[Paper Review] Enhancing Supply Chain Visibility with Knowledge Graphs and Large Language Models

Sara AlMahri, Liming Xu+1 more|arXiv (Cornell University)|Aug 5, 2024

Data Quality and ManagementDecision Sciences3 citations

TL;DR

This paper proposes a zero-shot, LLM-driven framework that constructs supply chain knowledge graphs from public web sources to enhance visibility without relying on direct information sharing. By leveraging zero-shot prompting for named entity recognition and relation extraction, the method maps multi-tier supplier networks—demonstrated on electric vehicle battery minerals—with high accuracy, revealing hidden dependencies and alternative sourcing options beyond tier-2 suppliers.

ABSTRACT

In today's globalized economy, comprehensive supply chain visibility is crucial for effective risk management. Achieving visibility remains a significant challenge due to limited information sharing among supply chain partners. This paper presents a novel framework leveraging Knowledge Graphs (KGs) and Large Language Models (LLMs) to enhance supply chain visibility without relying on direct stakeholder information sharing. Our zero-shot, LLM-driven approach automates the extraction of supply chain information from diverse public sources and constructs KGs to capture complex interdependencies between supply chain entities. We employ zero-shot prompting for Named Entity Recognition (NER) and Relation Extraction (RE) tasks, eliminating the need for extensive domain-specific training. We validate the framework with a case study on electric vehicle supply chains, focusing on tracking critical minerals for battery manufacturing. Results show significant improvements in supply chain mapping, extending visibility beyond tier-2 suppliers. The framework reveals critical dependencies and alternative sourcing options, enhancing risk management and strategic planning. With high accuracy in NER and RE tasks, it provides an effective tool for understanding complex, multi-tiered supply networks. This research offers a scalable, flexible method for constructing domain-specific supply chain KGs, addressing longstanding challenges in visibility and paving the way for advancements in digital supply chain surveillance.

Motivation & Objective

• Address the persistent challenge of limited supply chain visibility due to restricted information sharing among partners.
• Overcome the limitations of traditional visibility methods that depend on direct collaboration or proprietary data sharing.
• Develop a scalable, automated framework to extract and structure supply chain information from publicly available, unstructured sources.
• Enable end-to-end mapping of complex, multi-tiered supply networks, particularly for critical materials like those in electric vehicle batteries.
• Support improved risk management and strategic decision-making by uncovering hidden dependencies and alternative sourcing options.

Proposed method

• Utilize zero-shot prompting in large language models (LLMs) for named entity recognition (NER) and relation extraction (RE) to identify suppliers, materials, and relationships from unstructured text.
• Construct domain-specific knowledge graphs (KGs) by extracting structured triplets (subject, relation, object) from diverse public sources such as Wikipedia, news, and reports.
• Integrate LLMs to infer relationships between entities without fine-tuning on domain-specific datasets, enabling zero-shot generalization.
• Aggregate and normalize entities across multiple sources to build a unified, multi-tiered supply chain network representation.
• Apply prompt engineering to guide LLMs in identifying complex dependencies, such as material flows and supplier-customer chains, across tiers.
• Validate the KG with a case study on electric vehicle supply chains, focusing on critical minerals for battery production.

Experimental results

Research questions

• RQ1 Can LLMs effectively extract supply chain entities and relationships from unstructured public text without fine-tuning or domain-specific training?
• RQ2 To what extent can a zero-shot LLM-driven approach extend supply chain visibility beyond tier-1 and tier-2 suppliers?
• RQ3 How accurately can such a framework identify critical dependencies and alternative sourcing options for key materials in complex supply chains?
• RQ4 What is the impact of using publicly available data sources on the completeness and reliability of the resulting knowledge graph?
• RQ5 How can knowledge graphs constructed via LLMs support strategic risk management and decision-making in supply chain networks?

Key findings

• The zero-shot LLM approach achieved high accuracy in both named entity recognition and relation extraction tasks, enabling reliable extraction from unstructured public text.
• The framework successfully mapped supply chain networks extending beyond tier-2 suppliers, revealing previously unknown interdependencies in electric vehicle battery material flows.
• The constructed knowledge graph identified critical dependencies on specific suppliers for key minerals such as lithium and cobalt, enabling proactive risk assessment.
• Alternative sourcing options were uncovered, providing strategic insights for supply chain resilience and diversification planning.
• The framework demonstrated scalability and adaptability across different supply chain domains, with potential for application in pharmaceuticals, semiconductors, and consumer goods.
• The knowledge graph provided actionable insights for decision-makers, enhancing transparency and supporting ethical and sustainable sourcing strategies.