nlp.cs.aueb.gr: Two Stage Sentiment Analysis

Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 562–567, Atlanta, Georgia, June 14-15, 2013. c⃝2013 Association for Computational Linguistics nlp.cs.aueb.gr: Two Stage Sentiment Analysis Prodromos Malakasiotis, Rafael Michael Karampatsis Konstantina Makrynioti and John Pavlopoulos Department of Informatics Athens University of Economics and Business Patission 76, GR-104 34 Athens, Greece Abstract This paper describes the systems with which we participated in the task Sentiment Analysis in Twitter of SEMEVAL 2013 and specifically the Message Polarity Classification. We used a 2-stage pipeline approach employing a lin- ear SVM classifier at each stage and several features including BOW features, POS based features and lexicon based features. We have also experimented with Naive Bayes classi- fiers trained with BOW features. 1 Introduction During the last years, Twitter has become a very popular microblogging service. Millions of users publish messages every day, often expressing their feelings or opinion about a variety of events, top- ics, products, etc. Analysing this kind of content has drawn the attention of many companies and re- searchers, as it can lead to useful information for fields, such as personalized marketing or social pro- filing. The informal language, the spelling mis- takes, the slang and special abbreviations that are frequently used in tweets differentiate them from traditional texts, such as articles or reviews, and present new challenges for the task of sentiment analysis. The Message Polarity Classification is defined as the task of deciding whether a message M conveys a positive, negative or neutral sentiment. For instance M1 below expresses a positive sentiment, M2 a neg- ative one, while M3 has no sentiment at all. M1: GREAT GAME GIRLS!! On to districts Monday at Fox!! Thanks to the fans for coming out :) M2: Firework just came on my tv and I just broke down and sat and cried, I need help okay M3: Going to a bulls game with Aaliyah & hope next Thursday As sentiment analysis in Twitter is a very recent subject, it is certain that more research and improve- ments are needed. This paper presents our approach for the subtask of Message Polarity Classification (Wilson et al., 2013) of SEMEVAL 2013. We used a 2-stage pipeline approach employing a linear SVM classifier at each stage and several features includ- ing bag of words (BOW) features, part-of-speech (POS) based features and lexicon based features. We have also experimented with Naive Bayes clas- sifiers trained with BOW features. The rest of the paper is organised as follows. Sec- tion 2 provides a short analysis of the data used while section 3 describes our approach. Section 4 describes the experiments we performed and the cor- responding results and section 5 concludes and gives hints for future work. 2 Data Before we proceed with our system description we briefly describe the data released by the organisers. The training set consists of a set of IDs correspond- ing to tweet messages, along with their annotations. A message can be annotated …

Content truncated to 500 words