1. 서론

2. 검색 증강 생성(Retrieval Augmented Generation)

3. 한국어 문법 교정 RAG 구현 방안

4. 한국어 RAG 챗봇의 검토

5. 결론

The present study addresses the limitations of generative AI, including hallucinations and inadequate handling of specialized domain knowledge, which decrease accuracy in Korean grammar correction and grammatical explanations. To overcome these challenges, this study proposes a Retrieval-Augmented Generation (RAG) API integrated with ChatGPT, utilizing a VectorDB of Korean learners’ error cases and a corpus derived from grammar resources provided by the Korean Language Teaching and Learning Center. The effectiveness of the proposed RAG-based Korean grammar correction chatbot was compared to that of a conventional ChatGPT chatbot, which relied solely on prompt-based instructions. The results demonstrated that the RAG-based chatbot effectively provided learners with precise grammatical expressions suitable for their proficiency level through accurate retrieval from databases. Additionally, the chatbot’s inclusion of source URLs for referenced data enhanced credibility. Therefore, the findings confirm that integrating RAG technology can significantly enhance the quality of generative AI outputs, emphasizing its practical utility and potential for advancing Korean language education.