicon-based methods use the features of a lexicon to measure semantic relatedness. The most frequently used lexicon is Princeton’s WordNet (Fellbaum, 1998) which groups words into synonyms sets (called synsets) and includes various semantic relations between those synsets, in addition to their definitions (or glosses). WordNet contains 26 semantic relations that include: hypernymy, hyponymy, meronymy, and entailment. To measure relatedness, most of the lexicon-based approaches rely on the structure of the lexicon, such as the semantic link path, depth (Leacock and Chodorow, 1998; Wu and Palmer, 1994), direction (Hirst and St-Onge, 1998), or type (Tsat- saronis et al., 2010). Most of these approaches exploit the hypernym/hyponym relations, but a few approaches have also included the use of other semantic relations. Leacock and Chodorow (1998) for example, computed semantic related- ness as the length of the shortest path between synsets over the depth of the taxonomy. Wu and Palmer (1994) also used the hyponym tree to calculate relatedness by using the depth of the words in the taxonomy and the depth of the least common superconcept between the two words. Hirst and St-Onge (1998), on the other hand, used the lexical chains between words based on their synsets and the semantic edges that connect them. In addition to using the hypernym relations, they classified the relations into classes: “extra strong” for …

Content truncated to 500 words