Analyzing the supply chain network created from the conventional relation extraction model and ChatGPT: Focusing on S&P 500 companies

Abstract

Identifying key supply chains of companies to pinpoint potential threats in the supply chains of companies or items, and finding alternative suppliers is crucial. In this study, we built a pipeline to create supply chains using a relation extraction model based on 10K reports. We extracted relationships from the 2022 10K reports of S&P 500 companies using the conventional relation extraction model SSAN and the large-scale language model ChatGPT. We refined the network using word embeddings through SBERT. The outputs of both models were compared, and the created supply networks were classified into countries, companies, materials, and technologies, followed by visualization and statistical analysis to determine the most critical nodes in the supply chains. This research is expected to aid in utilizing large-scale language models and quickly and easily provide users with hidden connections between companies or specific information about a particular company.