Multi-Scale Adaptive Neighborhood Awareness Transformer For Graph Fraud Detection

Researchers propose MANDATE, a Multi-scale Neighborhood Awareness Transformer that improves graph fraud detection by addressing limitations of traditional graph neural networks. The system uses multi-scale positional encoding and different embedding strategies to better identify fraudulent behavior in financial networks and social media platforms.

• →MANDATE addresses inherent biases in graph neural networks including homogeneity assumptions and limited global modeling ability.
• →The system uses multi-scale positional encoding to capture information at various distances from central nodes.
• →Different embedding strategies are employed for homophilic and heterophilic connections to better distinguish benign from fraudulent nodes.
• →An embedding fusion strategy handles multi-relation graphs to reduce distribution bias from different relationship types.
• →Experiments on three fraud detection datasets demonstrate MANDATE's superior performance over existing methods.