Optimizing automated compliance checking with ontology-enhanced natural language processing: Case in the fire safety domain

Abstract

The fire safety compliance checking (FSCC) plays a crucial role in ensuring the quality of fire engineering design and eliminating inherent fire hazards. It requires an objective and rational interpretation of fire regulations. However, the texts of fire regulations are filled with numerous rules related to spatial limitations, which pose a significant challenge in interpreting them. The current method of interpreting these rules mostly relies on manual translation, which is not efficient. To address this issue, this study proposes an innovative automated framework for interpreting rules by combining ontology technology with natural language processing (NLP). Through the utilization of pre-trained language models (PLMs), concepts and relationships are extracted from sentences, a domain-specific ontology is established, spatial knowledge is transformed into language-agnostic tree structures based on the ontology, and the semantic components of spatial relationships are extracted. The tree structure is then mapped to logical clauses based on semantic consistency, thereby improving the efficiency of interpretation. Experimental results demonstrate that the architecture achieves an F1 score of 86.27 for entity extraction and 81.81 for spatial relationship joint extraction tasks, with an accuracy of 96.26% in the formalization of logical rules, highlighting its proficiency in automatically interpreting fire spatial rules. This study offers technical support to enhance public understanding of fire safety management and fire prevention predictions, thereby promoting the intelligent management of the building safety environment.