UO_UA: Using Latent Semantic Analysis to Build a Domain-Dependent Sentiment Resource

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 773–778, Dublin, Ireland, August 23-24, 2014. UO UA: Using Latent Semantic Analysis to Build a Domain-Dependent Sentiment Resource Reynier Ortega Adrian Fonseca Carlos Mu˜niz CERPAMID Ave Patricio Lumumba, S/N Santigo de Cuba, Cuba reynier.ortega@cerpamid.co.cu adrian@cerpamid.co.cu Yoan Guti´errez Andr´es Montoyo DLSI, University of Alicante Carretera de San Vicente, S/N Alicante, Spain ygutierrez@dlsi.ua.es montoyo@dlsi.ua.es Abstract In this paper we present our contribution to SemEval-2014 Task 4: Aspect Based Sen- timent Analysis (Pontiki et al., 2014), Sub- task 2: Aspect Term Polarity for Laptop domain. The most outstanding feature in this contribution is the automatic building of a domain-depended sentiment resource using Latent Semantic Analysis. We in- duce, for each term, two real scores that in- dicate its use in positive and negative con- texts in the domain of interest. The aspect term polarity classification is carried out in two phases: opinion words extraction and polarity classification. The opinion words related with an aspect are obtained using dependency relations. These rela- tions are provided by the Stanford Parser1. Finally, the polarity of the feature, in a given review, is determined from the pos- itive and negative scores of each word re- lated to it. The results obtained by our ap- proach are encouraging if we consider that the construction of the polarity lexicon is performed fully automatically. 1 Introduction Hundreds of millions of people and thousands of companies around the world, actively use So- cial Media2. Every day are more amazing web- sites and applications (Facebook, Twitter, MyS- pace, Amazon, etc.) that allow the easy sharing of information in near real time. For this rea- son, at present, the Web is flooded with subjec- tive, personal and affective data. Mining this huge This work is licensed under a Creative Commons At- tribution 4.0 International Licence. Page numbers and pro- ceedings footer are added by the organisers. Licence details: http://creativecommons.org/licenses/by/4.0/ 1http://nlp.stanford.edu:8080/parser/ 2http://en.wikipedia.org/wiki/Social media volume of information offer both interesting chal- lenges and useful intelligent applications such as recommendation systems (Dong et al., 2013; Sun et al., 2009) and customer’s reviews summariza- tion (Bafna and Toshniwal, 2013; Balahur and Montoyo, 2008). Nowadays, companies have redirected their marketing strategies toward the Web. Each one of them advertises that their products are the best, amazing, easy to use, long lasting and cheap. But are these advertisements really true? Obviously, not everything is true. The companies usually ex- aggerate the product’s quality and in many cases tend not to advertise the limitations of their prod- ucts. Therefore, taking a rational decision about which product is the best among the variety of ex- isting options can be very stressful. To avoid this situation, frequently we trust in the experiences gained by others who have pur- chased the product of our interest, or one similar. The existence of websites like Ciao3, Epinions4 and Cnet5 make possible to the customers to inter- change their experiences about a specific product, and to future clients avoid products advertising However, the existence …

Content truncated to 500 words