Abstract: The drill-and-blast method with favorable efficiency,applicability,and controllability has become one of thefrequently-used techniques in tunnel construction.However,insufficient attention to the spatiotemporal coupling effects of safety risk factors during construction is prone to accidents.The current research heavily relies on expert experience,overlooks accident case data and complex interactions between risk factors,thereby reducing the credibility of safety risk analysis and assessment.This study proposed a joint extraction method for safety risk factors and their relationships,based on Bert,BiLSTM,multi-head attention,and densely connected graph convolutional networks (BBi-MA-DCGCN).This study constructed a knowledge graph of risk coupling evolution,proposed an intelligent inference method for safety risk coupling evolution paths and risk factor importance assessment method by integrating knowledge graph data,path inference algorithms,and interaction matrix principles.The results show that:(1)The BBi-MA-DCGCN model achieves an F1 score of 79.89% on the self-constructed dataset,demonstrating strong robustness in entity and relationship extraction;(2)The coupling evolution inference system can quickly deduce the most likely evolution path from a given risk node,identifying 10 critical edges and 3 key chains related to collapse and mud inflow incidents;(3)The risk assessment method considers the coupling effects of risk factors and identifies 10 key points,including surrounding rock deformation and errors in advanced geological forecasting.The proposed data- and knowledge-driven method has improved the accuracy and reliability of safety risk analysis and assessment in drill-and-blast tunnel construction,with theoretical and practical value.