Supply Chain Financial Fraud Detection Based on Graph Neural Network and Knowledge Graph

Abstract

Supply chain financial fraud, characterized by extensive false fund circulation and fictitious business events, causes substantial financial losses and undermines the efficiency of supply chain operations. To address this challenge, we introduce an innovative research framework that utilizes knowledge graphs and spatial-temporal neural networks for effective fraud detection. Our approach involves constructing a supplier-customer knowledge graph from data of Chinese listed companies, capturing the complex supply-demand relationships within the supply chain. We designed a spatial-temporal Graph Neural Network (GNN) that models both node attributes and the time-evolving graph topology. By incorporating temporal and spatial dual attention mechanisms, our model adeptly identifies local topology and temporal changes in the knowledge graph. Empirical evaluations demonstrate that our Dual Attention Spatial-Temporal Graph Neural Network (DAST-GNN) outperforms existing methods, achieving an AUC of 93.64%, which is 10.41% higher than the leading machine learning methods. Furthermore, analyzing supplier-customer relationships across different historical periods enhances fraud detection, highlighting the robustness of our approach. This research offers a potent tool for regulators, investors, and researchers, advancing the security and efficiency of supply chain operations.

Keywords

• financial fraud
• graph neural network
• knowledge graph
• spatial-temporal attention
• supply chain network