Graph Neural Networks for Modeling Complex Dependencies in Global Supply Chain Networks

Abstract

Global supply chain networks exhibit intricate dependencies characterized by multi-tier supplier relationships, dynamic demand propagation, cascading disruption effects, and complex material flows across geographically distributed entities. Traditional analytical approaches struggle to capture these interdependencies due to their reliance on simplified assumptions about network structure and information flow. Graph neural networks (GNN) have emerged as a powerful framework for modeling complex relational data by learning representations that encode both node attributes and graph topology through message passing mechanisms. This review examines the application of GNN to supply chain network modeling, focusing on how these methods capture dependency structures, predict disruption cascades, optimize network configurations, and facilitate decision-making under uncertainty. We explore fundamental GNN architectures including graph convolutional networks (GCN), graph attention networks (GAT), and graph recurrent networks, analyzing their suitability for different supply chain modeling tasks. The paper investigates how GNN approaches model multi-tier visibility, demand forecasting with network effects, risk propagation analysis, and supplier relationship dynamics. Key applications examined include disruption prediction, inventory optimization across network echelons, supplier selection considering indirect dependencies, and resilience assessment for global supply chains. This comprehensive analysis reveals that GNN methods demonstrate superior performance in capturing non-local dependencies and modeling complex interaction patterns compared to traditional supply chain analytics. The paper concludes by identifying critical research directions including temporal GNN for dynamic supply chain evolution, heterogeneous GNN for multi-modal supply chain data

Topics & Concepts