“All I know about politics is what I read in Twitter”: Weakly Supervised Models for Extracting Politicians’ Stances From Twitter

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2966–2977, Osaka, Japan, December 11-17 2016. “All I know about politics is what I read in Twitter”: Weakly Supervised Models for Extracting Politicians’ Stances From Twitter Kristen Johnson and Dan Goldwasser Department of Computer Science Purdue University, West Lafayette, IN 47907 {john1187, dgoldwas}@purdue.edu Abstract During the 2016 United States presidential election, politicians have increasingly used Twitter to express their beliefs, stances on current political issues, and reactions concerning national and international events. Given the limited length of tweets and the scrutiny politicians face for what they choose or neglect to say, they must craft and time their tweets carefully. The content and delivery of these tweets is therefore highly indicative of a politician’s stances. We present a weakly supervised method for extracting how issues are framed and temporal activity patterns on Twitter for popular politicians and issues of the 2016 election. These behavioral components are combined into a global model which collectively infers the most likely stance and agreement patterns among politicians, with respective accuracies of 86.44% and 84.6% on average. 1 Introduction The trending decline in popularity of traditional media outlets and continued rise of social media usage emerged in the 2008 U.S. presidential election campaign and has continued to the present 2016 campaign. Social media platforms, such as the microblogging outlet Twitter, allow politicians to directly access the public, express their beliefs, and react to current events. Unlike its traditional media predecessors, Twitter requires politicians to compress their ideas, political stances, and reactions to 140 character long tweets. Consequently, politicians must cleverly choose how to frame controversial issues, as well as how and when to react to each other (Mejova et al., 2013; Tumasjan et al., 2010). Due to this limit, we argue that the stance of a tweet is not independent of the social context in which it was generated. Thus, for accurate predictions these social behaviors must also be modeled. Converse to previous works which predict stance per individual tweet (SemEval, 2016), we instead present a novel approach better suited to model the dynamic political arena of Twitter, which uses the overall Twitter behavior per politician to predict a politician’s stance on an issue. We explore two aspects of the problem, stance prediction over a wide array of issues, as well as stance agreement and disagree- ment patterns between politicians over these issues. While the two aspects are related, we argue they capture different information, as identifying agreement patterns reveals alliances and rivalries between candidates, across and within their party. In an extreme case, even the lack of Twitter activity on certain issues can be indicative of a stance. For example, consider the three tweets on the issue of gun control shown in Figure 1. To identify the stance taken by each politician, our model combines both content and behavioral features, accumulated from all of a politician’s tweets on that issue. First, the tweet’s relevance to an issue can be identified using …

Content truncated to 500 words