A Hybrid Approach to Emotional Sentence Polarity and Intensity Classification

Proceedings of the Fourteenth Conference on Computational Natural Language Learning, pages 153–161, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics A Hybrid Approach to Emotional Sentence Polarity and Intensity Classification Jorge Carrillo de Albornoz, Laura Plaza, Pablo Gervás Universidad Complutense de Madrid Madrid, Spain {jcalbornoz,lplazam}@fdi.ucm.es, pgervas@sip.ucm.es Abstract In this paper, the authors present a new ap- proach to sentence level sentiment analysis. The aim is to determine whether a sentence expresses a positive, negative or neutral sen- timent, as well as its intensity. The method performs WSD over the words in the sentence in order to work with concepts rather than terms, and makes use of the knowledge in an affective lexicon to label these concepts with emotional categories. It also deals with the ef- fect of negations and quantifiers on polarity and intensity analysis. An extensive evaluation in two different domains is performed in order to determine how the method behaves in 2- classes (positive and negative), 3-classes (posi- tive, negative and neutral) and 5-classes (strongly negative, weakly negative, neutral, weakly positive and strongly positive) classifi- cation tasks. The results obtained compare fa- vorably with those achieved by other systems addressing similar evaluations. 1 Introduction Sentiment analysis has gained much attention from the research community in recent years. It is concerned with the problem of discovering emotional meanings in text, and most common tasks usually include emotion labeling (assigning a text its main emotion), polarity recognition (classifying a statement into positive or negative) and subjectivity identification (determining whether a text is subjective or objective). The growing research interest is mainly due to the practical applications of sentiment analysis. Companies and organizations are interested in finding out costumer sentiments and opinions, while individuals are interested in others’ opi- nions when purchasing a product or deciding whether or not watching a movie. Many approaches have dealt with sentiment analysis as the problem of classifying product or service reviews (Pang et al., 2002; Turney, 2002), while others have attempted to classify news items (Devitt and Ahmad, 2007). The task is usually addressed as a 2-classes classification problem (positive vs. negative). Recent works have included the neutral class, trying to detect not only the polarity but also the absence of emo- tional meaning (Wilson et al., 2005; Esuli and Sebastiani, 2006). However, few approaches try to face a more fine-grained prediction of the in- tensity (e.g. classifying the polarity into strongly negative, weakly negative, neutral, weakly posi- tive and strongly positive). Another important problem of most of these approximations is that they usually work with terms, and so disregard the contextual meaning of those terms in the sentence (Martineau and Finin, 2009; Moilanen and Pulman, 2007). The use of word disambiguation is not usual in this task, due to the fact that most approaches use lexical resources created to work with terms. However, it is essential to correctly capture the meaning of these terms within the text. In this paper, we present a hybrid approach based on machine learning techniques and …

Content truncated to 500 words