FCC: Three Approaches for Semantic Textual Similarity

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 631–634, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics BUAP: Three Approaches for Semantic Textual Similarity Maya Carrillo, Darnes Vilari˜no, David Pinto, Mireya Tovar, Saul Le´on, Esteban Castillo Benem´erita Universidad Aut´onoma de Puebla, Faculty of Computer Science 14 Sur & Av. San Claudio, CU Puebla, Puebla, M´exico {cmaya, darnes, dpinto, mtovar}@cs.buap.mx saul.ls@live.com, ecjbuap@gmail.com Abstract In this paper we describe the three approaches we submitted to the Semantic Textual Similar- ity task of SemEval 2012. The first approach considers to calculate the semantic similar- ity by using the Jaccard coefficient with term expansion using synonyms. The second ap- proach uses the semantic similarity reported by Mihalcea in (Mihalcea et al., 2006). The third approach employs Random Indexing and Bag of Concepts based on context vectors. We consider that the first and third approaches ob- tained a comparable performance, meanwhile the second approach got a very poor behav- ior. The best ALL result was obtained with the third approach, with a Pearson correlation equal to 0.663. 1 Introduction Finding the semantic similarity between two sen- tences is very important in applications of natural language processing such as information retrieval and related areas. The problem is complex due to the small number of terms involved in sentences which are tipically less than 10 or 15. Additionally, it is re- quired to “understand” the meaning of the sentences in order to determine the “semantic” similarity of texts, which is quite different of finding the lexical similarity. There exist different works at literature dealing with semantic similarity, but the problem is far to be solved because of the aforementioned issues. In (Mihalcea et al., 2006), for instance, it is pre- sented a method for measuring the semantic simi- larity of texts, using corpus-based and knowledge- based measures of similarity. The approaches pre- sented in (Shrestha, 2011) are based on the Vector Space Model, with the aim to capture the contex- tual behavior, senses and correlation, of terms. The performance of the method is better than the base- line method that uses vector based cosine similarity measure. In this paper, we present three different ap- proaches for the Textual Semantic Similarity task of Semeval 2012 (Agirre et al., 2012). The task is de- scribed as follows: Given two sentences s1 and s2, the aim is to compute how similar s1 and s2 are, returning a similarity score, and an optional confi- dence score. The approaches should provide values between 0 and 5 for each pair of sentences. These values roughly correspond to the following consid- erations, even when the system should output real values: 5: The two sentences are completely equivalent, as they mean the same thing. 4: The two sentences are mostly equivalent, but some unimportant details differ. 3: The two sentences are roughly equivalent, but some important information differs/missing. 2: The two sentences are not equivalent, but share some details. 1: The two sentences are not equivalent, but are on the same …

Content truncated to 500 words