A Knowledge Graph for Automated Construction Workers' Safety Violation Identification

Identifying workers' safety violations on construction job sites is critical for improving construction safety performance. The advancement of sensing technologies makes automatic safety violation detection possible by encoding the safety knowledge into computer programs. However, it requires intensive human efforts in turning safety knowledge into computer rules, and the hard-coded rules limit the expandability of the developed applications. This study proposes a condition-based knowledge graph for the safety knowledge representation to support the reasoning on safety violations. The improved knowledge graph's structure solves the limitation by presenting the public knowledge and safety rules for condition structure, respectively. A natural language processing supported automatic knowledge graph development approach is developed in this paper to extract the safety knowledge from safety knowledge texts automatically and to construct the knowledge graph. To validate this construction framework, an initial knowledge graph containing 1,200 rules is developed based on construction safety regulations. The proposed automatic safety knowledge extraction model achieves an F1 value of 67%.