NILC_USP: A Hybrid System for Sentiment Analysis in Twitter Messages

Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 568–572, Atlanta, Georgia, June 14-15, 2013. c⃝2013 Association for Computational Linguistics NILC USP: A Hybrid System for Sentiment Analysis in Twitter Messages Pedro P. Balage Filho and Thiago A. S. Pardo Interinstitutional Center for Computational Linguistics (NILC) Institute of Mathematical and Computer Science, University of S˜ao Paulo S˜ao Carlos - SP, Brazil {balage, taspardo}@icmc.usp.br Abstract This paper describes the NILC USP system that participated in SemEval-2013 Task 2: Sentiment Analysis in Twitter. Our system adopts a hybrid classification process that uses three classification approaches: rule- based, lexicon-based and machine learning approaches. We suggest a pipeline architec- ture that extracts the best characteristics from each classifier. Our system achieved an F- score of 56.31% in the Twitter message-level subtask. 1 Introduction Twitter and Twitter messages (tweets) are a modern way to express sentiment and feelings about aspects of the world. In this scenario, understanding the sen- timent contained in a message is of vital importance in order to understand users behavior and for mar- ket analysis (Java et al., 2007; Kwak et al., 2010). The research area that deals with the computational treatment of opinion, sentiment and subjectivity in texts is called sentiment analysis (Pang et al., 2002). Sentiment analysis is usually associated with a text classification task. Sentiment classifiers are commonly categorized in two basic approaches: lexicon-based and machine learning (Taboada et al., 2011). A lexicon-based classifier uses a lexicon to provide the polarity, or semantic orientation, of each word or phrase in the text. A machine learning clas- sifier learns features (usually the vocabulary) from annotated corpus or labeled examples. In this paper, we present a hybrid system for senti- ment classification in Twitter messages. Our system combines three different approaches: rule-based, lexicon-based and machine learning. The purpose of our system is to better understand the use of a hybrid system in Twitter text and to verify the performance of this approach in an open evaluation contest. 2:Our system participated in SemEval-2013 Task Sentiment Analysis in Twitter (Wilson et al., 2013). The task objective was to determine the sen- timent contained in Twitter messages. The task in- cluded two sub-tasks: a expression-level classifi- cation (Task A) and a message-level classification (Task B). Our system participated in Task B. In this task, for a given message, our system should classify it as positive, negative, or neutral. Our system was coded using Python and the CLiPS Pattern library (De Smedt and Daelemans, 2012). This last library provides the part-of-speech tagger and the SVM algorithm used in this work1. 2 Related work Despite the significant number of works in senti- ment analysis, few works have approached Twit- ter messages. Agarwal et al. (2011) explored new features for sentiment classification of twitter mes- sages. Davidov et al. (2010) studied the use of hashtags and emoticons in sentiment classification. Diakopoulos and Shamma (2010) analyzed the peo- ple’s sentiment on Twitter for first U.S. presidential …

Content truncated to 500 words