DSIC-ELIRF at SemEval-2016 Task 4: Message Polarity Classification in Twitter using a Support Vector Machine Approach

Proceedings of SemEval-2016, pages 198–201, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics DSIC-ELIRF at SemEval-2016 Task 4: Message Polarity Classification in Twitter using a Support Vector Machine Approach V´ıctor Mart´ınez Ferran Pla LLu´ıs-F. Hurtado Universitat Polit`ecnica de Val`encia Cam´ı de Vera s/n, 46022 Val`encia {vmartinez2,fpla,lhurtado}@dsic.upv.es Abstract This paper contains the description of our par- ticipation at task 4 (sub-task A, Message Po- larity Classification) of SemEval-2016. Our proposed system consists mainly of three steps. Firstly, the preprocessing step includes the tokenization and identification of special elements including URLs, hashtags, user men- tions and emoticons. The second step aims at selecting and extracting the feature set. Fi- nally, a supervised approach, in particular a Support Vector Machine has been applied to tackle the classification problem. 1 Introduction In the last few years, Twitter has become a source of a huge amount of information which introduces endless possibilities of research in the field of Sen- timent Analysis. Sentiment Analysis, also called Opinion Mining, is a research area within Natural Language Processing whose aim is to identify the underlying emotion of a certain document, sentence or aspect (Liu, 2012). As a case in point, Opinion Mining has been applied for recognizing reviews as recommended or not recommended (Turney, 2002) and for generating aspect-based summaries (Hu and Liu, 2004). The goal of SemEval-2016 task 4 (Nakov et al., 2016) consists of categorizing tweets as positive, negative or neutral concerning the opinion that a user holds with regard to a certain topic. One issue to take into consiteration is that the language adopted in So- cial Media, especially in Twitter, needs to be treated differently than normalized language due to the use of specific characteristics such as users, hashtags, emoticons and slang as well as some linguistic phe- nomena including sarcasm and irony. Our system is closely related to (Gim´enez et al., 2015). Section 2 describes the proposed method which consists mainly of three steps. Firstly, the pre- processing step includes the tokenization and iden- tification of special elements including URLs, hash- tags, user mentions and emoticons. The second step aims at selecting and extracting the feature set. Fi- nally, a supervised approach such as Support Vector Machine (SVM) has been applied to tackle the clas- sification problem. In section 3, the experiments car- ried out are described. Finally, section 4 discusses the results obtained for the different experiments in the tuning phase and in the official competition. 2 System Overview In this section, we describe the steps carried out in this work to achieve the results obtained in Semeval 2016. In this approach, a matrix of ocurrences, in which tweets are represented as rows and features as columns, normalized by tf-idf was used to represent whether a certain feature appears or not in a tweet. 2.1 Preprocessing After fetching all the data from Twitter, our cor- pus needs to be preprocessed. As Twitter makes an extensive use of emoticons, URLs and con- crete elements such as @User mentions and #hash- …

Content truncated to 500 words