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

  • 2019-04-02 18:00:09
  • Dorottya Demszky, Nikhil Garg, Rob Voigt, James Zou, Matthew Gentzkow, Jesse Shapiro, Dan Jurafsky
  • 23


We provide an NLP framework to uncover four linguistic dimensions ofpolitical polarization in social media: topic choice, framing, affect andillocutionary force. We quantify these aspects with existing lexical methods,and propose clustering of tweet embeddings as a means to identify salienttopics for analysis across events; human evaluations show that our approachgenerates more cohesive topics than traditional LDA-based models. We apply ourmethods to study 4.4M tweets on 21 mass shootings. We provide evidence that thediscussion of these events is highly polarized politically and that thispolarization is primarily driven by partisan differences in framing rather thantopic choice. We identify framing devices, such as grounding and thecontrasting use of the terms "terrorist" and "crazy", that contribute topolarization. Results pertaining to topic choice, affect and illocutionaryforce suggest that Republicans focus more on the shooter and event-specificfacts (news) while Democrats focus more on the victims and call for policychanges. Our work contributes to a deeper understanding of the way groupdivisions manifest in language and to computational methods for studying them.


Introduction (beta)



Conclusion (beta)