AI-Based Explainable Classification Of News Articles For Supply Chain Risk Management

Sammanfattning

This study focuses on the challenges of using artificial intelligence (AI) in supply chain risk management, particularly the lack of transparency and trust that makes it harder for stakeholders to adopt AI technologies. To address this, the project aims to apply an explainable AI method to classify risks from unstructured data, such as news articles, making the results easier to understand and trust. By analyzing three years of news data, this research provides a comprehensive approach to detecting supply chain disruptions using interpretable AI models. Large Language Models (LLMs), such as Llama 2 and Llama 3, were utilized to automate data annotation, creating a dataset that reflects real-world supply chain risks. To ensure transparency, the proto-lm framework was integrated, combining prototypical networks with LLMs to provide a clear view of model decisions. This framework not only improves classification accuracy but also makes it possible to explain predictions through representative examples, enhancing stakeholders' trust in AI-generated insights. Results show that among all applied LLMs, Llama 2-70B achieved higher classification accuracy, aligning closely with human-annotated data and offering more reliable insights into supply chain risk factors. This research demonstrates that explainable AI can drive the adoption of advanced models in critical domains by addressing transparency and trust, enabling organizations to manage risks with informed data-driven decisions.