Leader-driven or Leaderless: How Participation Structure Sustains Engagement and Shapes Narratives in Online Hate Communities

TL;DR

This study investigates how internal participation structures in online hate communities shape engagement and the narratives they propagate. By analyzing a decade of Facebook data on anti-Semitic and Islamophobic groups, the authors quantify participation centralization with the Gini coefficient, extract narrative frames via a grounded-theory taxonomy, and map inter-group connections. They find that centralized leadership correlates with higher engagement, with Islamophobic groups showing more homogeneous framing under centralization and anti-Semitic groups exhibiting broader framing diversity; Islamophobic groups also display stronger within-ideology homophily. The work suggests tailored counter-extremism strategies—targeting centralized leadership in Islamophobic networks and adopting broader approaches for dispersed anti-Semitic communities—while highlighting methodological innovations in framing detection and topic modeling for online hate analysis.

Abstract

Extremist communities increasingly rely on social media to sustain and amplify divisive discourse. However, the relationship between their internal participation structures, audience engagement, and narrative expression remains underexplored. This study analyzes ten years of Facebook activity by hate groups related to the Israel-Palestine conflict, focusing on anti-Semitic and Islamophobic ideologies. Consistent with prior work, we find that higher participation centralization in online hate groups is associated with greater user engagement across hate ideologies, suggesting the role of key actors in sustaining group activity over time. Conversely, our narrative frame detection models - based on an eight-frame extremist taxonomy (e.g., dehumanization, violence justification) - reveal a clear contrast across hate ideologies, offering new insight into how discursive strategies vary despite similar structural dynamics. Analysis of the inter-group network indicates that, although centralization and homophily are not clearly linked, ideological distinctions emerge: Islamophobic groups cluster tightly, whereas anti-Semitic groups remain more evenly connected. Overall, these findings clarify how participation structure may shape the dissemination pattern and resonance of extremist narratives online and provide a foundation for tailored strategies to disrupt or mitigate online extremist discourse.

Figures (10)

• Figure 1: A decade of data on online hate groups related to the Israel/Palestine conflict on Facebook: (A) total unique users over time across four hate group ideologies, (B) the overall network of online hate groups, with nodes representing groups, node size indicating post count, edges representing shared users, and edge weights corresponding to the number of shared users.
• Figure 2: Distributions of Gini coefficients measuring inequality of user participation across hate group ideologies. The plots reveal similar participation structures across the three ideologies. The left-skewed distributions indicate that a small proportion of users produce most group content.
• Figure 3: Regression results predicting next-month engagement by group ideology. This figure includes only factors that are statistically significant ($p < 0.05$ and $95\% CI$ does not include 0) and practically significant ($|\beta| >= 0.1$) in at least one model. Predictors that are not shown for a particular model were excluded during model selection due to high multicollinearity ($VIF \le 4$) and/or poor model fit (no AIC improvement). Faded markers show coefficients that are neither statistically nor practically significant. The figure shows that participation structure centralization is a significant predictor of engagement across ideologies.
• Figure 4: A) Framing homogeneity by participation centralization across hate group ideologies (shades: darker = centralized; lighter = decentralized). Centralized Islamophobic groups exhibit greater framing homogeneity than their decentralized counterparts. In contrast, centralized Anti-Semitic groups display more diverse framings. The difference between centralized and decentralized Other-hate groups is not statistically significant. (B-D) Proportions of posts by narrative framing in Islamophobic, Anti-Semitic, and Other-hate groups by participation centralization. Markers numbered 1 to 6 indicate the key events referenced in this results section. Visualizations demonstrate significant distinctions linked to participation structure and group ideology in terms of framing diversity. Note: Although our taxonomy includes eight narrative framing types, no post in the final dataset had Imminent War/Crisis as its top framing, so it is excluded from the analysis.
• Figure 5: Homophily across group ideologies and structures (shades: darker = centralized; lighter = decentralized). Islamophobic groups are more homophilic than Anti-Semitic and Other-hate groups; Other-hate groups are heterophilic.