PolyUCOMP: Combining Semantic Vectors with Skip bigrams for Semantic Textual Similarity

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 524–528, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics PolyUCOMP: Combining Semantic Vectors with Skip bigrams for Semantic Textual Similarity Jian Xu Qin Lu Zhengzhong Liu The Hong Kong Polytechnic University Department of Computing Hung Hom, Kowloon, Hong Kong {csjxu, csluqin, hector.liu}@comp.polyu.edu.hk Abstract This paper presents the work of the Hong Kong Polytechnic University (PolyUCOMP) team which has participated in the Semantic Textual Similarity task of SemEval-2012. The PolyUCOMP system combines semantic vec- tors with skip bigrams to determine sentence similarity. The semantic vector is used to compute similarities between sentence pairs using the lexical database WordNet and the Wikipedia corpus. The use of skip bigram is to introduce the order of words in measuring sentence similarity. 1 Introduction Sentence similarity computation plays an im- portant role in text summarization, classification, question answering and social network applica- tions (Lin and Pantel, 2001; Erkan and Radev, 2004; Ko et al., 2004; Ou et al., 2011). The SemEval 2012 competition includes a task targeted at Semantic Textual Similarity (STS) between sen- tence pairs (Eneko et al., 2012). Given a set of sen- tence pairs, participants are required to assign to each sentence pair a similarity score. Because a sentence has only a limited amount of content words, it is not easy to determine sentence similarities because of the sparseness issue. Hatzivassiloglou et al. (1999) proposed to use lin- guistic features as indicators of text similarity to address the problem of sparse representation of sentences. Mihalcea et al. (2006) measured sen- tence similarity using component words in sen- tences. Li et al. (2006) proposed to incorporate the semantic vector and word order to calculate sen- tence similarity. In our approach to the STS task, semantic vector is used and the semantic relatedness between words is derived from two sources: WordNet and Wikipedia. Because WordNet is limited in its cov- erage, Wikipedia is used as a candidate for deter- mining word similarity. Word order, however, is not considered in se- mantic vector. As semantic information are coded in sentences according to its order of writing, and in our systems, content words may not be adjacent to each other, we proposed to use skip bigrams to represent the structure of sentences. Skip bigrams, generally speaking, are pairs of words in a sen- tence order with arbitrary gap (Lin and Och, 2004a). Different from the previous skip bigram statistics which compare sentence similarities through overlapping skip bigrams (Lin and Och, 2004a), the skip bigrams we used are weighted by a decaying factor of the skipping gap in a sentence, giving higher scores to closer occurrences of skip bigrams. It is reasonable to assume that similar sentences should have more overlapping skip bi- grams, and the gaps in their shared skip bigrams should also be similar. The rest of this paper is organized as followed. Section 2 describes sentence similarity using se- mantic vectors and the order-sensitive skip bigrams. Section 3 gives the performance evaluation. …

Content truncated to 500 words