# Application of Large Language Models in Automated Question Generation: A Case Study on ChatGLM's Structured Questions for National Teacher Certification Exams

**Authors:** Ling He, Yanxin Chen, Xiaoqiang Hu  
**arXiv:** [2408.09982](https://arxiv.org/abs/2408.09982)  
**Date:** 2024-08-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2408.09982v2/

## Key points

- Study by Ling He, Yanxin Chen, and Xiaoqiang Hu explores application potential of large language models (LLMs) like ChatGLM in automated question generation for National Teacher Certification Exams (NTCE)
- ChatGLM generated simulated questions compared with past examinee questions, evaluated by education experts
- Results show high rationality, scientificity, and practicality of ChatGLM-generated questions similar to real exam questions
- Model demonstrates accuracy and reliability in question generation but identified limitations in considering different rating criteria
- Research validates ChatGLM's potential in educational assessment and supports development of more efficient automated generation systems
- Findings contribute to advancing field of automated question generation and improving educational assessment processes

## Summary

The study by Ling He, Yanxin Chen, and Xiaoqiang Hu explores the application potential of large language models (LLMs) like ChatGLM in automated question generation for National Teacher Certification Exams (NTCE). Through careful prompt design, ChatGLM was able to generate a series of simulated questions that were then compared comprehensively with questions from past examinees. To ensure objectivity, experts in education evaluated these questions and their scoring criteria. The results demonstrate that the questions generated by ChatGLM exhibit high levels of rationality, scientificity, and practicality similar to real exam questions across various evaluation criteria. This highlights the accuracy and reliability of the model in question generation. However, limitations were also identified in terms of considering different rating criteria during question generation, indicating the need for further optimization. Overall, this research not only validates ChatGLM's potential in educational assessment but also provides empirical support for developing more efficient and intelligent educational automated generation systems in the future. These findings contribute to advancing the field of automated question generation and improving educational assessment processes.