Self-moderation in the Decentralized Era: Decoding Blocking Behavior on Bluesky

Abstract

Centralized social media platforms combine top-down moderation with user-level mechanisms like blocking, but access to behavioral and moderation-related data is typically restricted. In contrast, emerging decentralized networks like Bluesky provide researchers with open access to blocking actions, offering a unique opportunity to study self-moderation at scale. Understanding blocking behavior is critical in order to assess how community safety and user autonomy can be balanced in these environments. This study examines blocking on Bluesky, analyzing more than 100M actions by nearly 2M users over three months. We construct behavioral profiles from 86 features capturing user activity, content, and network interactions, and address two questions: (1) Can a user’s propensity to be blocked by others be estimated from in-platform behavior? and (2) Which behavioral features are most informative in predicting this propensity? Using state-of-the-art machine learning models, we achieve high accuracy in binary classification (AUROC > 0.85) and reliable performance in regression (R^2 ≈ 0.5). Our findings show that a small subset of features is sufficient for robust prediction, even when accounting for higher activity on the platform. Through explainability analyses, we identify the behavioral signals most strongly associated with blocking outcomes, enabling a transparent understanding of why certain users are more likely to be blocked. Our study provides a framework and empirical evidence that advances understanding of both the mechanisms and feature signals underlying user self-moderation, while also offering insights relevant to moderation design.