Detection of false financial statements in enterprises based on dual-layer knowledge graph and graph-driven approach

In recent years, enterprise financial fraud has become increasingly sophisticated. Traditional detection methods are difficult to capture the deep correlation between variables, which affects the healthy and sustainable development of enterprises. To address this challenge, this study proposes a novel framework integrating a dual-layer knowledge graph with graph-driven analysis. The research begins by organizing and preprocessing enterprise financial statement data to identify key indicators of false statements. Subsequently, the study proposes a false statement detection model based on a dual-layer knowledge graph, achieving fraud detection through the dual-layer knowledge graph and rule mining. Finally, based on the graph neural network, the research proposes a graph-driven detection model, which realizes the capture of complex dynamic fraud camouflage relationships through the design of a three-level attention mechanism. In accuracy tests across different feature scales, the research model achieved a peak accuracy of 93.4%, outperforming comparable models. In tests involving hidden fraudulent relationship data, the research model demonstrated the highest identification accuracy at 86.2%. Additionally, in the identification accuracy test of the enterprise-audit firm relationship diagram, the proposed model achieved an identification accuracy of 0.905 at 40 nodes, outperforming comparable models. Additionally, the research model demonstrated the best performance in both data processing time and resource utilization tests. Overall, the proposed technology shows promising application potential in detecting false statements within enterprises. This study provides technical support for enterprise risk early warning and financial fraud detection.