ECNUCS: A Surface Information Based System Description of Sentiment Analysis in Twitter in the SemEval-2013 (Task 2)

Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 408–413, Atlanta, Georgia, June 14-15, 2013. c⃝2013 Association for Computational Linguistics ECNUCS: A Surface Information Based System Description of Sentiment Analysis in Twitter in the SemEval-2013 (Task 2) Tian Tian ZHU and Fang Xi ZHANG and Man LAN∗ Department of Computer Science and Technology East China Normal University 51111201046,51111201041@ecnu.edu.cn; mlan@cs.ecnu.edu.cn Abstract This paper briefly reports our submissions to the two subtasks of Semantic Analysis in Twitter task in SemEval 2013 (Task 2), i.e., the Contextual Polarity Disambiguation task (an expression-level task) and the Message Polarity Classification task (a message-level task). We extract features from surface infor- mation of tweets, i.e., content features, Micro- blogging features, emoticons, punctuation and sentiment lexicon, and adopt SVM to build classifier. For subtask A, our system on twit- ter data ranks 2 on unconstrained rank and on SMS data ranks 1 on unconstrained rank. 1 Introduction Micro-blogging today has become a very popular communication tool among Internet users. Millions of messages are appearing daily in popular web sites that provide services for Micro-blogging and one popularly known is Twitter1. Through the twitter platform, users share either information or opin- ions about personalities, politicians, products, com- panies, events (Prentice and Huffman, 2008) etc. As a result of the rapidly increasing number of tweets, mining sentiments expressed in tweets has attracted more and more attention, which is also one of the basic analysis utility functions needed by various ap- plications. The task of Sentiment Analysis in Twitter is to identify the sentiment of tweets and get a bet- ter understanding of how sentiment is conveyed in 1http://www.twitter.com tweets and texts, which consists of two sub-tasks, i.e., the Contextual Polarity Disambiguation task (an expression-level task) and the Message Polarity Classification task (a message-level task). The con- textual polarity disambiguation task (subtask A) is to determine whether a given message containing a marked instance of a word or a phrase is positive, negative or neutral in that context. The message polarity classification task (subtask B) is to decide whether a given message is of positive, negative, or neutral sentiment and for messages conveying both a positive and negative sentiment, whichever is the stronger sentiment should be chosen (Wilson et al., 2013). We participate in these two tasks. In recent years, many researchers have proposed methods to analyze sentiment in twitter. For exam- ple, (Pak and Paroubek, 2010) used a Part of Speech (POS) tagger on the tweets and found that some POS taggers can help identify the sentiment of tweets. They found that objective tweets often contain more nouns than subjective tweets. However, subjective tweets may carry more adjectives and adverbs than objective tweets. Besides, (Davidov et al., 2010) proved that emoticon and punctuation like excla- mation mark are good features when distinguishing the sentiment of tweets. In addition, some senti- ment lexicons like SentiWordNet (Baccianella et al., 2010) and MPQA Subjectivity Lexicon (Wilson et al., 2009) have been …

Content truncated to 500 words