Research on Question Bank Construction Based on Retrieval-Augmented Generation and Large Language Models

The construction of high-quality question banks has always been a complex problem. To enhance the efficiency and quality of question bank construction and address challenges such as the limited number of question bank samples and the slow update speed in traditional methods, this paper proposes a new technique that combines Retrieval Enhancement Generation (RAG) technology with Large Language Models (LLMs). Firstly, a system framework combining RAG and LLMs is designed. Retrieving external knowledge enhances the accuracy and relevance of question generation bases. The natural language processing capability is used to generate various types of questions. The data set uses textbooks, academic literature, and other data sources. After preprocessing, the data is input into the system to generate questions. The experimental results show that the problems generated by this method achieve an accuracy rate of 85% to 90%, covering various disciplines such as mathematics and physics. Meanwhile, the automatic optimization strategy reduces the need for manual intervention. This paper shows that integrating RAG and LLMs effectively solves the problems, such as the small number of samples and slow update speed of traditional question banks, providing an up-and-coming solution for the intelligent construction of educational resources. Future work will optimize the model's knowledge retrieval scope and the problem generation's difficulty-adaptive mechanism.