LyS: Porting a Twitter Sentiment Analysis Approach from Spanish to English

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 411–415, Dublin, Ireland, August 23-24, 2014. LyS: Porting a Twitter Sentiment Analysis Approach from Spanish to English David Vilares, Miguel Hermo, Miguel A. Alonso, Carlos G´omez-Rodr´ıguez, Yerai Doval Grupo LyS, Departamento de Computaci´on, Facultade de Inform´atica Universidade da Coru˜na, Campus de A Coru˜na 15071 A Coru˜na, Spain {david.vilares, miguel.hermo, miguel.alonso, carlos.gomez, yerai.doval}@udc.es Abstract This paper proposes an approach to solve message- and phrase-level polarity classi- fication in Twitter, derived from an exist- ing system designed for Spanish. As a first step, an ad-hoc preprocessing is per- formed. We then identify lexical, psycho- logical and semantic features in order to capture different dimensions of the human language which are helpful to detect sen- timent. These features are used to feed a supervised classifier after applying an in- formation gain filter, to discriminate irrel- evant features. The system is evaluated on the SemEval 2014 task 9: Sentiment Anal- ysis in Twitter. Our approach worked com- petitively both in message- and phrase- level tasks. The results confirm the robust- ness of the approach, which performed well on different domains involving short informal texts. 1 Introduction Millions of opinions, conversations or just trivia are published each day in Twitter by users of dif- ferent cultures, countries and ages. This provides an effective way to poll how people praise, com- plain or discuss about virtually any topic. Compre- hending and analysing all this information has be- come a new challenge for organisations and com- panies, which aim to find out a way to make quick and more effective decisions for their business. In particular, identifying the perception of the public with respect to an event, a service or an entity are some of their main goals in a short term. In this respect, sentiment analysis, and more specifically polarity classification, is playing an important role This work is licensed under a Creative Commons Attribu- tion 4.0 International Licence. Page numbers and proceed- ings footer are added by the organisers. Licence details: http://creativecommons.org/licenses/by/4.0/ in order to automatically analyse subjective infor- mation in texts. This paper describes our participation at Sem- Eval 2014 task 9: Sentiment Analysis in Twit- ter. Specifically, two subtasks were presented: (A) contextual polarity disambiguation and (B) message polarity classification. The first sub- task consists on determining the polarity of words or phrases extracted from short informal texts, the scope of extracts being provided by the Se- mEval organisation. Subtask B focusses on clas- sifying the content of the whole message. In both cases, three possible sentiments are consid- ered: positive, negative and neutral (which in- volves mixed and non-opinionated instances). Al- though the training set only contains tweets, the test set also includes short informal texts from other domains, in order to measure cross-domain portability. You can test the model for subtask B at miopia.grupolys.org. 2 SemEval 2014-Task 9: Sentiment Analysis in Twitter Our contribution is a reduced version of a Span- ish sentiment classification system (Vilares et al., 2013a; …

Content truncated to 500 words