Mining the Text: Automating Safety Insights from Mining Accident Reports

Mining safety remains a critical focus globally, with continuous efforts to reduce accidents and enhance safety standards. A key component of these efforts involves analyzing mining accident reports—comprehensive textual resources that provide valuable insights into the root causes and contributing factors of accidents. This study applies tools from Natural Language Processing and Topic Modeling to the analysis of fatality reports obtained from the United States Mine Safety and Health Administration (MSHA). This work seeks to automate the identification of latent themes and patterns within accident reports, streamlining the process for enhanced safety insights. A Latent Dirichlet Allocation model, optimized using Term Frequency-Inverse Document Frequency (TF-IDF) weighting and the Coherence Value metric were used to determine the optimal number of topics. Extracted topic keywords were processed using ChatGPT-4o, a general large language model (LLM), to generate coherent and interpretable topic summaries. The results demonstrate the potential of combining topic modeling with LLMs to streamline mining accident analysis, reducing manual review efforts while extracting actionable safety insights. However, findings also highlight challenges in language ambiguity, domain-specific terminology, and misclassification, reinforcing the need for tools that consider domain adaptation. This study underscores the importance of responsible AI deployment in occupational safety, ensuring that AI-driven solutions complement human decision-making rather than replace it.