Analyzing Polarization in Social Media: Method and Application to Tweets on 21 Mass Shootings

Proceedings of NAACL-HLT 2019, pages 2970–3005 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 2970 Analyzing Polarization in Social Media: Method and Application to Tweets on 21 Mass Shootings Dorottya Demszky1 Nikhil Garg1 Rob Voigt1 James Zou1 Matthew Gentzkow1 Jesse Shapiro2 Dan Jurafsky1 1Stanford University 2Brown University {ddemszky, nkgarg, robvoigt, jamesz, gentzkow, jurafsky}@stanford.edu jesse shapiro 1@brown.edu Abstract We provide an NLP framework to uncover four linguistic dimensions of political polar- ization in social media: topic choice, framing, affect and illocutionary force. We quantify these aspects with existing lexical methods, and propose clustering of tweet embeddings as a means to identify salient topics for analy- sis across events; human evaluations show that our approach generates more cohesive topics than traditional LDA-based models. We ap- ply our methods to study 4.4M tweets on 21 mass shootings. We provide evidence that the discussion of these events is highly polarized politically and that this polarization is primar- ily driven by partisan differences in framing rather than topic choice. We identify framing devices, such as grounding and the contrasting use of the terms “terrorist” and “crazy”, that contribute to polarization. Results pertaining to topic choice, affect and illocutionary force suggest that Republicans focus more on the shooter and event-specific facts (news) while Democrats focus more on the victims and call for policy changes. Our work contributes to a deeper understanding of the way group di- visions manifest in language and to computa- tional methods for studying them.1 1 Introduction Elites, political parties, and the media in the US are increasingly polarized (Layman et al., 2010; Prior, 2013; Gentzkow et al., forthcoming), and the propagation of partisan frames can influence public opinion (Chong and Druckman, 2007) and party identification (Fiorina and Abrams, 2008). Americans increasingly get their news from internet-based sources (Mitchell et al., 2016), and political information-sharing is highly ide- ologically segregated on platforms like Twitter (Conover et al., 2011; Halberstam and Knight, 1All data and code is available at: https://github. com/ddemszky/framing-twitter 2016) and Facebook (Bakshy et al., 2015). Prior NLP work has shown, e.g., that polarized mes- sages are more likely to be shared (Zafar et al., 2016) and that certain topics are more polarizing (Balasubramanyan et al., 2012); however, we lack a more broad understanding of the many ways that polarization can be instantiated linguistically. This work builds a more comprehensive frame- work for studying linguistic aspects of polariza- tion in social media, by looking at topic choice, framing, affect, and illocutionary force. 1.1 Mass Shootings We explore these aspects of polarization by study- ing a sample of more than 4.4M tweets about 21 mass shooting events, analyzing polarization within and across events. Framing and polarization in the context of mass shootings is well-studied, though much of the lit- erature studies the role of media (Chyi and Mc- Combs, 2004; Schildkraut and Elsass, 2016) and politicians (Johnson et al., 2017). Several works find that frames have changed over time and be- tween such events (Muschert and Carr, 2006; …

Content truncated to 500 words