Bill Chain Risk Detection in Industrial Enterprises Through Multi-Relational Graph Inference

Abstract

Commercial bills are important financing and payment instruments for industrial enterprises, but bill-chain risk may arise from repeated endorsement, circular circulation, weak acceptor credit, abnormal discounting, related-party transfer, and hidden debt rollover. Conventional bill risk assessment mainly reviews single bill attributes and may overlook risk accumulation across endorsement networks. This study proposes a multi-relational graph inference method for bill chain risk detection in industrial enterprises. The method constructs a bill circulation graph linking drawers, acceptors, endorsers, discounters, guarantors, banks, invoice records, and payment outcomes. A relational graph neural network captures multi-hop risk dependency across endorsement chains, while a knowledge inference module detects suspicious patterns such as circular endorsement, short-cycle discounting, repeated transfer among affiliated firms, and acceptor credit deterioration. Experiments are conducted on an industrial bill dataset containing 62,000 enterprises, 1.28 million commercial bills, 3.46 million endorsement edges, 210,000 discounting records, 74,000 guarantee relations, and 11,600 bill-risk events over 36 months. The proposed method identifies 8,940 high-risk bill chains, including 2,360 circular endorsement structures and 1,780 abnormal short-cycle discounting paths. Compared with a single-bill scoring baseline, the model shortens median warning time before bill default or delayed acceptance from 49 days to 17 days. Graph-based aggregation compresses 26,500 bill-level alerts into 5,720 chain-level review cases. The inference engine processes 3.46 million endorsement edges in 16.4 minutes and completes daily incremental updates in 42 seconds. The results demonstrate that multi-relational graph inference can improve industrial bill-chain risk detection by revealing hidden financing pressure and contagion paths across enterprise payment networks.