GraphCoarsenFormer: Integrating Transformer with Graph Coarsening for Enhanced Social Bot Detection
Xinjun Ma, Dong Qiu · 2024
Detecting social bots is essential for preserving social media integrity and combating misinformation. Traditional methods often struggle not only because they rely on homophily assumptions, which bots exploit by mimicking genuine user connections, but also due to their inability to effectively capture long-range dependencies. To address these problems, we propose a novel framework that integrates multi-granularity feature extraction with adaptive node sampling strategies. By using graph coarsening, we capture long-range dependencies while preserving the global structure, thereby enhancing the model’s ability to capture distant relationships. By combining multi-granularity and fine-grained features, we ensure a comprehensive representation of the structural and relational properties within the social network. Experiments on benchmark datasets show our method significantly outperforms existing approaches in accuracy, offering a robust solution for social network bot detection.