Mapping Election Toxicity on Social Media across Issue, Ideology, and Psychosocial Dimensions

A team of researchers from USC and KAIST has published a comprehensive study mapping toxic political discourse on social media during the 2024 U.S. presidential election. The paper, titled 'Mapping Election Toxicity on Social Media across Issue, Ideology, and Psychosocial Dimensions,' analyzed five weeks of discourse on X (Twitter) using a novel methodology. They employed a human-in-the-loop LLM-assisted process to categorize posts into 10 major campaign issues, estimate poster ideology, and detect harmful content with an LLM-based toxicity model. This allowed for a fine-grained examination of how toxicity varies by context, moving beyond blanket measurements.

The results reveal significant issue heterogeneity in online toxicity. Identity-related issues like immigration and race displayed the highest toxicity intensity, while economic issues were less toxic. In terms of specific harms, harassment was the most prevalent and intense across most issues, whereas hate speech concentrated heavily in identity-centered debates. The study also found that partisan posts contained more harmful content than neutral posts, though ideological asymmetries in toxicity varied by issue domain, challenging simple left-right narratives.

A key psychosocial finding is that toxic discourse is dominated by high-arousal negative emotions like anger and disgust. Interestingly, left- and right-leaning posts often exhibited similar emotional profiles within the same issue, suggesting a phenomenon of emotional mirroring. Furthermore, partisan groups frequently relied on overlapping moral foundations (like care and fairness), with the issue context strongly determining which foundations became most salient. This underscores that online political toxicity is highly context-dependent, necessitating issue-sensitive approaches for measurement and mitigation, rather than one-size-fits-all content moderation.