LLT-PolyU: Identifying Sentiment Intensity in Ironic Tweets

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 673–678, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics LLT-PolyU: Identifying Sentiment Intensity in Ironic Tweets Hongzhi Xu, Enrico Santus, Anna Laszlo and Chu-Ren Huang The Department of Chinese and Bilingual Studies The Hong Kong Polytechnic University {hongz.xu, esantus}@gmail.com mandarin1985@yahoo.de churenhuang@gmail.com Abstract In this paper, we describe the system we built for Task 11 of SemEval2015, which aims at i- dentifying the sentiment intensity of figurative language in tweets. We use various features, including those specially concerned with the identification of irony and sarcasm. The fea- tures are evaluated through a decision tree re- gression model and a support vector regres- sion model. The experiment result of the five- cross validation on the training data shows that the tree regression model outperforms the sup- port vector regression model. The former is therefore used for the final evaluation of the task. The results show that our model per- forms especially well in predicting the senti- ment intensity of tweets involving irony and sarcasm. 1 Introduction Sentiment analysis aims to identify the polarity and intensity of certain texts in order to shed light on people’s sentiments, perceptions, opinions, and be- liefs about a particular product, service, scheme, etc. Knowing what people think can, in fact, help com- panies, political parties, and other public entities in strategizing and decision making. While impressive results have been achieved in analysing literal texts (Abbasi et al., 2008; Yan et al., 2014), the study of polarity shifting in sentiment analysis still requires much research. For example, Li, et.al. (2010), explores the polarity shifters in En- glish which significantly improve the performance of sentiment analysis. Besides, figurative uses of language, such as irony or sarcasm, are also able to invert the polarity of the surface text. Theoreti- cal research in irony and sarcasm often emphasize that humans have difficulties in deciphering mes- sages with underlying meaning (Hay, 2001; Kot- thoff, 2003; Kreuz and Caucci, 2007). Factors that can facilitate the understanding of these messages include prosody (e.g. stress or intonation), kinesics (e.g. facial gestures), co-text (i.e. immediate textual environment) and context (i.e. wider environment), as well as cultural background. Computers, howev- er, cannot always rely on this kind of information. Currently, there is no method that can guaran- tee the unequivocal recognition of irony or sarcas- m. Training a computer to perform such a high- ly pragmatic task does indeed pose a challenge to computational linguists. A good number of studies have been recently devoted to finding a solution to the problem. Most of them have focused on tweet- s (Gonz´alez-Ib´a˜nez et al., 2011; Reyes et al., 2013; Liebrecht et al., 2013; Riloff et al., 2013; Barbieri et al., 2014; Vanzo et al., 2014). Identifying figurative language in short messages (generally consisting of no more than 140 character- s) that do not make use of conventional language, but employ “little space-consuming” elements, such as emoticons (“:D”), abbreviations (“abbr.”) and s- lang (“slng”) is not …

Content truncated to 500 words