Review of Supply Chain Risk Propagation Mechanisms: A Graph Neural Network Perspective

As the division of labor deepens within global industrial chains and international trade uncertainties intensify, supply chain security issues are becoming increasingly prominent. In recent years, emergencies such as the COVID-19 pandemic and international conflicts have frequently triggered supply chain disruptions, exposing the vulnerability of supply chain networks to external shocks. Meanwhile, supply chain networks are inherently complex, highly interconnected, and exhibit strong contagion effects, leading to risks that propagate along network structures in cascading and nonlinear patterns. Accurately characterizing these mechanisms has thus become a critical issue in supply chain risk management. Existing studies primarily rely on traditional approaches such as complex network analysis and Bayesian networks to model supply chain risk propagation. However, due to the structural complexity and dynamic evolution of such networks, these methods face limitations in capturing higher-order dependencies, nonlinear propagation behaviors, and dynamic interaction mechanisms. With the advancement of Graph Neural Networks (GNNs), their strengths in modeling graph-structured data, learning relational dependencies, and extracting dynamic features have made them well-suited for supply chain contexts. Consequently, GNN-based approaches have been increasingly applied to risk identification, prediction, and propagation analysis. Nevertheless, a systematic synthesis of GNN applications and methodological developments in supply chain risk propagation remains lacking. To address this gap, this paper reviews representative studies from the perspectives of network construction, mechanism modeling, and feature enhancement, analyzes their key ideas, contributions, and limitations, and provides theoretical insights and practical guidance for future research.