Causation analysis of crane-related accident reports by utilizing ChatGPT and complex networks

This study integrates ChatGPT and complex network (CN) techniques into an accident analysis framework designed to reduce manual effort in accident causation analysis. The proposed framework supports construction stakeholders in extracting causal factors (CFs) from accident reports and identifying both critical CFs and key causal paths. A multistep research design was adopted to develop and validate this novel framework for analyzing crane-related construction accident reports using ChatGPT and CN techniques. First, ChatGPT was prompted to extract CFs from a database of crane-related accident reports. Second, evaluation metrics and an expert questionnaire survey were developed to assess ChatGPT’s performance in CF extraction. Finally, CN analysis was conducted to explore the relationships among CFs and to identify critical causal paths. A total of 95 crane-related accidents from Hong Kong (2011-2020) were analyzed using the proposed framework. The critical CFs identified included: “carelessness”, “operation error”, “crane unbalanced”, “machine failure”, “parts of a crane fall”, “object strike”, “worker fall”, “trapping”, “collapse of crane”, and “load drop”. The critical path identified was: “broken/failed rope” → “load drop” → “object strike”. The primary contribution of this study lies in developing an AI-driven framework that combines the contextual understanding of ChatGPT with the structural analysis capabilities of CN methods—offering a novel and scalable approach to accident causation analysis in the construction industry. Safety managers and practitioners can leverage this framework to improve the automation, consistency, and interpretability of construction accident reporting.