Cyber Risk Identification in Construction with Language Models: The Next Generation

Abstract: The digitalization of the construction industry enhances efficiency through advanced technologies but increases cyber vulnerabilities. Existing research lacks comprehensive frameworks for risk identification and automation, leaving gaps in addressing cybersecurity challenges. Advances in language models offer potential, but limitations like outdated datasets and small architectures hinder their effectiveness. This study addresses these issues by collecting an up-to-date dataset to fine-tune the GPT-4o Mini model, renowned for its size and reasoning capabilities. The fine-tuned model outperforms others in identifying phase-specific cyber risks, generating a more thorough risk checklist. Its scalability suggests potential applications in broader risk management tasks, enabling industry-wide adoption.