Abstract: To systematically identify the causes and correlations of rib spalling and roof fall accidents in underground metal mines, 139 accident investigation reports in China from 2013 to 2024 were selected as research samples. A text mining method for accident causation based on the 24Model was proposed. First, based on the theoretical framework of the 24Model, a MacBERT−BiLSTM text classification model was constructed for the hierarchical classification of accident causation, integrating MacBERT (Masked Language Model as Correction BERT) and BiLSTM (Bidirectional Long Short−Term Memory) to achieve automatic classification of accident causes. Second, based on the combined weights of KeyBERT and TF−IDF (Term Frequency−Inverse Document Frequency), 50 accident causation factors were extracted. Finally, the Apriori algorithm was used to mine association rules among accident causation factors, and social network analysis was conducted to reveal the correlations among these factors. The results show that the proposed text classification model achieves an accuracy of 93.11% in the cause classification task and can accurately identify the categories of accident causes. Among them, defects in the safety management system are the most important hierarchical factors leading to accidents. Inadequate safety education and training, chaotic safety management, inadequate safety supervision, insufficient risk identification ability, weak safety awareness, and rock instability and collapse are strongly correlated and are key causes of accidents. These factors should be the focus of risk control to effectively reduce the occurrence of accidents.