NTUA-SLP at SemEval-2018 Task 3: Tracking Ironic Tweets using Ensembles of Word and Character Level Attentive RNNs

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 613–621 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics NTUA-SLP at SemEval-2018 Task 3: Tracking Ironic Tweets using Ensembles of Word and Character Level Attentive RNNs Christos Baziotis1,3, Nikos Athanasiou1, Pinelopi Papalampidi1, Athanasia Kolovou1,2, Georgios Paraskevopoulos1,4, Nikolaos Ellinas1 Alexandros Potamianos1,4 1School of ECE, National Technical University of Athens, Athens, Greece 2 Department of Informatics, University of Athens, Athens, Greece 3 Department of Informatics, Athens University of Economics and Business, Athens, Greece 4 Behavioral Signal Technologies, Los Angeles, CA cbaziotis@mail.ntua.gr, el12074@central.ntua.gr el12003@central.ntua.gr, akolovou@di.uoa.gr geopar@central.ntua.gr, nellinas@central.ntua.gr potam@central.ntua.gr Abstract In this paper we present two deep-learning systems that competed at SemEval-2018 Task 3 “Irony detection in English tweets”. We design and ensemble two independent mod- els, based on recurrent neural networks (Bi- LSTM), which operate at the word and charac- ter level, in order to capture both the semantic and syntactic information in tweets. Our mod- els are augmented with a self-attention mech- anism, in order to identify the most informa- tive words. The embedding layer of our word- level model is initialized with word2vec word embeddings, pretrained on a collection of 550 million English tweets. We did not utilize any handcrafted features, lexicons or external datasets as prior information and our models are trained end-to-end using back propagation on constrained data. Furthermore, we provide visualizations of tweets with annotations for the salient tokens of the attention layer that can help to interpret the inner workings of the proposed models. We ranked 2nd out of 42 teams in Subtask A and 2nd out of 31 teams in Subtask B. However, post-task-completion enhancements of our models achieve state-of- the-art results ranking 1st for both subtasks. 1 Introduction Irony is a form of figurative language, considered as “saying the opposite of what you mean”, where the opposition of literal and intended meanings is very clear (Barbieri and Saggion, 2014; Liebrecht et al., 2013). Traditional approaches in NLP (Tsur et al., 2010; Barbieri and Saggion, 2014; Karoui et al., 2015; Farías et al., 2016) model irony based on pattern-based features, such as the contrast be- tween high and low frequent words, the punctua- tion used by the author, the level of ambiguity of yay its fucking monday life is so perfect and magical i love everything Label: ironic by clash b e a u t i f u l w a y t o s t a r t m y m o r n i n g . Label: ironic by clash Figure 1: Attention heat-map visualization. The color intensity of each word / character, corre- sponds to its weight (importance), as given by the self-attention mechanism (Section 2.6). the words and the contrast between the sentiments. Also, (Joshi et al., 2016) recently added word em- beddings statistics to the feature space and further boosted the performance in irony detection. Modeling irony, especially in Twitter, is a chal- lenging task, since in ironic comments literal meaning can be misguiding; irony is …

Content truncated to 500 words