This study investigates the multi-actor network in disinformation campaigns to uncover the structural mechanisms driving their dissemination. By analyzing the roles of diverse actors and their interaction networks, this study aims to identify key propagation patterns and their implications for computational propaganda.
This study combines methods such as reasoning and fine-tuning of large language models (LLMs) and social network analysis to identify various types of users, discover astroturfing users and ultimately establish an exponential random graph model. It integrates node-level attributes with topological metrics to model network diffusion and formation.
The results reveal that while homogeneity shapes community structures, astroturfing actors appear less constrained by these tendencies and instead exhibit heterogeneity-driven connections, which complicates conventional detection methods. Influence follows asymmetric “core-periphery” patterns, with high-contribution users forming connections driven by difference, which invites further exploration of reciprocity norms in propaganda networks.
This study offers a multidimensional framework analyzing how actor attributes and network structure facilitate disinformation spread. Its key value lies in refining governance strategies for platform and developing scalable LLM-based methods for researchers.
