Deep learning-based relation extraction and knowledge graph-based representation of construction safety requirements

Abstract

Field compliance checking aims to check the compliance of site operations with applicable construction safety regulations for detecting violations. Relation extraction provides an automated solution to extract relations that describe construction safety requirements from unstructured text. However, previous relation extraction efforts are limited in their extraction capabilities, representation, and automation. To address this gap, this paper proposes a deep learning-based method to automatically extract and represent relations that describe fall protection requirements. The proposed method: (1) uses a CNN-based model, with pre-trained word and position embeddings, to automatically extract domain-specific relations, and (2) represents the extracted requirements in the form of knowledge graph-based queries, which helps decompose complex requirements into manageable units while keeping these units connected in a scalable graph structure. The proposed method was tested on 20 OSHA sections, and has achieved 87.5% precision, 83.4% recall, and 85.4% F-1 measure, which indicates good relation extraction performance.