An artificial-intelligence driven method for quantitative assessment of influencing factors in community fire resilience

Community fire resilience is a complex system involving multiple stages and factors, and its enhancement is critical to reducing urban fire risks. This study proposes an intelligent method to assess community fire resilience and identify key factors. Firstly, a community fire resilience concept is proposed as the theoretical foundation. Subsequently, a BERT-BiLSTM-CRF model is trained to automatically extract influencing factors using fire accident reports. A modified technique for order preference by similarity to ideal solution (TOPSIS) method is then employed to screen factors and construct an indicator system. Finally, a Bayesian network (BN) model is constructed to assess resilience and identify key influencing factors. Results indicate that the BERT-BiLSTM-CRF model achieves strong performance with 90.98% F1-score. The indicator system encompasses 23 influencing factors across four resilience stages. BN inference results indicate that Chinese communities exhibit a high-resilience probability of 62%, while recovery and improvement capacities remain relatively weak. Electrical equipment failures, inadequate resident emergency response, underdeveloped emergency medical assistance mechanisms and accident case learning mechanisms are the dominant contributors across the four resilience stages, representing critical vulnerabilities in community fire governance. This study establishes a data-driven assessment framework for community fire resilience, which can be extended to other accident scenarios.