Graph State Detection for Identifying Fictitious and Related Transaction Chains in Financial Networks

Abstract

This study addresses the difficulty of identifying fictitious transactions and related transaction chains in complex financial networks, where multi-hop dependencies, weak path signals, and concealed structural propagation often remain undetected by traditional methods. It proposes a graph state detection framework that focuses on global structural consistency. The method first encodes node attributes, transaction path features, and local topology into a unified multi-granularity structural representation. It then applies multi-layer structural propagation to extract cross-node dependencies and chain-level relations, allowing the model to capture hidden abnormal patterns within weak connections. A path attention fusion module is introduced to assign dynamic importance to different transaction chains and generate a graph state vector that reflects global structural variations. A graph-level aggregation mechanism further integrates multi-scale information to understand abnormal chain propagation from a holistic perspective. The framework maintains stable detection performance under noise, topology perturbation, and temporal structural drift, and it reveals the key structural features behind fictitious and related transaction chains. Experimental results show clear improvements in accuracy, precision, recall, and F1 score. The proposed method enhances the detection of concealed transactional chains and provides an efficient, scalable, and structure-sensitive solution for graph-based risk modeling in complex financial environments.