Social Approval and Network Homophily as Motivators of Online Toxicity

TL;DR

The paper addresses why online hate messages are produced, testing the social approval theory against a harm-motivated account. It analyzes millions of historical tweets from known hateful users, builds retweet and mention networks, and measures toxicity with the Perspective API, applying bot-filtering thresholds at $0.5$ and $0.8$. The results show toxicity is homophilous in social networks (positive network assortativity and neighbor-to-user toxicity correlations, with significant $p<0.001$), and that social approval signals—particularly retweets—predict increases in subsequent toxicity, while insufficient approval predicts reductions. These findings support the social approval mechanism of online hate and imply moderation or design interventions to disrupt reinforcement chains, though the study remains observational and acknowledges limitations like possible data bias and lack of randomized control.

Abstract

Online hate messaging is a pervasive issue plaguing the well-being of social media users. This research empirically investigates a novel theory positing that online hate may be driven primarily by the pursuit of social approval rather than a direct desire to harm the targets. Results show that toxicity is homophilous in users' social networks and that a user's propensity for hostility can be predicted by their social networks. We also illustrate how receiving greater or fewer social engagements in the form of likes, retweets, quotes, and replies affects a user's subsequent toxicity. We establish a clear connection between receiving social approval signals and increases in subsequent toxicity. Being retweeted plays a particularly prominent role in escalating toxicity. Results also show that not receiving expected levels of social approval leads to decreased toxicity. We discuss the important implications of our research and opportunities to combat online hate.

Figures (9)

• Figure 1: The four types of social engagement on dimensions of rebroadcast and endorsement. Retweets represent rebroadcast and endorsement, likes represent endorsements, quotes are rebroadcasts that can be either positive or negative, and replies do not rebroadcast and can be either positive or negative.
• Figure 2: Changes in toxicity (y-axis) when an anchor tweet received lower (red bars) or higher (blue bars) than the predicted amount of social engagement at different windows $k$ (x-axis). Changes that are significantly different between the lower- and the higher-than-predicted groups are indicated (Mann-Whitney U test, ** $p<0.01$, *** $p<0.001$).
• Figure 3: Having higher amounts of likes and retweets than predicted would result in the biggest increase in future toxicity, and vice versa ($k=50$).
• Figure 4: When an anchor tweet receives substantially lower (red) or higher (blue) amount of retweets than expected, the difference in maximum toxicity ($k=50$) is statistically significant (Mann-Whitney U test, $**p<0.01$). More retweets lead to an increase in maximum toxicity, and vice versa
• Figure 5: Distribution of user bot scores.