Journal Search Engine
The Journal of The Korea Institute of Intelligent Transport Systems Vol.24 No.5 pp.54-76
Fine-tuning과 RAG를 활용한 교통안전 분야 특화 언어모델 구축을 위한 프로세스 제시
A Process for Developing a Traffic Safety Domain Specific Language Model Using Fine-Tuning and RAG
Abstract
The field of traffic safety requires policies to be established and implemented based on clear evidence and real-world cases, which limits the applicability of general-purpose large language models (LLMs). This study investigates the feasibility and applicability of domain-specific LLMs for traffic safety and proposes a systematic development process. To this end, a small language model (SLM) was developed using fine-tuning and retrieval-augmented generation (RAG) techniques to generate countermeasures based on traffic accident types and related information. The model’s responses were qualitatively evaluated through actual improvement project cases. Key considerations derived from this process include the structuring of training data, incorporation of multidimensional contextual information, and establishment of a rigorous evaluation framework. This study provides both academic and practical implications by presenting the direction and foundation for developing traffic safety domain-specific SLMs.
