Identifying Supply Chain Risks using Large Language Models

Abstract

Supply chains (SC) are large and complex networks that consist of the flow of physical material, information, and assets between companies. The complexity of SCs has become a considerable problem in recent years, causing SCs to become vulnerable to a growing number of risks. The SC environment is also becoming more uncertain, as accurate data about the SC is scarce, and the number of disruptions in the SC is ever-growing. In the face of an uncertain SC environment, conventional SC risk management (SCRM) tools are becoming outdated. Large language models (LLMs) have shown their ability to perform unformalized tasks, due to their exceptional natural language understanding and generation capabilities and specific domain knowledge acquired from a vast training dataset. LLMs have been successfully tailored and trialled for use cases in many academic fields, including SCRM. LLMs can be tailored to identify SC risks (SCR) from textual data by utilizing prompt engineering and hyperparameter tuning. Natural language processing tasks, such as semantic text similarity analysis and named entity recognition, can be used to extract SCRs from textual data. LLM-based implementations are robust and proactive methods to support the conventional SCRM process in the face of the modern uncertain SC environment. SCRs identified by these implementations, however, lack the universal support of SCRM domain experts. To develop the SCR identification capabilities of LLMs, AI and SCRM domain experts should co-operate in creating a general taxonomy of SCRs.