A Macro-to-Micro Analytical Framework for Interpreting Context-Stratified Accident Patterns in Construction

Accident knowledge graphs can represent rich relational structures. However, interpreting recurring accident-pattern associations remains difficult when analysis relies on a single scale. Macro-level analysis tends to remain descriptive, whereas microlevel mining can be confounded by globally frequent regularities, blurring trade- and context-specific differences. To address this gap, this study proposes an ontology-based accident knowledge graph, a macro-to-micro framework, and models 29,463 accident records (2019–2025) as an event-centric heterogeneous graph. Each record is represented as an ordered full-chain attribute combination to support reproducible pathway-pattern comparison. Importantly, the proposed pathways are not intended to establish causal chains; rather, they represent frequency-supported association patterns among structured accident attributes. At the macro stage, PageRank-based centrality and total variation (TV) distance screening align comparison baselines, and robustness checks verify the stability of work-type selection for micro analysis. At the micro stage, context strata enable like-for-like comparisons to examine persistence and differentiation of pathway patterns given unfiltered versus condition-fixed settings. Results identify a cross-work-type backbone aligned with temporary facilities and installation/dismantling phases and clarify conditional pivots by workforce size and project progress rate. By linking global network signals with context-specific pathway patterns, this study provides a two-layer safety-planning logic that distinguishes consistent monitoring targets from context-sensitive control priorities and offers an interpretable, stability-oriented foundation for data-driven construction safety management.