Graph Neural Network-Driven Social Bot Detection

Abstract

The rapid expansion of online social networks has fundamentally transformed how information is produced, disseminated, and consumed. While these platforms enable open communication and global connectivity, they have also become fertile ground for malicious automation in the form of social bots. Social bots are automated or semi-automated accounts designed to imitate human behavior, often deployed to manipulate public opinion, spread misinformation, amplify extremist narratives, or distort online discourse through coordinated activities. Traditional detection techniques have struggled to keep pace with the increasing sophistication of these bots, particularly as modern bot systems leverage advanced natural language generation and complex interaction strategies that closely resemble genuine human users. In this context, this research introduces LGB, a novel social bot detection framework that integrates Language Models (LMs) with Graph Neural Networks (GNNs) to jointly analyze linguistic content and relational network structures. The proposed approach addresses the inherent limitations of content-only or structure-only detection systems by modeling both semantic intent and coordinated behavior patterns. By capturing deep contextual language representations alongside graph-based dependencies among users, posts, and interactions, LGB aims to deliver robust, scalable, and adaptive bot detection performance in dynamic social media environments