Political Polarization on Social Media

• In Event: Social Media and Polarization

Abstract

This paper uses a joint embedding model, based on graph convolutional networks and RoBERTa, for textual and social network analysis to estimate the partisan orientation of Twitter users during the 2020 presidential election and its aftermath. By combining information on social networks, likes, hashtags, retweets, and the content of tweets, the model is capable of estimating the partisan orientation of users who were active during the campaign. These estimates are based on more than 400 million tweets made by 20 million users. We validate our results by using several different gold standard measures, such as the voting records of Member of Congress (DW-NOMINATE scores), the party affiliation of users, and an ideological index based on a survey of twitter users in our data. We also present a novel approach for estimating the degree of partisan polarization on a daily basis throughout the campaign by clustering users and analyzing changes in intra- and inter-cluster distances. Preliminary findings indicate that polarization increased during the 2020 campaign, with important shifts around the Capitol riot.

Authors

• Kellin Pelrine

• Reihaneh Rabbany, McGill University

• André Blais, University of Montreal

• Jean-Francois Godbout, Universite de Montreal