A knowledge graph for the vulnerability of construction safety system in megaprojects based on accident inversion

The increasing vulnerability of construction safety systems in megaprojects (CSSMs) poses significant challenges to their safety management and control. To address this obstacle, this study retrodicts the accidents based on text mining using the Bidirectional Encoder Repre-sentations from Transformers Topic (BER-Topic) model to uncover topic and topic words related to the vulnerabilities of CSSMs. The vulnerability indicator system (VIS) is established by considering the exposure, sensitivity, and adaptability of the vulnerability of CSSMs. Subsequently, an improved Decision-making Trial and Evaluation Laboratory (DEMATEL) method based on association rules is proposed to reduce the subjectivity in assigning weights to vulnerability indicators, and a topological network based on complex network is constructed to identify the characteristics of VIS. Based on this, a knowledge graph of vulnerabilities in CSSMs is developed. Finally, taking into account the occurrence probability and the actual losses incurred of vulnerability indicators, a vulnerability assessment model for CSSMs is proposed. The research findings are: 1) Based on the BER-Topic model, 32 topics and topic words related to the vulnerability of CSSMs are mined. 2) A VIS for CSSMs is constructed, including 42 indicators across three dimensions of exposure (19), sensitivity (14), and adaptability (9), involving four aspects: humans, machines, environment, and management. 3) The key points for vulnerability management and control in CSSMs are Inaccurate implementation of geological remediation plans, Rusting of connecting components, and Unlicensed personnel operating, among others, which have strong intermediary roles.