Abstract: To investigate the temporal evolution of the causes of coal mine accidents, we utilized a dynamic knowledge graph analysis framework to trace the evolution of accident-related knowledge. Using 82 coal mine accident reports from 2017 to 2025 as the primary data source, we constructed a semantic model of coal mine accident knowledge entities that integrates temporal features; Using a method that combines the BERT-BiLSTM-CRF model with rule templates, we automatically extracted knowledge regarding accident causes and constructed a dynamic knowledge graph of coal mine accidents; employing complex network analysis methods, we analyzed the evolutionary characteristics of accident cause knowledge in terms of network structure, nodes, risk-cause mappings, and pathways. The results indicate that the network structure of accident causes has evolved, with a dispersed structure gradually becoming more tightly coupled; accident causes have shifted from the behavioral-operational level to the institutional and systemic levels; management, technical, and behavioral risks exhibit significant differentiation in terms of exposure and transformation capacity; and early, relatively short key causal pathways have gradually evolved into cross-level, multi-stage transmission chains. Based on these evolutionary patterns, it is recommended that safety governance shift from post-incident correction to pre-incident prevention, and from single-point control to systemic coordination. This provides a knowledge foundation and decision-making basis for the transformation of China’s coal mine safety governance from behavioral control to systemic governance.