Turing at SemEval-2017 Task 8: Sequential Approach to Rumour Stance Classification with Branch-LSTM

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 475–480, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics Turing at SemEval-2017 Task 8: Sequential Approach to Rumour Stance Classification with Branch-LSTM Elena Kochkina12, Maria Liakata12, Isabelle Augenstein3 1 University of Warwick, Coventry, United Kingdom 2 Alan Turing Institute, London, United Kingdom 3 University College London, London, United Kingdom {E.Kochkina, M.Liakata}@warwick.ac.uk, I.Augenstein@ucl.ac.uk Abstract This paper describes team Turing’s sub- mission to SemEval 2017 RumourEval: Determining rumour veracity and support for rumours (SemEval 2017 Task 8, Sub- task A). Subtask A addresses the challenge of rumour stance classification, which in- volves identifying the attitude of Twitter users towards the truthfulness of the ru- mour they are discussing. Stance classi- fication is considered to be an important step towards rumour verification, therefore performing well in this task is expected to be useful in debunking false rumours. In this work we classify a set of Twit- ter posts discussing rumours into either supporting, denying, questioning or com- menting on the underlying rumours. We propose a LSTM-based sequential model that, through modelling the conversational structure of tweets, which achieves an ac- curacy of 0.784 on the RumourEval test set outperforming all other systems in Sub- task A. 1 Introduction In stance classification one is concerned with de- termining the attitude of the author of a text to- wards a target (Mohammad et al., 2016). Targets can range from abstract ideas, to concrete entities and events. Stance classification is an active re- search area that has been studied in different do- mains (Ranade et al., 2013; Chuang and Hsieh, 2015). Here we focus on stance classification of tweets towards the truthfulness of rumours circu- lating in Twitter conversations in the context of breaking news. Each conversation is defined by a tweet that initiates the conversation and a set of nested replies to it that form a conversation thread. The goal is to classify each of the tweets in the conversation thread as either supporting, denying, querying or commenting (SDQC) on the rumour initiated by the source tweet. Being able to detect stance automatically is very useful in the context of events provoking public resonance and associ- ated rumours, as a first step towards verification of early reports (Zhao et al., 2015). For instance, it has been shown that rumours that are later proven to be false tend to spark significantly larger num- bers of denying tweets than rumours that are later confirmed to be true (Mendoza et al., 2010; Proc- ter et al., 2013; Derczynski et al., 2014; Zubiaga et al., 2016b). Here we focus on exploiting the conversational structure of social media threads for stance clas- sification and introduce a novel LSTM-based ap- proach to harness conversations. 2 Related Work Single Tweet Stance Classification Stance classification for rumours was pioneered by Qazvinian et al. (2011) as a binary classification task (support/denial). Zeng et al. (2016) perform stance classification for rumours emerging during crises. Both works use tweets related to …

Content truncated to 500 words