Retrieval-Augmented Generation (RAG) Large Language Model For Educational Chatbot

In the digital age, educational chatbots are increasingly essential for providing personalized learning experience. This study explores the potential of the Retrieval-Augmented Generation (RAG) method to enhanced the performance of educational chatbots. By integrating Large Language Models (LLMs) with real-time data retrieval from external sources, the RAG approach improves the accuracy, relevance, and safety of chatbot responses. Tested using an Indonesian elementary school e-book dataset on Physical Education, Sports, and Health (PJOK), the RAG model demonstrated significantly improvements in generating accurate and contextually appropriate answers while minimizing harmful content. The model achieved a mean context precision of 0.69, context recall of 0.70, faithfulness of 0.60, answer relevancy of 0.86, and harmfulness reduction to 0.26. The findings underscore the effectiveness of RAG advancing educational chatbot performance.