Identifying Bots on Social Media through Coordinated Group Perception

Abstract

Identifying bots on social media has become a crucial and challenging task for regulating online discourse. Existing detection methods primarily focus on individual account-level information, identifying potential threats by detecting inconsistencies between genuine humans and anomalous bots in personal profiles, textual content, and social relationships. However, these approaches generally overlook the coordinated behavior characteristics inherent in groups of bot accounts. To address this research gap, we propose a novel Bot detection network based on Coordinated Group Perception (BotCGP), which enhances bot identification performance by uncovering the collective coordinated features among bot groups. Specifically, our method jointly models account profiles, textual content, and social relationships using a Student’s t-distribution kernel function and a differentiable modularity function to capture potential coordinated characteristics. Experimental results demonstrate that BotCGP significantly outperforms existing methods in bot detection across three real-world X/Twitter datasets. Our code is available at https://github.com/QQQQQQBY/BotCGP.