Knowledge graph attention network with multilevel contrastive learning for safety hazard rectification

In the process of coal mine safety hazard rectification, rapidly eliminating hazards without compromising safe production or endangering workers’ lives is a critical challenge in coal mine safety management. Effectively preventing the propagation and escalation of safety hazards within a limited time frame remains an urgent issue to be addressed. To this end, this study first constructs a domain-specific knowledge graph tailored to coal mine safety hazards. Based on this structured knowledge, we then propose an enhanced Knowledge Graph Attention Network model integrated with multilevel contrastive learning. The proposed model helps alleviate the adverse effects of long-tail distributions commonly encountered in knowledge graph–based recommendation tasks and enables intelligent recommendation of rectification measures for coal mine safety hazards. To verify the effectiveness of the proposed approach, comparative experiments are first conducted on two public benchmark datasets. Based on the experimental performance on these datasets, further validation is carried out on the coal mine safety hazard knowledge graph dataset constructed in this study. The results demonstrate that the proposed model exhibits superior recommendation performance and robustness. This research provides a novel perspective for the intelligent recommendation of coal mine safety hazard rectification measures and offers effective technical support for the timely control and mitigation of safety hazards in coal mines.