Graph Contrastive Learning for Internet Financial Fraud Detection

With the rapid development of Internet financial business, the risk of financial fraud is increasing. How to effectively detect fraud has always been a highly concerned issue. Based on graph contrastive learning (GCL), we propose a fraud detection model called FFD-GCL. For the FFD-GCL model, we design an adaptive graph data augmentation strategy, including topology augmentation based on graph attention mechanism and node attribute augmentation based on feature importance, and use the contrast loss that considers the topology of the network to enhance the applicability of the model for financial transaction networks. The experimental results demonstrate the effectiveness of the FFD-GCL model.