This is the authors' abstract. We don't add key points for this paper.
Industrial explosion accidents often cause severe casualties, property damage, and environmental impacts, posing major challenges to process safety and emergency management. This study constructs an industrial-explosion eventic graph grounded in a domain-specific ontology and implements a Retrieval-augmented Generation (RAG) Q&A system powered by large language models (LLM) to support emergency decision-making. We designed an accident-emergency ontology that systematically captured accident characteristics and response workflows. A zero-shot information-extraction framework automatically identifies events from historical reports, and template-based matching extracts inter-event relations. Uncertainty modeling is introduced to ensure accurate knowledge representation. A semantic-similarity-driven knowledge-fusion method improves event abstraction and consistency, and the resulting graph is stored in Neo4j for efficient querying and analysis. By integrating the eventic graph with RAG, we created a Q&A system that significantly outperforms baseline models and traditional reasoning methods. A case study of the 8·12 Tianjin Port explosion demonstrates the framework’s ability to represent accident evolution patterns and causal chains. This integrated approach provides a practical tool for accident investigation, risk assessment, and emergency decision-making, contributing to improved safety management in industrial processes.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: