Quiz generation using RAG and Self-Refine

Abstract

In our study, we propose an automated high-quality quiz generation system by integrating the use of Large Language Models (LLM) with RAG (Retrieval-Augmented Generation) and Self-Refine mechanisms. RAG collects relevant information from external databases such as Wikipedia and utilizes this as prompts for the GPT model. On the other hand, the Self-Refine mechanism generates feedback using LLM for the initially produced quizzes and iteratively refines the quizzes based on this feedback. The primary goal of this system is to efficiently generate quizzes with high accuracy and relevance, and the results have been validated through evaluations by people. Experimental results have shown that the Self-Refine mechanism significantly improves the initial quiz generation process, enabling the creation of high-quality quizzes.