Graph-Driven Knowledge Reasoning for Human AI Collaborative Strategic Decision Support in Supply Chain Networks

Abstract

Strategic decision support in supply chain networks remains constrained by fragmented data structures, limited visibility of multi-tier dependencies, and weak integration between relational analytics and managerial prioritization. This study proposes a graph-driven knowledge reasoning framework that combines knowledge graph construction, graph-based inference, multi-criteria recommendation ranking, and scenario-based strategic evaluation to support decision-making in complex supply chain environments. The modeled network comprised 248 nodes and 612 directed edges representing suppliers, factories, warehouses, transport routes, retailers, and disruption events. Structural analysis showed an average degree of 4.94, with suppliers forming the largest entity class at 72 nodes, while factories exhibited the highest relational intensity with an average degree of 7.4. The reasoning engine identified several high-risk propagation corridors, including the path S12 → F3 → R9 with a risk score of 0.91, indicating concentrated supplier dependence and high downstream delay exposure. Additional critical paths, such as S21 → F6 → W2 and S4 → W7 → T5, recorded scores of 0.84 and 0.81, revealing capacity bottlenecks, bridge-node fragility, and regionally amplified service disruption. Entity-level inference further showed that Warehouse W7 and Route R9 were strategically important despite not being dominant in raw node volume, confirming that graph-based reasoning can reveal hidden structural vulnerabilities overlooked by conventional analysis. In the decision support phase, supplier diversification obtained the highest composite recommendation score at 8.14, followed by inventory repositioning at 7.96 and route redundancy at 7.88. Comparative scoring indicated that supplier diversification produced the strongest resilience value at 8.9, while inventory repositioning achieved the best balance between feasibility and cost efficiency. Scenario analysis demonstrated that diversification performance increased from 7.6 under normal conditions to 8.8 under severe disruption, whereas inventory repositioning declined from 8.0 to 7.1, indicating that short-term stabilization strategies lose effectiveness under deeper structural stress. Expert-oriented validation also reported high managerial acceptance, with average scores of 4.6 for strategic relevance, 4.5 for interpretability, and 4.4 for practical usefulness. These findings show that graph-driven knowledge reasoning can improve strategic visibility, detect hidden dependencies, and generate interpretable, context-sensitive recommendations for supply chain decision support.