Serendio: Simple and Practical lexicon based approach to Sentiment Analysis

Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 543–548, Atlanta, Georgia, June 14-15, 2013. c⃝2013 Association for Computational Linguistics Serendio: Simple and Practical lexicon based approach to Sentiment Analysis Prabu Palanisamy, Vineet Yadav and Harsha Elchuri Serendio Software Pvt Ltd Guindy, Chennai 600032, India {prabu, vineet, harsha}@serendio.com Abstract This paper describes the system developed by the Serendio team for the SemEval-2013 Task 2 competition (Task A). We use a lexicon based approach for discovering sentiments. Our lexicon is built from the Serendio tax- onomy. The Serendio taxonomy consists of positive, negative, negation, stop words and phrases. A typical tweet contains word varia- tions, emoticons, hashtags etc. We use prepro- cessing steps such as stemming, emoticon de- tection and normalization, exaggerated word shortening and hashtag detection. After the preprocessing, the lexicon-based system clas- sifies the tweets as positive or negative based on the contextual sentiment orientation of the words. Our system yields an F-score of 0.8004 on the test dataset. 1 Introduction Social media websites like Twitter, Facebook etc. are a major hub for users to express their opinions online. On these social media sites, users post com- ments and opinions on various topics. Hence these sites become rich sources of information to mine for opinions and analyze user behavior and provide in- sights for: • User behavior • Product feedback • User intentions • Lead generation Businesses spend an enormous amount of time and money to understand their customer opinions about their products and services. Thus Sentiment Analysis has become a hot research area since 2002. Sentiment Analysis is used to determine sentiments, emotions and attitudes of the user. The text used for analysis can range from big document (e.g. Product reviews from Amazon, blogs) to small status mes- sage (e.g. Tweets, Facebook comments). In this pa- per, we confine to Twitter data i.e classify a tweet to have a positive, negative or neutral sentiment. The rest of the paper is organized as follows. In Section 2, we study relevant previous work on Sen- timent Analysis on Twitter data. In Section 3, we describe each processing step of our approach in de- tail. In Section 4, we experiment with the training and the lexicon. In Section 5, we report and evaluate the final result obtained from the test data published by the SemEval team. In Section 6, we present our conclusions and outline our future work. 2 Related Work Sentiment Analysis on raw text is a well known problem. The Liu (2012) book covers the entire field of Sentiment Analysis. Sentiment Analysis can be done using Machine learning or a Lexicon-based approach. We use our lexicon based approach in our study. The rest of the paper is confined to Lexicon based approach 2.1 Lexicon based approach The lexicon based approach is based on the assump- tion that the contextual sentiment orientation is the 543 sum of the sentiment orientation of each word or phrase. Turney (2002) identifies sentiments …

Content truncated to 500 words