Research on assisting coal mine hazard investigation for accident prevention through text mining and deep learning

Hazard investigation, an effective means of preventing accidents in coal mines, is in a dilemma of “lack of understanding and management”, because the hazard data recorded and saved in text cannot be effectively used. Against this background, this research combines deep learning and text mining technologies to efficiently analyze hazard records for assisting hazard investigation: (1) 3900 historical hazard data from the Fenghuangtai coal mine are used for statistical analysis and text preprocessing; (2) Latent Dirichlet Allocation (LDA) topic analysis is utilized to obtain dangerous topics and feature words; (3) The convolutional neural network (CNN) algorithm is used to classify the danger automatically; (4) The rules between hazard categories and units are visually displayed through images. The findings demonstrate that this paper can effectively classify new text data and identify latent hazard themes that represent reality from coal mine data without relying on expert judgment. The visualization method is appropriate for translating the mining results into specific visual information, which is practical for worker safety education and training and managing hazard investigation. This study provides an innovative perspective on efficiently discovering and utilizing coal mine concealed danger information for accident prevention. •Latent Dirichlet Allocation (LDA) topic analysis is utilized to obtain dangerous topics and feature words.•The convolutional neural network (CNN) algorithm is used to classify the danger automatically without manual annotation.•The network analysis technique is used to examine the relationship between hidden risks and units.