Application of News Analysis Based on Large Language Models in Supply Chain Risk Prediction

This study investigates using large-scale linguistic models (LLM) in media analysis to predict stock market risk. This research combines the best language processing techniques with traditional real-time analysis to create a comprehensive approach to identifying and predicting product impact. A database of 200,000 articles from 2018 to 2023 is used to train and evaluate the needs of the LLM. The model's performance has been rigorously compared with the baseline methods, including TF-IDF with logistic regression, BERT, and LSTM with monitoring methods. The results showed the best performance of the LLM-based method, achieving an F1 score of 0.883 for risk classification and a percentage of uncertainty of 9.3 % for risk estimation. Case studies of specific events, including the COVID-19 pandemic and the Suez Canal blockage, add validity to the forecasting model's capabilities, with operational time ranging from 1 to 4 weeks. The research also addresses the interpretation of the model through visualisation and factor analysis, providing insight into the principal risks of different groups. This study contributes to the risk management industry by providing a new, data-driven approach that leverages the power of LLMs for early risk detection and awareness. Decision-making in the global supply chain is complex.