LGB Language Model and Graph Neural Network-Driven Social Bot Detection

Abstract

Social bots on platforms like Twitter and Facebook pose serious threats, spreading misinformation and manipulating public opinion. Traditional bot detection methods rely heavily on feature engineering and shallow machine learning models, often failing to capture complex social behaviors. This research proposes a hybrid approach using LGB (Light Gradient Boosting) Language Models combined with Graph Neural Networks (GNNs) to enhance bot detection accuracy. LGB models analyze textual content efficiently, while GNNs capture network structures, relationships, and propagation patterns of messages. Experimental evaluation demonstrates improved detection rates, reduced false positives, and robustness against evolving bot strategies. This approach enables real-time social bot detection, contributing to safer online communities and providing a scalable framework for social media platforms.