Abstract: To improve the accuracy and efficiency of traditional fire risk assessment methods for commercial premises, this paper proposes a fire risk assessment method based on the combination of the BERT algorithm and Bayesian Network (BN). By analyzing fire accident reports and expert assessment reports, a fire risk indicator system applicable to various types of commercial premises is established. The BERT algorithm is utilized to construct an automatic indicator generation model, enabling the automatic generation of fire risk indicators that match the characteristics of commercial premises. Based on the generated indicators, the Bayesian Network is employed to calculate the fire risk of commercial premises. Taking a large commercial complex as a case study, the matching indicators are obtained through the automatic generation model, and the Bayesian Network is used to calculate the probability distribution of each indicator node and the overall fire risk, thereby verifying the feasibility of the proposed method.

Key words: commercial buildings, fire risk, BERT algorithm, Bayesian Networks, indicator system