A Few Topical Tweets are Enough for Effective User Stance Detection

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 2637–2646 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 2637 A Few Topical Tweets are Enough for Effective User Stance Detection Younes Samih Dept. of Computational Linguistics, University of D¨usseldorf D¨usseldorf, Germany Samih@phil.hhu.de Kareem Darwish Qatar Computing Research Institute, HBKU Doha, Qatar kdarwish@hbku.edu.qa Abstract User stance detection entails ascertaining the position of a user towards a target, such as an entity, topic, or claim. Recent work that em- ploys unsupervised classification has shown that performing stance detection on vocal Twit- ter users, who have many tweets on a target, can be highly accurate (+98%). However, such methods perform poorly or fail completely for less vocal users, who may have authored only a few tweets about a target. In this paper, we tackle stance detection for such users using two approaches. In the first approach, we im- prove user-level stance detection by represent- ing tweets using contextualized embeddings, which capture latent meanings of words in con- text. We show that this approach outperforms two strong baselines and achieves 89.6% accu- racy and 91.3% macro F-measure on eight con- troversial topics. In the second approach, we expand the tweets of a given user using their Twitter timeline tweets, which may not be top- ically relevant, and then we perform unsuper- vised classification of the user, which entails clustering a user with other users in the train- ing set. This approach achieves 95.6% accu- racy and 93.1% macro F-measure. 1 Introduction Stance detection entails identifying the position of a user towards a topic, an entity, or a claim (Mo- hammad et al., 2016b). Effective stance detection, particularly in the realm of social media, can be instrumental in gauging public opinion, identifying intersecting and diverging groups, and understand- ing issues of interest to different user communities (Magdy et al., 2016a). Much recent works have explored varying stance detection methods includ- ing supervised, semi-supervised, and unsupervised user classification (Darwish et al., 2020; Magdy et al., 2016a; Pennacchiotti and Popescu, 2011; Wong et al., 2013), and much of the work has fo- (a) (b) Figure 1: Sample 2018 US midterm election related tweets that either express a very clear stance (a) or not (b) cused on stance detection for Twitter users. The different approaches have advantages and disad- vantages. For example, supervised methods are simple to implement, but they require manually annotated training data and their accuracy varies widely based on classification features, classifica- tion techniques, and the number of training and test examples (Magdy et al., 2016a). Though semi- supervised and unsupervised methods typically use user interactions and often may yield perfect classi- fication, they are effective in classifying highly vo- cal users with many topical tweets (Darwish et al., 2020). Most of these methods produce sub-optimal results for users who rarely express their opinion, and for whom we may only have one or two topi- cally related tweets. Though a single tweet might be explicitly clear, …

Content truncated to 500 words