Retrieving and Analyzing Toxic and Polarized Discourse on BlueSky: A Study of Decentralized Social Network

Abstract

The proliferation of hate speech on decentralized social media platforms, such as BlueSky, presents a growing challenge for ensuring healthy online discourse. Unlike traditional centralized networks, decentralized platforms lack a single governing authority, making moderation and content regulation more complex. This thesis focuses on the retrieval and analysis of data from BlueSky to better understand the nature and spread of hate speech and polarization within such decentralized ecosystems. By collecting and processing real-world data from the platform(13699 posts, 58788 users), we explore patterns of information dissemination and evaluate the potential of applying machine learning approaches—such as Federated Learning (FL) and Graph Neural Networks (GNN), for future disinformation detection in decentralized contexts.