Binarizer at SemEval-2018 Task 3: Parsing dependency and deep learning for irony detection

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 628–632 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics Binarizer at SemEval-2018 Task 3: Parsing dependency and deep learning for irony detection Nishant Nikhil IIT Kharagpur Kharagpur, India nishantnikhil@iitkgp.ac.in Muktabh Mayank Srivastava ParallelDots, Inc. muktabh@paralleldots.com Abstract In this paper, we describe the system submit- ted for the SemEval 2018 Task 3 (Irony de- tection in English tweets) Subtask A by the team Binarizer. Irony detection is a key task for many natural language processing works. Our method treats ironical tweets to consist of smaller parts containing different emotions. We break down tweets into separate phrases using a dependency parser. We then embed those phrases using an LSTM-based neural network model which is pre-trained to predict emoticons for tweets. Finally, we train a fully- connected network to achieve classification. 1 Introduction The micro-blogging site Twitter has created an abundance of data about opinions and sentiments regarding almost every aspect of daily life. deeper study of the public opinion can be ob-A tained by applying natural language processing techniques on this data. However, the performance of these NLP models is detrimentally affected by irony (Pozzi et al., 2016). As per the Oxford En- glish Dictionary, irony is the expression of one’s meaning by using language that normally signi- fies the opposite, typically for humorous or em- phatic effect. This deviation between what is said and what is intended makes irony hard to detect. Being a platform where users are free to commu- nicate and express themselves colloquially, Twit- ter generates considerable data injected with irony. Studying this would provide us with a better sen- timent analysis of these tweets. Prior work on irony detection includes the use of unigrams and emoticons (Gonz´alez-Ib´anez et al., 2011; Carvalho et al., 2009; Barbieri et al., 2014). Maynard and Greenwood (2014) describe an unsupervised pattern mining approach where the sentiment of the hashtag in the tweet is pro- posed to be a key indicator of sarcasm. If the sen- timent of the tweet does not match the sentiment of the hashtag, it is predicted to be sarcastic. Riloff et al. (2013) illustrates a semi-supervised approach where rule-based classifiers are used to look for negative situation phrases and positive verbs in a sentence. Tsur et al. (2010) build pattern-based features that detect the presence of discriminative patterns as extracted from a large sarcasm-labelled corpus. N-gram-based approaches have also been used (Davidov et al., 2010; Pt´aˇcek et al., 2014; Joshi et al., 2015) with sentiment features. Joshi et al. (2017) use similarity between word embed- dings as feature and Poria et al. (2016) use con- volutional neural networks to extract sentiment, emotion and personality features for sarcasm de- tection. SemEval-2018 Task 3 (the 12th workshop on semantic evaluation) specifies two subtasks in relation to irony detection in English tweets (Van Hee et al., 2018). In subtask A the goal was to train a binary classifier that detects whether a given tweet is ironic or …

Content truncated to 500 words