스포츠 도메인 특화 검색증강생성(RAG) 개발 및 적용 : 스포츠경영관리사를 중심으로

초록

Purpose: This study aims to develop and validate a sports domain-specific Retrieval-Augmented Generation (RAG) system, SBM-RAG, focusing on the Sports Business Manager certification exam. The goal is to establish a professional and efficient learning environment within the field of sports. Method: The SBM-RAG system utilizes vector-based semantic document embeddings and prompt optimization techniques to provide accurate question-answering, automated problem generation, and solution functionalities aligned with key concepts of the Sports Business Manager examination. Results: Performance evaluation revealed that SBM-RAG achieved a precision@k retrieval accuracy of 76.67% and a 100% correctness rate for the generated questions. In addition, qualitative feedback from domain experts indicated that the generated items were nearly indistinguishable from the actual exam questions, underscoring the system’s high practical relevance and content validity. Conclusion: Although the system demonstrated strong performance, several technical limitations were identified, including constraints related to copyright data and increased computational demands. To address these challenges, this study highlights the need for improved data-licensing protocols and scalable GPU-based infrastructure. Future research should explore the integration of item response theory (IRT) for test equating, which would enhance difficulty calibration and broaden the applicability of RAG-based systems across various peofessional education domains.

키워드