Political Tendency Identification in Twitter using Sentiment Analysis Techniques

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 183–192, Dublin, Ireland, August 23-29 2014. Political Tendency Identification in Twitter using Sentiment Analysis Techniques Ferran Pla and Llu´ıs-F. Hurtado Departament de Sistemes Inform`atics i Computaci´o Universitat Polit`ecnica de Val`encia Cam´ı Vera s/n 46022 Val`encia (Spain) {lhurtado|fpla}@dsic.upv.es Abstract This paper describes an approach for political tendency identification of Twitter users. We define some metrics that take into account the polarity of the political entities in the tweets of each user. To obtain this polarities we present the sentiment analysis system developed. The evaluation was performed on the general corpus developed at TASS2013 workshop for Spanish. To our knowl- edge, the results obtained for the sentiment analysis task and the political tendency identification task are the best results published until now using this data set. 1 Introduction Social media are usually used to express opinions and feelings about companies, products, services, hobbies, politics, etc. Therefore, enterprises, organizations, governments, and different groups in general have shown interest in the opinions that users have for their activities. They are also interested to known the way users use these media, the communication behaviour, and some users attributes such as gender, age, geographical location, political orientation, etc. In general, the main aim is to provide personalized services, particularized offers, or simply to know what people think about something in order to improve their activities. The scientific community has made a great effort to provide effective solutions to analyse, structure, and process the large amount of on-line reviews in social media. A wide set of techniques of Senti- ment Analysis (SA) are used in micro-blogging texts to extract the polarity (positive, negative, mixed or neutral) that users express in these texts. In this respect, Twitter has become a popular micro-blogging site in which users express their opinions on a variety of topics in real time. The texts used in Twit- ter are called tweets, which are short texts of a maximum of 140 characters and a language that does not have any restriction on the form and content. The nature of these texts poses new challenges for researchers in Natural Language Processing (NLP). In some cases, the tweets are written with ungram- matical sentences with a lot of emoticons, abbreviations, specific terminology, slang, etc. Therefore, the usual techniques of NLP must be adapted to these characteristics of the language, and new approaches must be proposed in order to successfully address this problem. NLP tools like POS taggers, parsers, or Named Entity Recognition (NER) tools usually fail when processing tweets because they generally are trained on grammatical texts and they perform poorly in micro-blogging texts. In this work we present a system for addressing the task of political tendency identification of Twitter users based on SA techniques. For each user, we collect all their tweets and we extract all the entities related to the political subject. Then, we automatically assign a polarity to these entities and we define a political tendency metric that …

Content truncated to 500 words