Characterization of Political Polarized Users Attacked by Language Toxicity on Twitter

TL;DR

This paper investigates how language toxicity propagates among Left, Right, and Center Twitter users during politically charged discussions tied to the COVID-19 era. It analyzes a massive English tweet dataset (over 542 million posts) and 25 million replies, labeling users by political leaning through domain-based annotations from Allsides and quantifying toxicity with the Perspective API, reporting both maximum and median toxicity per user. The key finding is that Left users receive more toxic replies than Right and Center, driven by outliers in the Left category, while median toxicity is similar across groups. The work highlights implications for misinformation spread and echo-chamber dynamics, informing moderation strategies and cross-platform considerations for healthier online political discourse.

Abstract

Understanding the dynamics of language toxicity on social media is important for us to investigate the propagation of misinformation and the development of echo chambers for political scenarios such as U.S. presidential elections. Recent research has used large-scale data to investigate the dynamics across social media platforms. However, research on the toxicity dynamics is not enough. This study aims to provide a first exploration of the potential language toxicity flow among Left, Right and Center users. Specifically, we aim to examine whether Left users were easier to be attacked by language toxicity. In this study, more than 500M Twitter posts were examined. It was discovered that Left users received much more toxic replies than Right and Center users.

Figures (4)

• Figure 1: The maximum toxicities of replies of each politically-leaning category of replied-to users received. The X-axis,repied times, is in log scale. The Y axis indicates the times each category of users were replied. Reds indicate users were engaging with Right domains, exclusively; Greens indicate users were engaging with Center domains, exclusively; Blues indicate uses were engaging with Left domains, exclusively. Top toxic to-replied users were indicated by arrows. Left category outliers were indicated by arrows. Replies to the Left category users were significantly more toxic than the ones to the Right and Center category ($p < 0.005$ by Mann–Whitney U-test with a Bonferroni correction.).
• Figure 2: Boxplots represent the maximum toxicity of the replies for each user category. The median for each category is shown in each bar.
• Figure 3: The median toxicities of replies of each politically-leaning category of replied-to users received. The three categories shared a similar distribution for median toxicity. The X-axis, repied times, is in log scale. The Y axis indicates the times each category of users were replied. Reds indicate users were engaging with Right domains, exclusively; Greens indicate users were engaging with Center domains, exclusively; Blues indicate uses were engaging with Left domains, exclusively. The Left and Center outlier users (indicated by arrows) received much more toxic replies.
• Figure 4: Boxplots represent the median toxicity of the replies for each user category. The median for each category is shown in each bar. The overall reply toxicities were similar across the three categories, although the Left and Center outliers received much more toxic contents.