Coal mine safety accidents are characterized by sudden occurrence, severe consequences, complex causal structures, and deeply embedded causal chains, posing significant threats to personnel safety and operational stability. Under such circumstances, conducting structured, systematic, and inference-capable causation analysis is essential for improving root-cause identification efficiency and supporting rescue operations and subsequent risk control. To address limitations in traditional accident analysis—such as dispersed causal information, insufficient structural representation, and difficulty in tracing deep causes—a knowledge-graph-based approach for coal mine accident causation analysis is constructed. By extracting semantic entities and relations such as “Coal Mine—Accident Type—Direct Cause—Indirect Cause” from accident case texts, a traceable causal chain structure is formed, enabling complex accident causes to be represented and inferred within a unified graph framework. Using real accident cases, extensive causal relations involving human behaviors, management deficiencies, equipment status, ventilation conditions, and gas control issues are extracted, and multiple analyses including causal chain length, high-frequency cause distribution, PageRank ranking are performed to characterize the structural features of the accident causation network and identify critical risk nodes. To evaluate the effectiveness of the knowledge-graph-based inference, comparisons are conducted with Bayesian Networks, BERT-GNN, and large language models, using Precision, Recall, and F1-score as evaluation metrics. Experimental results show that the knowledge graph achieves the highest performance across all metrics, demonstrating stronger capability in revealing multi-factor coupling mechanisms within accident evolution and identifying key causative factors related to human errors, management issues, equipment failures, poor ventilation, and insufficient gas control. Overall, the knowledge-graph-based causation analysis framework provides a structured representation and transparent inference mechanism suitable for complex industrial safety scenarios, supporting accident tracing, risk monitoring, and hazard management, and offering a technological pathway for advancing the intelligent development of coal mine safety governance.