DGA-GNN: Dynamic Grouping Aggregation GNN for Fraud Detection

• DOI: 10.1609/aaai.v38i10.29067
• Publication Date: 2024-03-25T10:42:22Z
• Authors: Mingjiang Duan; Tongya Zheng; Yang Gao; Gang Wang; Zunlei Feng; Xinyu Wang

Fraud detection has increasingly become a prominent research field due to the dramatically increased incidents of fraud. The complex connections involving thousands, or even millions nodes, present challenges for fraud tasks. Many researchers have developed various graph-based methods detect from these intricate graphs. However, those neglect two distinct characteristics graph: non-additivity certain attributes and distinguishability grouped messages neighbor nodes. This paper introduces Dynamic Grouping Aggregation Graph Neural Network (DGA-GNN) detection, which addresses by dynamically grouping attribute value ranges In DGA-GNN, we initially propose decision tree binning encoding transform non-additive node into bin vectors. approach aligns well with GNN’s aggregation operation avoids nonsensical feature generation. Furthermore, devise feedback dynamic strategy classify graph nodes groups then employ hierarchical aggregation. method extracts more discriminative features Extensive experiments on five datasets suggest that our proposed achieves 3% ~ 16% improvement over existing SOTA methods. Code is available at https://github.com/AtwoodDuan/DGA-GNN.