SSA-UO: Unsupervised Sentiment Analysis in Twitter

Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 501–507, Atlanta, Georgia, June 14-15, 2013. c⃝2013 Association for Computational Linguistics SSA-UO: Unsupervised Twitter Sentiment Analysis Reynier Ortega, Adrian Fonseca CERPAMID, University of Oriente Ave Patricio Lumumba S/N Santiago de Cuba, Cuba Yoan Guti´errez DI, University of Matanzas Autopista a Varadero Km 31 2 Matanzas, Cuba Andr´es Montoyo DLSI, University of Alicante Carretera de San Vicente S/N Alicante,Spain Abstract This paper describes the specifications and re- sults of SSA-UO, unsupervised system, pre- sented in SemEval 2013 for Sentiment Analy- sis in Twitter (Task 2) (Wilson et al., 2013). The proposal system includes three phases: data preprocessing, contextual word polarity detection and message classification. The preprocessing phase comprises treatment of emoticon, slang terms, lemmatization and POS-tagging. Word polarity detection is car- ried out taking into account the sentiment as- sociated with the context in which it appears. For this, we use a new contextual sentiment classification method based on coarse-grained word sense disambiguation, using WordNet (Miller, 1995) and a coarse-grained sense in- ventory (sentiment inventory) built up from SentiWordNet (Baccianella et al., 2010). Fi- nally, the overall sentiment is determined us- ing a rule-based classifier. As it may be ob- served, the results obtained for Twitter and SMS sentiment classification are good consid- ering that our proposal is unsupervised. 1 Introduction The explosion of Web 2.0 has marked a new age for the human society. The huge use of Social Me- dia such as Facebook1 , MySpace2 , LinkedIn3 and Twitter4 , offers a place for people to share informa- tion in real time. Twitter is one of the most popular 1https://www.facebook.com 2http://www.myspace.com/ 3http://www.linkedin.com 4https://www.twitter.com/ social network websites and has been growing at a very fast pace. The number of active users exceeds 500 million and the number of tweets posted by day exceeds 500 million (as of May 2012)5. Through the twitter applications, users shared opinions about per- sonalities, politicians, products, companies, events, etc. This has been attracting the attention of dif- ferent research communities interested in analyz- ing its content and motivated many natural language tasks, such as sentiment analysis, emotions detec- tion, opinions retrieval, product recommendation or opinion summarization. One of the most popular sentiment analysis tasks is polarity classification. This task is a new field that classifies opinion texts as positive, negative or neutral (Pang et al., 2002; Turney, 2002; Esuli and Sebastiani, 2006; Wilson et al., 2006; Wiegand et al., 2010). Determining polarity might seem an easy task, as many words have some polarity by them- selves. However, words do not always express the same sentiment, and in most cases the polarity of a word depends on the context in which the word is used. So, terms that clearly denote negative feel- ings can be neutral, or even positive, depending on their context. Hence, sentiment analysis sys- tems should include semantic-level analysis in order to solve word ambiguity and correctly capture the meaning of each word …

Content truncated to 500 words