A Social Bot Detection Framework based Heterogeneous Graph Attention and Similarity Graph Features

Social bot detection represents a critical and complex challenge in social network analysis and security.While graph neural networks (GNNs) have gained widespread adoption in recent years, significantly improving detection methodologies, existing approaches often struggle to comprehensively extract and integrate the rich relational information embedded in social network edges and node attributes.Specifically, current methods typically fail to effectively process and fuse different graph features.To address these limitations, We introduce an innovative framework based Heterogeneous Graph Attention and Similarity Graph Features.Our methodology encompasses a comprehensive framework: First, we construct a multi-relational graph network.We then employ Heterogeneous Graph Attention to learn node representations.Subsequently, we develop similarity graph features by Graph Convolutional Networks(GCNs), extracting a similarity graph feature that is merged with learned node features to generate a comprehensive node graph representation.We further implement semantic attention networks to combine messages from different users and relations, and ultimately utilize a Softmax for bot detection.Extensive experimental validation across three real-world datasets demonstrates the advanced performance of our approach, conclusively establishing its effectiveness in enhancing detection capabilities.