Industrial Safety Management in the Digital Era: Constructing a Knowledge Graph from Near Misses

This paper proposes a methodology to extract knowledge from near-miss reports in oil refinery operations using knowledge graphs built in Neo4j. The approach applies an ontological model to natural language reports from Seveso Directive-regulated facilities, enabling systematic explorative analysis of incident precursors. For MOC in process safety, learning from near misses is essential to inform change risk assessments. When evaluating a proposed modification, an AI-powered knowledge graph of near-miss data can reveal historical patterns showing which types of changes have previously led to elevated risk, what causal factors were involved, and which safeguards proved most effective—enabling evidence-based MOC decision-making rather than relying solely on expert judgment.