Uncovering Discussion Groups on Claims of Election Fraud from Twitter

Abstract

Twitter was widely used during the 2020 U.S. election to disseminate claims of election fraud. As a result, a number of works have examined this phenomenon from a variety of perspectives. However, none of them focus on analyzing topics behind the general fraud claims and associating them with user communities. To fill this gap, we propose to uncover and characterize groups of Twitter users engaging in discussions about election fraud claims during the 2020 U.S. election using a large dataset that spans seven weeks during this period. To accomplish this, we model a sequence of co-retweet networks and employ a backbone extraction method that controls for inherent traits of social media applications, particularly, user activity levels and the popularity of tweets (which together generate many spurious edges in the network), thus allowing us to reveal topics of tweets that lead users to retweet them. After extracting the backbones, we identify user groups representative of the communities present in the network backbones and finally analyze the topics behind the retweeted tweets to understand how they contributed to the spread of fraud claims at that time. Our main results show that (i) our approach uncovers better-structured communities than the original network in terms of users spreading discussions about fraud; and (ii) these users discuss 25 topics with specific psycholinguistic and temporal characteristics.