Can Rumour Stance Alone Predict Veracity?

Proceedings of the 27th International Conference on Computational Linguistics, pages 3360–3370 Santa Fe, New Mexico, USA, August 20-26, 2018. 3360 Can Rumour Stance Alone Predict Veracity? Sebastian Dungs�, Ahmet Aker��, Norbert Fuhr� and Kalina Bontcheva� � University of Duisburg-Essen, Lotharstraße 65, 47057 Duisburg, Germany � University of Sheffield, 211 Portobello, Sheffield S1 4DP, UK firstname.lastname@uni-due.de k.bontcheva@dcs.shef.ac.uk Abstract Prior manual studies of rumours suggested that crowd stance can give insights into the actual rumour veracity. Even though numerous studies of automatic veracity classification of social media rumours have been carried out, none explored the effectiveness of leveraging crowd stance to determine veracity. We use stance as an additional feature to those commonly used in earlier studies. We also model the veracity of a rumour using variants of Hidden Markov Models (HMM) and the collective stance information. This paper demonstrates that HMMs that use stance and tweets’ times as the only features for modelling true and false rumours achieve F1 scores in the range of 80%, outperforming those approaches where stance is used jointly with content and user based features. 1 Introduction Social media are rife with rumours, which are fast-spreading, unverified pieces of information (Zubiaga et al., 2018). With fake news and misinformation now widely recognised as a major problem for journalists, media, online platforms, and citizens, automatic rumour detection and analysis has become a hot research topic too. Rumour analysis research has focused on Twitter in particular, as it has established itself as the go-to social platform for real-time news (Hu et al., 2012). Twitter’s unmoderated nature is also the perfect ground for spreading rumours (Qazvinian et al., 2011). A key focus of prior work on rumour analysis has been rumour stance classification, where the stance of each tweet on a given rumour is classified as supporting, denying, questioning or commenting on the rumour (Procter et al., 2013). Our work builds on the hypothesis that, as rumours evolve over time, so does the stance expressed by the public towards those rumours. In the early stages of a rumour, its actual veracity tends to be unknown. However, as new evidence emerges over time, Twitter users take more pronounced and continuously evolving stance towards the information asserted in the rumour. For instance, in the early stages of a rumour supporting tweets might prevail, simply due to the lack of information to the contrary. However, when authoritative sources or reliable evidence emerge either for or against the rumour, a similar trend tends to be observed in the collective rumour stances. This was first noted in a manual analysis by Mendoza et al. (2010), who found that true rumours tended to be affirmed more than 90% of the time, whereas false rumours were challenged (questioned or denied) 50% of the time. This encourages the use of crowd (or collective) stance as a feature in an automatic rumour veracity classifier. A rumour consists of a source tweet and several other tweets that responded to the source one—the source tweet contains the rumour. Each of those …

Content truncated to 500 words