LDR at SemEval-2018 Task 3: A Low Dimensional Text Representation for Irony Detection

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 531–536 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics LDR at SemEval-2018 Task 3: A Low Dimensional Text Representation for Irony Detection Bilal Ghanem Universitat Politecnica de Valencia, Spain bigha@doctor.upv.es Francisco Rangel Universitat Politecnica de Valencia, Spain francisco.rangel@autoritas.es Paolo Rosso Universitat Politecnica de Valencia, Spain prosso@dsic.upv.es Abstract In this paper we describe our participation in the SemEval-2018 task 3 Shared Task on Irony Detection. We have approached the task with our low dimensionality representation method (LDR), which exploits low dimensional fea- tures extracted from text on the basis of the occurrence probability of the words depend- ing on each class. Our intuition is that words in ironic texts have different probability of occur- rence than in non-ironic ones. Our approach obtained acceptable results in both subtasks A and B. We have performed an error analysis that shows the difference on correct and incor- rect classified tweets. 1 Introduction With the existence of online social networks, a huge amount of information rapidly pervades, which attracting the attention of researchers to in- vestigate the linguistic phenomenon that appears. One of these complex phenomenon is irony, where the speaker uses words that mean the opposite of the literal meaning and what others really think, especially in order to be funny1. Moreover, irony can be considered as a strategy intended to criticise or to praise (Hern´andez-Far´ıas et al., 2015). The detection of irony recently is quite a hot research topic due to its importance for efficient sentiment analysis (Ghosh et al., 2015). Also, another figu- rative language device noticed recently is sarcasm, where the writer intend to offend someone rather than creating a humor situation. In many research works, irony and sarcasm are often viewed as the same language device, or they considered irony as an umbrella term that covers also sarcasm (Wang, 1As defined in the Merriam Webster Dictionary, http://www.merriam-webster.com/dictionary/irony; accessed on Feb. 2018. 2013). Several approaches have been proposed to detect irony, where most of them have turned the problem into a binary classification task using a set of features. (Carvalho et al., 2009) proposed one of the first works on irony detection. They worked on the identification of a set of patterns to identify ironic sentences. The adopted features were the use of punctuation marks and emoticons. (Reyes et al., 2013) proposed a model that em- ployed four types of conceptual features: signa- tures, unexpectedness, style and emotional sce- narios. (Barbieri and Saggion, 2014) proposed a model using lexical features, such as frequency of rare and common terms, synonyms, adjectives, emoticons, punctuation marks, positive and nega- tive terms. Their results showed that the most im- portant features are structure, frequency and syn- onyms for detecting irony in multiple datasets. (Karoui et al., 2015) presented a model to detect irony using a vector composed of six main groups of features: surface features (such as punctuation marks), sentiment (positive and negative words), sentiment shifter (positive and negative words in …

Content truncated to 500 words