With the rapid advancement of artificial intelligence (AI), generative AI technologies are increasingly applied in education. This study focuses on middle school mathematics, proposing a systematic methodology integrating cognitive quantitative analysis and multi-phase quality assurance. Leveraging the DeepSeek-7B model, we construct a dynamic cognitive load quantification framework to achieve dynamic alignment between question difficulty and student cognitive profiles. An evaluation system spanning three dimensions—question quality, cognitive adaptability, and instructional practicality—is established to systematically analyze the efficacy of generative AI in automatic exercise generation. The methodology effectively validates output quality and offers novel insights for educational technology by optimizing cognitive load alignment. Future research could explore multimodal input optimization and real-time compensatory mechanisms to further enhance generative performance.

Yan Bingpeng, Luo Meifen, Qu Yujing. Generative AI-Based System for Automatic Exercise Generation in Middle School Mathematics: Design and Evaluation Methodology. Frontiers in Educational Research (2025), Vol. 8, Issue 12: 78-85. https://doi.org/10.25236/FER.2025.081212.

[1] HU H, YANG L. AI Empowering the Construction of a Country Strong in Education: Innovation in Artificial Intelligence Teaching and Application[J]. Journal of Teacher Education, 2025, 12(4): 62-70. DOI: 10.13718/j.cnki.jsjy.2025.04.007.

[2] LIN X. Gagné’s Information Processing Learning Theory and Instructional Design [J]. Fujian Tribune, 2010, (S1): 100-101.

[3] YANG X., PAN Y., YAN X. Exploration of Geography Teaching Model Based on Gagné’s Information Processing Learning Theory [J]. Teaching Reference of Middle School Geography, 2023, (09): 17-19+23.

[4] LI X. How to Design Primary School Math Exercises under the New Curriculum Standard [J]. Tianjin Education, 2021, (33): 112-113.

[5] LIU S., HAO X. The Challenges and Approaches of AIGC in Facilitating Educational Innovation [J]. Tsinghua Journal of Education,2024,45(03):1-12.DOI:10.14138/j.1001-4519.2024.03.000112.

[6] XU S. Generative Artificial Intelligence: Development Evolution and Industrial Opportunities [J]. AI-View, 2023, (04): 43-50. DOI: 10.16453/j.2096-5036.2023.04.005.

[7] QI Y., ZHOU H. Technology, System and Ideology: the Evolutionary Logic of Generative Artificial Intelligence in Education [J]. e-Education Research, 2024, 45(08): 28-34. DOI: 10.13811/j.cnki.eer.2024.08.004.

[8] GUO J., FENG J. Research on the Impact of Technological Burden on Primary and Secondary School Teachers under Educational Digital Transformation—Based on Cognitive Load Theory [J]. Education Review, 2025, (07): 103-112.

[9] ZHANG Q. Research on Generative Artificial Intelligence Technology Empowering Primary School Mathematics Education [J]. Intelligence, 2025, (10): 34-37.

[10] GUO G., LIANG D. Should Large Language Models Have “Intrinsic Language Generation Ability”? ——On Chomsky’s Critique of ChatGPT’s Limitations [J]. Studies in Philosophy of Science and Technology, 2025, 42(01): 1-9.

[11] CAI Z., DU G., YIN Y., et al. Research on AI-Assisted Student Writing Norms and Guidance Strategies Based on Educational Ethics [J]. Zhongguancun, 2025, (03): 129-131.

[12] QIAN Y., ZHANG L. Data Cleaning Technology and Its Application Based on Big Data [J]. Digital Technology & Application, 2023, 41(03): 84-86+113. DOI: 10.19695/j.cnki.cn12-1369.2023.03.25.

[13] ZHANG C., XIAO X., GU Y. Review of Data Cleaning Methods [J]. Journal of Beijing Institute of Graphic Communication, 2025, 33(03): 49-55. DOI: 10.19461/j.cnki.1004-8626.2025.03.005.

[14] ZHANG H., ZHANG D., HUANG R. The Development, Application and Reflection of Cognitive Loading Theory in the Intelligent Era——A Review for the 11th International Cognitive Load Theory Conference [J]. Modern Distance Education Research, 2018, (06): 37-44.

[15] LI L. Research on the Effective Application of Auxiliary Lines in Middle School Mathematics [J]. Mathematical and Physical Problem Solving Research, 2023, (08): 11-13.

[16] ZHU Y., GAO Y. Fine-Tuning Methods for General Large Language Models Based on LoRA and Its Variants [J]. Journal of Shanghai University of Electric Power, 2025, 41(01): 90-95.

[17] WEI Y., JIA K., ZENG R., et al. AI Innovation Development and Governance Transformation under the Breakthrough Effect of DeepSeek [J]. E-Government, 2025, (03): 2-39. DOI: 10.16582/j.cnki.dzzw.2025.03.001.

[18] HAO J., NIU H., DU J., et al. Typical Applications of Knowledge Graphs and Large Language Models Empowering Education and Teaching [J]. China Modern Educational Equipment, 2025, (11): 1-5. DOI: 10.13492/j.cnki.cmee.2025.11.003.

[19] CAO J., XIAO J., CAO Y. Automatic Geometric Problem Solution by Integrating Knowledge Points Information [J]. Journal of Chinese Information Processing, 2023, 37(10): 86-96.

[20] CHEN Y., GU J. The Creativity and Imagination of Artificial Intelligence [J]. Journal of Foshan University (Social Science Edition), 2025, 43(01): 26-33. DOI: 10.13797/j.cnki.jfosu.1008-018x.2025.0015.