Industrial chain risk monitoring with explainable dynamic knowledge graph learning

The increasing complexity and interconnectedness of global supply chains have made proactive systemic risk identification a critical challenge. Existing methods often struggle to bridge the semantic gap between high-frequency multi-modal behavioral data and structured domain knowledge, particularly in unsupervised settings where risk labels are scarce. To address this issue, we propose UCGF, an unsupervised cross-graph fusion framework for resilient supply chain risk identification. UCGF consists of three stages: a self-supervised dynamic graph encoder for capturing spatio-temporal anomalies from micro-level enterprise behaviors, a contrastive alignment module for mapping dynamic behavioral representations into a static knowledge graph space without manual supervision, and a neuro-symbolic reasoner for performing multi-hop propagation on the knowledge graph to identify interpretable risk conduction paths. Experiments on three complementary benchmarks, including a proprietary Channel Supply Chain dataset, a public US Supply Chain Risk Analysis dataset, and a macro-level OECD-ICIO benchmark, show that UCGF consistently outperforms state-of-the-art baselines across evaluation metrics. • Proposes an unsupervised cross-graph fusion framework for supply chain risk identification. • Aligns dynamic anomaly subgraphs with knowledge graph semantics via contrastive learning. • Delivers accurate and explainable risk prediction with evidence paths across three benchmarks.