Cause analysis of hot work accidents based on text mining and deep learning

Abstract

Hot work accidents have significant consequences. Admittedly, preventing hot work accidents requires managers to analyze the accident profoundly and learn from the requisite documents that contain the detailed process and causes of the accident. However, the analysis process of unstructured records is manual, leading managers to failing to quickly analyze the cause of the accident and identify the key causes. Therefore, we used deep learning to automatically classify and predict the cause of accidents by the textual features and text mining methods that quickly extract the cause information to optimize accident record analysis. Initially, the latent Dirichlet allocation model was adopted to extract the topic words in the accidents to form cause topics, and convolutional neural networks were trained to predict the cause of the accident based on the previous cause topics. Then, the key causes of hot work accidents were extracted by qualifying the importance score of the cause, which included no gas detection and continuous monitoring, uncleaned combustibles, improper protection measures, and violations of regulations by workers. Managers can utilize these key causes to formulate optimal safety strategies to reduce the number of accidents. This study provides valuable suggestions for improving the process safety management of hot work in China. Moreover, the results of our method can be used as important documents for the safety education and training of the enterprise.