Graph Neural Networks for Multi-Tier Supply Chain Disruption Risk with Incomplete Supplier Knowledge
Graph Neural Networks for Multi-Tier Supply Chain Disruption Risk with Incomplete Supplier Knowledge
Description
Details
Context and Problem Statement
Organizations usually know direct suppliers but have limited visibility into second- and third-tier dependencies. Existing risk models often assume a complete supply graph, while inferred links can themselves introduce uncertainty and false alarms.
Research Question
How can uncertainty in supply-graph reconstruction be propagated into downstream disruption-risk prediction so that confirmed risk can be distinguished from risk arising from uncertain inferred links?
Proposed Approach
Construct a supply knowledge graph from open and enterprise data, predict missing links with confidence scores, propagate risk through a GNN that incorporates edge uncertainty, and identify critical dependency paths for explanation.
Expected Contribution
An evaluation framework separating errors caused by graph reconstruction from errors caused by the risk model.
Expected Prototype
A risk-analysis application showing critical dependency paths, confidence levels, and simulated disruption scenarios.
Datasets
SupplyGraph, open customs and company-register data, and partner enterprise data.
Challenges
Commercial confidentiality, sparse disruption labels, open-source bias, and potential misuse of reconstructed supplier networks.
Research Question
Innovation
Expected Deliverable
Technologies
Required Skills
- Graph Machine Learning
- Knowledge Graphs and Data Modeling
- Statistics and Uncertainty Quantification
- Supply Chain Management Fundamentals
Datasets
- SupplyGraph
- Open customs and company-register data
- Internal data from an industrial partner