Amrita_CEN at SemEval-2016 Task 1: Semantic Relation from Word Embeddings in Higher Dimension

Proceedings of SemEval-2016, pages 706–711, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics Amrita CEN at SemEval-2016 Task 1: Semantic Relation from Word Embeddings in Higher Dimension Barathi Ganesh HB Artificial Intelligence Practice Tata Consultancy Services Kochi - 682 042 Kerala, India barathiganesh.hb@tcs.com Anand Kumar M and Soman KP Center for Computational Engineering and Networking Amrita School of Engineering, Coimbatore Amrita Vishwa Vidyapeetham Amrita University, India m anandkumar@cb.amrita.edu kp soman@.amrita.edu Abstract Semantic Textual Similarity measures similar- ity between pair of texts, even though the sim- ilar context is projected using different words. This work attempted to incorporate the con- text space of the sentence from that sentence alone. It proposes combination of Word2Vec and Non-Negative Matrix Factorization to rep- resent the sentence as context embedding vec- tor in context space. Distance and correla- tion values between context embedding vector pairs used as a features for Support Vector Re- gression to built the domain independent sim- ilarity measuring model. The proposed model yielding performance 0.41 in terms of correla- tion. 1 Introduction Semantic Textual Similarity (STS) assess the degree to which two snippets of text mean the same thing (Agirrea et al., 2015). The modules developed for successful STS systems have a broad range of po- tential applications including: Discourse Analysis, Information Retrieval, Machine Reading, Machine Translation, Question Answering, Text Summariza- tion and Plagiarism Detection. Degree of dependence between sentences vary even-though there exist similar words present in them (Example 1) whereas the dependence remains unchanged when the context is being projected with different words (Example 2). For instance, S1 : Boy chases the cat. S2 : Cat chases the boy. Example 1 S1 : The rat jumps inside the tub. S2 : Mouse dives into the vessel. Example 2 From the above example, representing sentence as context dependent vector in a context space is more informative than the traditional frequency based rep- resentation methods. Thus by considering this, the proposed approach measures the similarity between the sentences as prescribed in STS task, which is given in the Table 1. From the Table 1 it is clear that simple frequency based representation will fail to achieve the objec- tive. Our approach proposes measuring similarity based on contextual information provided by the other words in the sentences instead of measuring similarity using just the words. Words tends to have different meaning with respect to their appearance with context. In this proposed approach, sentence embedding will be found from word embedding in which words are represented as word embedding vectors with re- spect to context they occurs. Thereafter the simi- larity measure is done by finding correlation of the features in the sentence embedding. Remaining paper details about the related works done on STS in section 2, detailed mathematical ex- planation is given in section 3 and statistics about the data-set, experiment and observations are explained in section 4. 2 Related Works In this section we discuss the related works carried on the STS and how the current proposal has …

Content truncated to 500 words