TFMR-RAG: a knowledge base for traffic facility management regulations

The acceleration of urbanisation has highlighted traffic safety issues, imposing higher demands on the application efficiency and assessment accuracy of traffic facility management regulations. Existing frameworks face challenges such as delayed updates, regional disparities, and regulatory complexity. To address these, a novel Traffic Facility Management Regulation-Retrieval-Augmented Generation (TFMR-RAG) framework is proposed for constructing a specialised knowledge base. This framework integrates large language models with Retrieval-Augmented Generation techniques, leveraging an external corpus of traffic regulations to store and retrieve risk identification standards for road facilities, thereby enhancing contextual understanding and enabling precise regulatory interpretation and recommendations. The TFMR-RAG framework comprises four components: semantic-based document segmentation, content-relevant text chunk embedding, a multi-query mechanism and a hybrid retrieval algorithm. Experimental results demonstrate its significant effectiveness and superiority in building a knowledge base for traffic road facility management regulations.