Industrial-Scale Data Processing and LLM Evaluation for Coal Mine Safety
Year: 2025
Type: conference-paper
Abstract: Industrial information processing in safety-critical domains requires the integration of heterogeneous data sources and the transformation of domain expertise into actionable knowledge. To address these challenges, this paper proposes a data processing framework for industrial large language models (LLMs) that standardizes multisource data integration, semantic cleaning, and quality monitoring for large-scale model training. Based on this framework, we construct Mine Safety Intelligence(MSI), a multisource and multitask corpus containing 10 GB of domain text and 10,000 high-quality instruction–response pairs covering safety regulations, emergency response, and risk management. Using this data set, we develop MS-GLM, a domain-adaptive LLM for coal mining safety applications. To ensure reliable evaluation, we establish a dedicated benchmark (CMS-QED) and an expert-driven scoring framework (CMS-QEM) that jointly assess accuracy, interpretability, and user experience. The experimental results demonstrate that MS-GLM achieves substantial improvements in factual accuracy, reasoning precision, and understanding of safety compliance compared to baseline models. These findings highlight the effectiveness of structured data processing and domain-aligned evaluation frameworks in advancing intelligent decision support for industrial information systems.
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7127045794
