LT3: Sentiment Analysis of Figurative Tweets: piece of cake #NotReally

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 684–688, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics LT3: Sentiment Analysis of Figurative Tweets: piece of cake #NotReally Cynthia Van Hee, Els Lefever and Véronique hoste LT3, Language and Translation Technology Team Department of Translation, Interpreting and Communication – Ghent University Groot-Brittanniëlaan 45, 9000 Ghent, Belgium Firstname.Lastname@UGent.be Abstract This paper describes our contribution to the SemEval-2015 Task 11 on sentiment analysis of figurative language in Twitter. We consid- ered two approaches, classification and regres- sion, to provide fine-grained sentiment scores for a set of tweets that are rich in sarcasm, irony and metaphor. To this end, we combined a variety of standard lexical and syntactic fea- tures with specific features for capturing fig- urative content. All experiments were done using supervised learning with LIBSVM. For both runs, our system ranked fourth among fif- teen submissions. 1 Introduction Handling figurative language is currently one of the most challenging tasks in NLP. Figurative lan- guage is often characterized by linguistic devices such as sarcasm, irony, metaphors, and humour. Their meaning goes beyond the literal meaning and is therefore often hard to capture, even for humans. However, as an increasing part of our daily commu- nication takes place on social media (e.g. Twitter, Facebook), which are prone to figurative language use, there is an urgent need for automatic systems that recognize and understand figurative online con- tent. This is especially the case in the field of senti- ment analysis where the presence of figurative lan- guage in subjective text can significantly undermine the classification accuracy. Understanding figurative language often requires world knowledge, which cannot easily be accessed by machines. Moreover, figurative language rapidly evolves due to changes in vocabulary and language, which makes it difficult to train machine learning algorithms. Nevertheless, the identification of non- literal uses of language has attracted a fair amount of research interest recently. Veale (2012) investi- gated the relation between irony and our stereotyp- ical knowledge of a domain and showed how the insight in stereotypical norms helps to recognize and understand ironic utterances. Reyes et al. (2013) built an irony model for Twitter for which they re- lied on a set of textual features for capturing ironic tweets. Their model obtained promising results con- cerning recall (84%). In what relates to the detec- tion of metaphors, Turney et al. (2011) introduced an algorithm for distinguishing between metaphor- ical and literal word usages based on the degree of abstractness of a word’s context. More recent work by Tsvetkov et al. (2014) presents a cross-lingual model based on lexical semantic word features for metaphor detection in English, Spanish, Farsi and Russian. To date, most studies on figurative language use have focussed on the detection of linguistic devices such as sarcasm, irony and metaphor. By contrast, only a few studies have investigated how these de- vices affect sentiment analysis. Indeed, as stated by Maynard (2014), it is not sufficient to determine whether a text contains …

Content truncated to 500 words