The Virality of Hate Speech on Social Media

TL;DR

Important determinants that explain differences in the spreading of hateful vs. normal content are identified and novel insights into the virality of hate speech on social media are offered.

Abstract

Online hate speech is responsible for violent attacks such as, e.g., the Pittsburgh synagogue shooting in 2018, thereby posing a significant threat to vulnerable groups and society in general. However, little is known about what makes hate speech on social media go viral. In this paper, we collect N = 25,219 cascades with 65,946 retweets from X (formerly known as Twitter) and classify them as hateful vs. normal. Using a generalized linear regression, we then estimate differences in the spread of hateful vs. normal content based on author and content variables. We thereby identify important determinants that explain differences in the spreading of hateful vs. normal content. For example, hateful content authored by verified users is disproportionally more likely to go viral than hateful content from non-verified ones: hateful content from a verified user (as opposed to normal content) has a 3.5 times larger cascade size, a 3.2 times longer cascade lifetime, and a 1.2 times larger structural virality. Altogether, we offer novel insights into the virality of hate speech on social media.

Figures (6)

• Figure 1: Complementary cumulative distribution functions (CCDFs) for cascade size, cascade lifetime, and structural virality of hateful vs. normal content on social media.
• Figure 2: Parameter estimates (standardized) and 95% confidence intervals for (a) cascade size, (b) cascade lifetime, and (c) structural virality. The direct effects (top) explain the spread of all tweets (regardless of whether these embed hate or not), while the interactions (bottom) explain differences between the spread of hateful. vs. normal content. Significance levels: ***$\bm{p<0.001}$, **$\bm{p<0.01}$, and *$\bm{p<0.05}$.
• Figure 3: Predicted marginal effects of the key determinants on (a) cascade size, (b) cascade lifetime, and (c) structural virality. Shown is the mean estimate with 95% confidence intervals for hateful content (in red) and for normal content (in blue). Continuous variables (here tweet length) were $\bm{z}$-standardized. Thus, their values on the $x$-axis represent the mean ($\bm{\mu}$) and standard deviations ($\bm{\sigma}$).
• Figure 4: Mediation analysis captioning the average direct effect (ADE) and the average causal mediation effect (ACME) through infectiousness (mediator) of the EV (i.e., hate) on (a) cascade lifetime and (b) structural virality (DVs). Significance levels: ***$\bm{p<0.001}$, **$\bm{p<0.01}$, and *$\bm{p<0.05}$.
• Figure A.1: Exemplary cascade with computed cascade metrics at each level.