[AML] AI-empowered Intelligent Education: Question Generation based on LLMs
Keywords: AI-empowered teaching assistant, personalized learning, question generation, retrieval-augmented generation, supervised fine-tuning
Abstract: College students often struggle with grasping complex materials in courses like physics and engineering due to limited access to personalized practice. Current AI-based question generation systems cover shallow knowledge and superficial formats, limiting their effectiveness as learning tools. In this work, we propose an AI-powered teaching assistant based on ChatGLM, which leverage Retrieval-Augmented Generation (RAG) and Supervised Fine-Tuning (SFT) to automatically generate specialized exercise questions in University Physics and Chemical Engineering Thermodynamics, fulfilling personalized requirements and effectively assisting college students with their study. Utilizing a multi-dimensional evaluation framework, our results show that multi-level RAG achieves the best performance, significantly improving question relevance and quality over the baseline. SFT with reflection also enhances question quality but remains inferior to RAG. These findings demonstrate the practical value of our approach while highlighting the need for improved reasoning capabilities to generate more complex and challenging questions.
Submission Number: 28
Loading