Predicting Hidden Supply Network Dependencies




Predicting and extracting dependencies from the "hidden" or "invisible" parts of a supply network is a key area of ongoing research for the Supply Chain AI Lab, which has led to a number of innovative and impactful approaches, collaborations and a start up,


Manufacturing companies lack visibility of the procurement interdependencies between the suppliers within their supply network. However, knowledge of these interdependencies is useful to plan for operational disruptions, as well as tracking wider,  cascading risk in the supply chain ecosystem, which may range from identifying suppliers that engage in unethical and  unsustainable practices, to identifying geopolitical risk concentrations.


The Supply Chain AI Lab leads a number of activities Supply Network Reconstruction and Link Prediction, in order to increase visibility of supply networks, which in turn enables us to create strategies for improved resilience, and sustainability.


These efforts began with the 2015 VIPr project, where we developed the Supply Chain Miner, a Natural Language Processing approach that enabled the extraction and prediction of company to company relationships from the World Wide Web. 


We have expanded on this approach by using NLP to extract a knowledge graphs, where we are not only extracting supply-buy relations, but also other dependencies, such as production capabilities.  


While NLP provides us with a baseline to start from, it needs to be complemented by other mechanisms to inform dependencies. We have also developed the Supply Network Link Predictor (SNLP) to infer supplier interdependencies using the manufacturer’s incomplete knowledge of the network. SNLP uses topological data to extract relational features from the known network to train a classifier for predicting potential links. 


In the Aviva sponsored ChainedRisk project, we created multiple approaches that predict supplier links using Graph Neural Networks,  Knowledge Graphs and ultimately, Graph Reasoning. Each of these approaches build on public and private datasets, giving rise to exciting tools that make automated supply chain mapping and Digital Supply Chain Surveillance possible. 


See our research in action at the Office for National Statistics here


ResearchersEdward Kosasih, Liming Xu, Sara Almahri, Nils Brockmann, Phillip Schaffer, Jonathan Chen


Collaborators and funders: Aviva, Boeing, EPSRC


Share This