Mahtab at SemEval-2017 Task 2: Combination of Corpus-based and Knowledge-based Methods to Measure Semantic Word Similarity

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 256–260, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics Mahtab at SemEval-2017 Task 2: Combination of Corpus-based and Knowledge-based Methods to Measure Semantic Word Similarity Niloofar Ranjbar ni.ranjbar@Mail.sbu.ac.ir Fatemeh Mashhadirajab fa.mashhadi@gmail.com Mehrnoush Shamsfard m-shams@sbu.ac.ir Rayehe Hosseini pour Hosseinipour.r1@gmail.com Aryan Vahid pour aryanvahidpour@gmail.com Faculty of Computer Science and Engineering / Shahid Beheshti University Abstract In this paper, we describe our proposed method for measuring semantic similar- ity for a given pair of words at SemEval- 2017 monolingual semantic word simi- larity task. We use a combination of knowledge-based and corpus-based tech- niques. We use FarsNet, the Persian Word- Net, besides deep learning techniques to extract the similarity of words. We eval- uated our proposed approach on Persian (Farsi) test data at SemEval-2017. outperformed the other participants andIt ranked the first in the challenge. 1 Introduction Semantic similarity represents a special case of se- mantic relatedness: for example, cars and gasoline would seem to be more closely related than, say, cars and bicycles, but the latter pair are certainly more similar(Resnik et al., 1999). Semantic simi- larity has been used in many application in natural language processing. At SemEval-2017 monolin- gual semantic word similarity task, given a pair of words, we have to automatically measure their semantic similarity and score them according to a [0-4] similarity scale where 4 denotes that the two words are synonymous and 0 indicates that they are completely dissimilar(Camacho-Collados et al., 2017). In subtask 1 in which we partici- pated, the two words in the pair belong to the same language. This subtask provides five monolin- gual word similarity datasets in English, German, Italian, Spanish and Farsi. The language whose dataset we used is Farsi. The rest of the paper is organized as follows. Section 2 reviews published work related to the semantic word similarity task. Section 3 explains the proposed algorithm. The experimental results are discussed in Section 4 and the conclusion and future work are reported in Section 5. 2 Related Works There are different methods for finding seman- tic similarity and relation between two words. Christoph(Christoph, 2015)generally divides sim- ilarity measurement techniques into two cate- gories: knowledge-based and corpus-based tech- niques. Some of the available techniques use a combination of these two methods. In knowledge- based techniques, a taxonomy or ontology like WordNet(Miller, 1995) usually is used to ex- tract taxonomic information like path length and depth in the hierarchy(Pilehvar and Nav- igli, 2015). For example, the proposed meth- ods by Resnik(Resnik, 1995), Lin(Lin et al., 1998)and FaITH(Pirr´o and Euzenat, 2010) fall in this category. In corpus-based techniques, usually a large corpus is used to extract sta- tistical information. Christoph(Christoph, 2015) also divides corpus-based methods into two cat- egories. One category contains simple distribu- tional approaches, which check co-occurrences of words like SemSim(Bollegala et al., 2007) and PMI(Church and Hanks, 1990) . Another category contains dense vector representations- based methods, which usually use dimensionally reduction techniques in vector representations. …

Content truncated to 500 words