Anomaly Behavior Analysis for Blockchain Social Networks Using Heterogeneous Graph Neural Networks

Social networks have exploded in popularity, leading to growing concerns about abnormal behaviors. Existing solutions for identifying anomalies often rely on limited information and synthetic data due to restricted access to comprehensive social network datasets. However, these data fail to accurately capture real-world scenarios, leading to potential misclassification when applied to actual contexts. To overcome this issue, we introduce HiveABN, a novel graph-structured dataset derived from Hive blockchain social network, for comprehensive anomaly analysis. By leveraging the decentralized system characteristics, our dataset encompasses a complete record of all user profiles and activities within the network, facilitating straightforward analysis for labeling anomalous behavior. We further conduct experiments utilizing our proposed graph anomaly behavior framework, GABA, based on various graph neural networks to maximize rich relationships within the network. The experiments demonstrate the effectiveness of our framework and dataset across various anomaly analysis tasks, providing a solid foundation for future research.