Deep Learning and Text Mining: Classifying and Extracting Key Information from Construction Accident Narratives

Construction accidents can lead to serious consequences. To reduce the occurrence of such and strengthen execution capabilities in on-site safety management, managers must analyze accident report texts depth extract valuable information from them. However, are usually presented unstructured or semi-structured forms; analyzing these manually requires a lot time effort, it is difficult cope with demand large number texts, quality key extracted may be poor. Therefore, this study proposes classification method based on natural language processing (NLP) technology. First, we developed text model convolutional neural network (CNN) that automatically classify categories features. Next, taking classified fall as an example, narratives using term frequency-inverse document frequency (TF-IDF) visually word clouds. The results show overall accuracy CNN reaches 84%, which better than other three shallow machine-learning models. Then, eight areas accident-prone operations were identified TF-IDF algorithm. This provide important guidance for project used management help prevent production accidents.