GAA-BD: Graph Adversarial Augmentation-based Social Bot Detection
Nan Hu, Le Cheng, Botao Wang, Jiwei Xu, Keke Tang, Peican Zhu · 2024
Online social networks are crucial for information acquisition nowadays, yet they are increasingly jeopardized by malicious attacks from social bots. This underscores the urgent need for robust social bot detection methods. To address the significant challenge posed by the imbalance in the number of human and bot users, we introduce a Graph Adversarial Augmentation-based method for social Bot Detection (GAA-BD) to improve detection efficacy. Our method incorporates a graph convolutional neural network as its core architecture and enhances it with adversarial augmentation applied to both the dataset and the training process. By strategically generating synthetic samples and employing targeted adversarial training, our approach effectively resolves sample size imbalances and bolsters model robustness. Comprehensive experimental results demonstrate that our proposed method outperforms existing baselines, establishing its effectiveness in combatting social bot infiltration in online networks.