Dual-channel Heterophilic Message Passing for Graph Fraud Detection

TL;DR

A novel framework, Dual-channel Heterophilic Message Passing (DHMP), that leverages a heterophily separation module to divide the graph into homophilic and heterophilic subgraphs, mitigating the low-pass inductive bias of traditional GNNs.

Abstract

Fraudulent activities have significantly increased across various domains, such as e-commerce, online review platforms, and social networks, making fraud detection a critical task. Spatial Graph Neural Networks (GNNs) have been successfully applied to fraud detection tasks due to their strong inductive learning capabilities. However, existing spatial GNN-based methods often enhance the graph structure by excluding heterophilic neighbors during message passing to align with the homophilic bias of GNNs. Unfortunately, this approach can disrupt the original graph topology and increase uncertainty in predictions. To address these limitations, this paper proposes a novel framework, Dual-channel Heterophilic Message Passing (DHMP), for fraud detection. DHMP leverages a heterophily separation module to divide the graph into homophilic and heterophilic subgraphs, mitigating the low-pass inductive bias of traditional GNNs. It then applies shared weights to capture signals at different frequencies independently and incorporates a customized sampling strategy for training. This allows nodes to adaptively balance the contributions of various signals based on their labels. Extensive experiments on three real-world datasets demonstrate that DHMP outperforms existing methods, highlighting the importance of separating signals with different frequencies for improved fraud detection. The code is available at https://github.com/shaieesss/DHMP.

Figures (5)

• Figure 1: Illustration of two key challenges in fraud detection: (1) The pronounced disparity between the number of fraudulent and benign users, with fraudsters often being significantly outnumbered, increases the difficulty of identifying and mitigating fraudulent activities. (2) Homophilic connections promote the blending of information between entities within the same categories, while heterophilic connections blur the boundaries between different classes, further complicating the detection of deceitful interactions.
• Figure 2: The framework of DHMP. First, DHMP leverages a heterophily separation module to split the low-pass and high-pass signals in the original graph. Then, DHMP leverages the re-scaled residual message aggregation module to propagate the homophilic and heterophilic information. Finally, DHMP fuses the relational information and detects the fraudulent entities. DHMP is trained using the joint loss functions, including classification and heterophily loss.
• Figure 3: The results of sensitivity experiments of hyperparameter $\epsilon$: (a) AUC of DHMP on three datasets; (b) F1-macro of DHMP on three datasets
• Figure 4: The results of sensitivity experiments of hyperparameter $\lambda$: (a) AUC of DHMP on three datasets; (b) F1-macro of DHMP on three datasets
• Figure 5: The results of visualization of the representation learning on the YelpChi dataset.