Using Large Language Models and Knowledge Graphs to Improve the Interpretability of Machine Learning Models in Manufacturing
Summary: arXiv:2604.16280v1 Announce Type: new
Abstract
Explaining Machine Learning (ML) results in a transparent and user-friendly manner remains a challenging task of Explainable Artificial Intelligence (XAI). In this paper, we present a method to enhance the interpretability of ML models by using a Knowledge Graph (KG). We store domain-specific data along with ML results and their corresponding explanations, establishing a structured connection between domain knowledge and ML insights.
Introduction
The rapid integration of Machine Learning (ML) in manufacturing processes has led to significant advancements in efficiency and productivity. However, the black-box nature of many ML models presents challenges in terms of transparency and trust. To address these issues, our research focuses on the intersection of Explainable Artificial Intelligence (XAI) and practical manufacturing applications.
Methodology
To make insights accessible to users, we designed a selective retrieval method in which relevant triplets are extracted from the Knowledge Graph (KG) and processed by a Large Language Model (LLM) to generate user-friendly explanations of ML results. The process can be summarized as follows:
- Data Storage: Domain-specific data and ML results are stored in the KG.
- Knowledge Retrieval: Relevant triplets are identified based on user queries.
- Explanation Generation: The LLM generates coherent explanations tailored to user needs.
Evaluation
We evaluated our method in a manufacturing environment using the XAI Question Bank. Our evaluation included:
- Standard questions to assess basic interpretability.
- Complex, tailored questions designed to highlight the strengths of our approach.
In total, we analyzed 33 questions, using both quantitative metrics such as accuracy and consistency, and qualitative metrics such as clarity and usefulness.
Results
The results indicated that our method significantly improved the interpretability of ML models in the manufacturing context. Users reported higher satisfaction with the explanations generated, noting that they were more understandable and relevant to their decision-making processes.
Conclusion
Our contribution is both theoretical and practical. From a theoretical perspective, we present a novel approach for effectively enabling LLMs to dynamically access a KG to improve the explainability of ML results. From a practical perspective, we provide empirical evidence showing that such explanations can be successfully applied in real-world manufacturing environments, supporting better decision-making in manufacturing processes.
Future Work
Future research will focus on expanding the application of our method to other sectors, enhancing the scalability of the KG, and improving the integration of user feedback to continuously refine the explanation generation process.