Deep Learning-Based Relation Extraction from Construction Safety Regulations for Automated Field Compliance Checking

Information extraction provides an opportunity to automatically extract safety requirements from construction safety regulations to support automated safety compliance checking for detecting field non-compliances with these regulations. However, previous efforts on automating the safety compliance checking process fall short in their scalability and ability to automatically extract safety requirements, due to the complexity in unstructured text. Therefore, this paper proposes a deep learning-based information extraction method for extracting relations that link fall protection-related entities extracted from construction safety regulations for supporting automated field compliance checking. The proposed method uses an attention-based convolutional neural network model for recognizing and classifying relations. The proposed method was implemented and tested on two selected Occupational Safety and Health Administration (OSHA) sections related to fall protection. It has achieved a weighted precision, recall, and F-1 measure of 82.7%, 81.1%, and 81.3%, respectively, which indicates good relation extraction performance.