Enhancing Supply Chain Visibility with Generative AI: An Exploratory Case Study on Relationship Prediction in Knowledge Graphs

TL;DR

The paper tackles the challenge of limited visibility in complex supply networks by marrying pretrained language models with knowledge graphs to predict contextual, multi-step relationships, termed quintuplets. It proposes a hybrid embedding framework where GenAI-derived textual snippets anchor a KG in the LM’s relational knowledge, while a downstream ML model grounds predictions to factual graph data, mitigating hallucinations. Through a real-world case study on automotive and EV battery supply chains, the approach consistently outperforms traditional ML baselines in predicting quintuplets, and demonstrates robustness across multiple LM backbones and model architectures. The work offers a practical pathway for enhancing supply chain visibility and risk management, with clear implications for governance, data sharing, and future extensions to additional contextual knowledge types.

Abstract

A key stumbling block in effective supply chain risk management for companies and policymakers is a lack of visibility on interdependent supply network relationships. Relationship prediction, also called link prediction is an emergent area of supply chain surveillance research that aims to increase the visibility of supply chains using data-driven techniques. Existing methods have been successful for predicting relationships but struggle to extract the context in which these relationships are embedded - such as the products being supplied or locations they are supplied from. Lack of context prevents practitioners from distinguishing transactional relations from established supply chain relations, hindering accurate estimations of risk. In this work, we develop a new Generative Artificial Intelligence (Gen AI) enhanced machine learning framework that leverages pre-trained language models as embedding models combined with machine learning models to predict supply chain relationships within knowledge graphs. By integrating Generative AI techniques, our approach captures the nuanced semantic relationships between entities, thereby improving supply chain visibility and facilitating more precise risk management. Using data from a real case study, we show that GenAI-enhanced link prediction surpasses all benchmarks, and demonstrate how GenAI models can be explored and effectively used in supply chain risk management.

Figures (5)

• Figure 1: (a) company-level relationships in a supply chain network brintrup2018predicting; (b) multiple relationships including (company, supplies to, company), (company, has, certificate) and (company, has, product) in a knowledge graph kosasih2022towards; (c) quintuplet based relationships on a knowledge graph. For example, company A supplies product 1 to company B, and company A with certificate 1 has product 1. Figure 1 Alt-text: The diagram shows company-level relationships within a supply chain network, multiple relationships in a supply chain knowledge graph, and quintuplet-based relational structures within the knowledge graph.
• Figure 2: The pretrained LM-enhanced supply chain link prediction framework. Figure 2 Alt-text: The diagram presents the pretrained LM-enhanced supply chain link prediction framework.
• Figure 3: (a) results on quintuplet (company, supplies, product, to, company) achieved by different machine learning models (b) results achieved by pretrained LM-enhanced machine learning models with "all-MiniLM-L12-v2" Figure 3 Alt-text: The diagram shows results on quintuplet (company, supplies, product, to, company) achieved by different machine learning models and also the results achieved by pretrained LM-enhanced machine learning models with "all-MiniLM-L12-v2".
• Figure 4: (a) shows results of predicting relationships in the quintuplet of (company, with, certificate, has, product) achieved by five machine learning models on all countries' datasets while (b) presents the results achieved by our proposed approach using five machine learning models empowered by the pretrained LM, "all-MiniLM-L12-v2". Figure 4 Alt-text: The diagram presents results of predicting relationships in the quintuplet of (company, with, certificate, has, product) achieved by five machine learning models on all countries' datasets, and also the results achieved by our proposed approach using five machine learning models empowered by the pretrained LM, "all-MiniLM-L12-v2".
• Figure 5: (a) results of predicting relationships in a quintruplet of (company, supplies, product, to, company) achieved by pretrained LM-enhanced CNN (b) results of predicting relationships in quintuplet (company, with, certificate, has, product). Figure 5 Alt-text: The diagram shows results of predicting relationships in a quintuplet of (company, supplies, product, to, company) achieved by pretrained LM-enhanced CNN, and also the results of predicting relationships in quintuplet (company, with, certificate, has, product).