SU-Sentilab : A Classification System for Sentiment Analysis in Twitter

Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 471–477, Atlanta, Georgia, June 14-15, 2013. c⃝2013 Association for Computational Linguistics SU-Sentilab : A Classification System for Sentiment Analysis in Twitter Gizem Gezici, Rahim Dehkharghani, Berrin Yanikoglu, Dilek Tapucu, Yucel Saygin Sabanci University Istanbul, Turkey 34956 {gizemgezici,rdehkharghani,berrin,dilektapucu,ysaygin}@sabanciuniv.edu Abstract Sentiment analysis refers to automatically ex- tracting the sentiment present in a given natu- ral language text. We present our participation to the SemEval2013 competition, in the senti- ment analysis of Twitter and SMS messages. Our approach for this task is the combination of two sentiment analysis subsystems which are combined together to build the final sys- tem. Both subsystems use supervised learning using features based on various polarity lexi- cons. 1 Introduction Business owners are interested in the feedback of their customers about the products and services pro- vided by businesses. Social media networks and micro-blogs such as Facebook and Twitter play an important role in this area. Micro-blogs allow users share their ideas with others in terms of small sen- tences; while Facebook updates may indicate an opinion inside a longer text. Automatic sentiment analysis of text collected from social media makes it possible to quantitatively analyze this feedback. In this paper we describe our sentiment analy- sis system identified as SU-Sentilab in the SemEval 2013 competition, Task 2: Sentiment analysis in Twitter. One or the problems in this competition was to label a given tweet or sms message with the cor- rect sentiment orientation, as positive, negative or neutral. In the second task of the same competition, the polarity of a given word or word sequence in the message was asked. Details are described in (Man- andhar and Yuret, 2013). We participated in both of these tasks using a classifier combination consisting of two sub-systems that are based on (Dehkharghani et al., 2012)(Gezici et al., 2012) and adapted to the tweet domain. Both sub-systems use supervised learning in which the system is trained using tweets with known polari- ties and used to predict the label (polarity) of tweets in the test set. Both systems use features that are based on well-known polarity resources namely SentiWordNet (Baccianella et al., 2010), SenticNet (Cambria et al., 2012) and NRC Emotion Lexicon (Mohammad, 2012). Also a set of positive and neg- ative seed words (Liu et al., 2005) is used in feature extraction. The remainder of paper is organized as follows: Related works are presented in Section 2; the pro- posed approach is described in Section 3 and exper- imental evaluation is presented in Section 4. 2 Related Works There has been much work on sentiment analysis in the last ten years (Riloff and Wiebe, 2003) (Wilson et al., 2009) (Taboada et al., 2011) (Pang and Lee, 2008). The two fundamental methods for sentiment analysis are lexicon-based and supervised methods. The lexicon-based technique adopts the idea of de- termining the review sentiment by obtaining word polarities from a lexicon, such as the SentiWordNet …

Content truncated to 500 words