BUAP: Polarity Classification of Short Texts

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 154–159, Dublin, Ireland, August 23-24, 2014. BUAP: Polarity Classification of Short Texts David Pinto1, Darnes Vilari˜no1, Saul Le´on1, Miguel Jasso1,2, Cupertino Lucero2 1 Benem´erita Universidad Aut´onoma de Puebla 14 Sur y Av. San Claudio, CU, 72570 Puebla, Puebla, M´exico {dpinto,darnes,saul.leon}@cs.buap.mx 2 Universidad Tecnol´ogica de Iz´ucar de Matamoros Prolongaci´on Reforma 168, Santiago Mihuacan, 74420 Iz´ucar de Matamoros, Puebla, M´exico migueljhdz18@yahoo.com.mx, cuper lucero@hotmail.com Abstract We report the results we obtained at the sub- task B (Message Polarity Classification) of Se- mEval 2014 Task 9. The features used for representing the messages were basically tri- grams of characters, trigrams of PoS and a number of words selected by means of a graph mining tool. Our approach performed slightly below the overall average, except when a cor- pus of tweets with sarcasm was evaluated, in which we performed quite well obtaining around 6% above the overall average. 1 Introduction Analyzing polarity in texts is an important task that may have various applications in real life. There ex- ist plenty of tasks that may be benefited of computa- tional procedures that automatically allow to detect if the author intention has been to express himself as a positive, negative, neutral or objective manner. Let us consider, for instance, when a public figure (such as a politician, celebrity, or business leader) would like to investigate its reputation in public media. An- other example would be to calculate the reputation of a public or private institution. In any case, the construction of methods for determining the polar- ity of messages at Internet would help to investigate their reputation. In this paper, we present the results we obtained when we carried out experiments for the subtask B This work is licensed under a Creative Commons At- tribution 4.0 International Licence. Page numbers and pro- ceedings footer are added by the organisers. Licence details: http://creativecommons.org/licenses/by/4.0/ of Semeval 2014 Task 9, which was named “Mes- sage Polarity Classification”, and was defined as fol- low: “Given a message, decide whether the mes- sage is of positive, negative, or neutral sentiment. For messages conveying both a positive and nega- tive sentiment, whichever is the stronger sentiment should be chosen”. The remaining of this paper is structured as fol- lows. In Section 2 we present some related work found at the literature with respect to the identifica- tion of emotions in short texts such as twitter. Sec- tion 3 presents the description of the features and classification model used in our experiments. The results obtained together with a discussion of these results are given in Section 4. Finally, the conclu- sions are given in Section 5. 2 Related Work There exist a number of works in literature associ- ated to the automatic identification of emotions in Twitter, mainly due to the massification of this so- cial network around the world and the easy manner we can access to the Tweets from API’s provided by Twitter itself. Some of these works have focused …

Content truncated to 500 words