Graph-based intelligent accident hazard ontology using natural language processing for tracking, prediction, and learning

Abstract

This paper addresses the challenge of dispersed accident-related information on construction sites, which hinders consensus among employers, workers, supervisors, and society. A robust NLP-based framework is presented to analyze and structure accident-related textual data into a comprehensive knowledge base that reveals accident patterns and risk information. Accident scenarios, including frequency and severity scores, are structured into a graph database through knowledge modeling, establishing an ontology to elucidate keyword relationships. Network analysis identifies accident patterns, quantifies scenario likelihood and severity, and predicts criticality, forming an accident hazard ontology. This vectorized ontology supports accident tracking, prediction, and learning with potential applications. The framework ensures reliable data integration, real-time hazard assessment, and proactive safety measures.