Twits, Toxic Tweets, and Tribal Tendencies: Trends in Politically Polarized Posts on Twitter

Social media platforms are often blamed for exacerbating political polarization and worsening public dialogue. Many claim that hyperpartisan users post pernicious content slanted toward their political views, inciting contentious and toxic conversations. However, what factors are actually associated with increased online toxicity and negative interactions? In this work, we explore the role that partisanship and affective polarization play in contributing to toxicity both at the individual user level and at the topic level on Twitter/X. To do this, we train and open-source a DeBERTa-based toxicity detector that outperforms the Google Jigsaw Perspective API toxicity detector on the Civil Comments test dataset. After collecting 89.6~million tweets from 43,151 US-based Twitter/X users, we then examine how several account-level characteristics-including partisanship along the US left-right political spectrum-predict how often users post toxic content. Using a Generalized Additive Model (GAM), we find that both the diversity of views and the toxicity of other accounts with which users engage have a marked effect on users' own toxicity. Specifically, toxicity is correlated with users who engage with a wider array of political views. Performing topic analysis on the toxic content posted by these accounts using the large language model MPNet and a version of the DP-Means clustering algorithm, we find similar patterns across 5,288 topics, with users becoming more toxic as they engage with a broader diversity of politically charged topics.