LLM-Based Reconstruction of Mining Supply Chain Networks: Evidence on Concentration, Geography, and Fragility
Name
applsci-16-04536-v2.pdf
Size
911.19 KB
Format
Adobe PDF
Checksum (MD5)
261526b9f0778d8d3d637a1d4f29d14d
Author(s)
Kegenbekov, Zhandos
Alipova, Alima
Jackson, Ilya
Date Issued
May 5, 2026
Journal
Applied Sciences
Publisher
MDPI
Citation
Kegenbekov, Z.; Alipova, A.; Jackson, I. LLM-Based Reconstruction of Mining Supply Chain Networks: Evidence on Concentration, Geography, and Fragility. Appl. Sci. 2026, 16, 4536.
Version
Final published version
Abstract
Mining supply chains are strategically important yet difficult to observe because firm-to-firm relationships are fragmented across unstructured public disclosures. This paper applies a retrieval-augmented large language model (LLM) framework to reconstruct mining supply chain networks from unstructured textual data. While the underlying methodology builds on prior work on supply chain mapping using generative AI, the focus here is on its deployment in the mining sector, where structured data on firm-to-firm relationships is particularly limited. The resulting dataset contains 1231 focal mining firms, 4602 directly connected firms, and 8279 directed supplier–customer relationships. The reconstructed network exhibits a pronounced core–periphery structure, with a small number of highly connected firms sustaining a disproportionate share of overall connectivity. We also document strong concentration across countries and mining subsectors, particularly among major mining finance and corporate hubs. Robustness analysis shows that the network is relatively resilient to random node removal but fragments rapidly under targeted removal of central firms. These findings suggest that mining supply chains combine diffuse peripheral structure with systemic dependence on key hub firms, and illustrate how LLM-based methods can serve as a scalable measurement technology for supply chain mapping in data-scarce settings.
Terms of Use
Creative Commons Attribution
Persistent DSpace Link
DOI of Published Version
https://doi.org/10.3390/app16094536
