国际教育成就评价协会（IEA）：2023年使用大语言模型进行自动题库生成：TIMSS四年级的发展与验证报告

The study aimed to validate the quality of assessment items generated by Large Language Models for use in mathematics and science assessment on the example of the TIMSS Grade 4. The validation process included expert ratings and qualitative assessment, as well as using the generated items in a field test that enabled the assessment of the substantial and psychometric properties of the generated items. Since LLM-generated items were mixed with original TIMSS items, we were able to compare the