A GNN-based Fraud Detection with Label Embedding and Propagation

Graph Neural Networks (GNNs) have been widely applied in fraud detection in recent years. However, few studies have considered the role of labels in information propagation. In fact, label attributes contain rich semantic information that can be leveraged. For instance, nodes with dissimilar attributes can reinforce the confidence that their labels are also dissimilar. Based on this assumption, we propose a novel fraud detection method called GNN-based Fraud Detection with Label Embedding and Propagation (GFDL). By embedding label information, we integrate node features with label attribute features. We then employ a Graph Isomorphism Network (GIN) for information propagation and utilize gated residual connections during the process, allowing the neural net-work to achieve greater depth. Extensive experiments conducted on two public benchmark fraud detection tasks demonstrate that the proposed method significantly outperforms state-of-the-art baselines.