UWaterloo at SemEval-2017 Task 8: Detecting Stance towards Rumours with Topic Independent Features

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 461–464, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics UWaterloo at SemEval-2017 Task 8: Detecting Stance towards Rumours with Topic Independent Features Hareesh Bahuleyan and Olga Vechtomova University of Waterloo, ON, Canada {hpallika, ovechtomova}@uwaterloo.ca Abstract This paper describes our system for subtask-A: SDQC for RumourEval, task- 8 of SemEval 2017. Identifying rumours, especially for breaking news events as they unfold, is a challenging task due to the ab- sence of sufficient information about the exact rumour stories circulating on social media. Determining the stance of Twitter users towards rumourous messages could provide an indirect way of identifying po- tential rumours. The proposed approach makes use of topic independent features from two categories, namely cue features and message specific features to fit a gra- dient boosting classifier. With an accu- racy of 0.78, our system achieved the sec- ond best performance on subtask-A of Ru- mourEval. 1 Introduction In the recent years, with the increasing popular- ity of smartphones, social media has become one of the top sources of news. However, because all the content is user-generated, the truth behind such news stories may become difficult to verify. Spread of misinformation during the event of an emergency can potentially have negative impacts. Although a few studies in the literature have de- veloped rumour classification algorithms for Twit- ter (Qazvinian et al., 2011), these studies assume that the circulating stories about a topic or an event are known a priori (Eg: Is Barack Obama muslim ?). On the other hand, identifying rumour stories for breaking news events, as they unfold, is even more challenging (Zubiaga et al., 2016b). This is because during these early stages, the exact ru- mour stories propagating about the event are still unknown. In such a scenario, studying the conversation between users discussing the event on Twitter can possibly give insights about the veracity of a cir- culating rumour story (Zubiaga et al., 2016c). By making use of the so-called ’wisdom of the crowd’, the idea here is to understand how other users respond to rumourous tweets. It would be useful to identify if users may reply with an intent to support the story, deny the rumour by providing counter evidence or pose questions about the in- formation stated (Zubiaga et al., 2016a). Collat- ing the stance of other users could indirectly help in resolving the veracity of a rumour. The rest of this paper is organized as follows. Section 2 is a brief overview of the task. The features used and the modeling technique are de- scribed in section 3. The results are analyzed in section 4 and the conclusions from the study are provided in section 5. 2 Task Description The objective of subtask-A of RumourEval was to identify the stance of Twitter users towards rumour tweets. Given a rumourous tweet (source) and its conversation thread, the participants were required to classify the stance of each tweet (including the source tweet) …

Content truncated to 500 words