A novel method for anomalous behavior detection based on static and dynamic features in social networks

Anomalous activity within social networks poses significant security risks. Detecting such activity is a critical aspect of social network analysis. These activities are often implicitly correlated and temporally dynamic, complicating the representation of discrete actions with evolving relational features. The central challenge is to capture both the distinctiveness of individual behaviors and their complex, evolving interrelations. To address this challenge, this paper introduces an approach that integrates multi-structural topology and temporal knowledge for anomaly detection in social networks. The method first constructs a behavioral temporal knowledge graph to organize activity characteristics and their multiple interactions. Subsequently, a comprehensive embedding strategy is introduced to represent individual action with multi-relational and temporal information in evolving interaction scenarios. Specifically, a time-series tensor is utilized to integrate association relationships within each time slice and to incorporate temporal information into each behavioral representation. An anomalous behavior identification model, combining a variational autoencoder with a gated recurrent unit, is then employed to capture both dynamic and static interaction patterns of user activity. This approach improves the detection of dynamically diverse anomalous behaviors. Extensive experiments demonstrate the effectiveness of the proposed method in detecting anomalous activities within social networks.