GPLSI: Supervised Sentiment Analysis in Twitter using Skipgrams

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 294–299, Dublin, Ireland, August 23-24, 2014. GPLSI: Supervised Sentiment Analysis in Twitter using Skipgrams Javi Fern´andez, Yoan Guti´errez, Jos´e M. G´omez, Patricio Mart´ınez-Barco Department of Software and Computing Systems University of Alicante {javifm,ygutierrez,jmgomez,patricio}@dlsi.ua.es Abstract In this paper we describe the system sub- mitted for the SemEval 2014 Task 9 (Sen- timent Analysis in Twitter) Subtask B. Our contribution consists of a supervised ap- proach using machine learning techniques, which uses the terms in the dataset as fea- tures. In this work we do not employ any external knowledge and resources. The novelty of our approach lies in the use of words, ngrams and skipgrams (not- adjacent ngrams) as features, and how they are weighted. 1 Introduction The Web 2.0 has become one of the most im- portant sources of data to extract useful and het- erogeneous knowledge from. Texts can provide factual information, such as descriptions and lists of features, and opinion-based information, which would include reviews, emotions, or feelings. This subjective information can be expressed through different textual genres, such as blogs, forums, so- cial networks and microblogs. An example of microblogging social network is Twitter1, which has gained much popularity in the last years. This website enables its users to send and read text-based messages of up to 140 char- acters, known as tweets. This site can be a vast source of subjective information in real time; mil- lions of users share opinions on different aspects of their everyday life. Extracting this subjective information has a great value for both general and expert users. However, it is difficult to exploit it accordingly, mainly because of the short length of 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: creativecommons.org/licenses/by/4.0/ 1http://twitter.com the tweets, the informality, and the lack of context. Sentiment Analysis (SA) systems must be adapted to face the challenges of this new textual genre. International competitions related to the assess- ment of SA systems in Twitter have taken place. Some of them include the TASS workshop in the SEPLN conference (Villena-Rom´an et al., 2013), the RepLab workshop in the CLEF conference (Amig´o et al., 2012), and the Sentiment Analysis in Twitter task (Task 2) in the last SemEval work- shop (Nakov et al., 2013). In this paper we describe the system submit- ted for the SemEval 2014 Sentiment Analysis in Twitter task (Task 9 Subtask B)2 (Rosenthal et al., 2014). This task consists of performing an au- tomatic sentiment analysis to determine whether a message expresses a positive, negative, or neu- tral sentiment. The organisers of this task provide three datasets: training, development training, and development test. The participants can use the training and development training datasets to train and validate their models, but the development test dataset can only be used for validation. The size and distribution of polarities of these datasets is shown in Table 1. …

Content truncated to 500 words