Improving Instructional Design Proficiency of Master's Students in Mathematics Education Through Intelligent Educational Technologies

Synthesis: Zhu and colleagues address a gap in Learning Design for discipline-specific smart education by proposing and testing a smart classroom model for Master of Education (M.Ed.) programs, using mathematics instructional design as a case study. Rather than an algorithmic approach, they take a pedagogically oriented route grounded in a three-dimensional framework (learning effectiveness, ICT, and classroom organization) and build the D–T–E Model (Disciplinary Demand–Technological Empowerment–Evaluation Loop). A quasi-experimental study with 68 mathematics education M.Ed. students showed the model significantly improves the precision and professionalism of their instructional objective design, offering a transferable template for Professional Development and Educational Development.

Key Findings

• The D–T–E Model significantly improved instructional design competence. The experimental group (M = 84.15, SD = 6.48) outperformed the control group (M = 80.09, SD = 6.47) on instructional objective design, t(66) = 2.576, p = 0.012, with a medium-to-large effect (Cohen's d = 0.62) and a total score 4.06 points higher.
• The model deepened both "precision" and "professionalism." Gains were significant across all three first-level indicators (curriculum standards, mathematics textbooks, and student learning conditions), with the largest advantages on "competency development" (+4.56) and "integration & alignment" (+4.02), alongside progress in objective-formulation precision (T1, +2.86).
• Multi-AI feedback powered an iterative learning loop. After-class use of three large language models (DeepSeek, Doubao, Kimi) as "intelligent reviewers" — diagnosed across scientific rigor, adaptability, operability, and developmental potential — enabled students to cycle through "design–feedback–reflection–optimization," transforming single-submission assignments into closed-loop practice aligned with Intelligent Tutoring and "Assessment as Learning."
• The smart classroom restructured interaction dynamics. The smart platform extended teacher–student interaction beyond single classroom Q&A into multi-turn, multimedia, and cross-forum exchanges, supporting open resource ecosystems and personalized resource recommendations within AI in Education.
• Learning effectiveness was framed across three levels — knowledge content, research ability, and practical ability — reflecting the training objectives of M.Ed. programs in China, with the model cultivating metacognitive skills and self-improvement methods.
• A pre-class analytics routine surfaced five student difficulty types (task comprehension ambiguity, textbook-content translation, information overload, superficial cognitive processing, and technical inefficiency), enabling stratified grouping and personalized learning paths.

What this means for practice

• Teacher educators. Adapt technologies to a discipline's abstraction, logic, and applicability rather than deploying generic integration, since gains appeared specifically on discipline-grounded objective design.
• Instructional designers. Build a design–feedback–reflection–optimization loop into assignments by using multiple LLMs as "intelligent reviewers," which converted single-submission work into closed-loop practice aligned with Assessment as Learning.
• Instructional designers. Favor continuous, multidimensional, actionable feedback over one-way summative evaluation, and keep the three-model triangulation (DeepSeek, Doubao, Kimi) rather than a single reviewer.
• Teacher educators. Run pre-class analytics to surface student difficulty types and use them to stratify groups and set personalized paths, rather than assuming a uniform entry level.
• Instructors. Extend interaction beyond single classroom Q&A using the smart platform's multi-turn, multimedia, and cross-forum exchanges to sustain open resource ecosystems.

Limitations

• The quasi-experimental comparison rests on N = 68 mathematics education M.Ed. students in one case, and the authors note this disciplinary context constrains direct generalizability to other subject areas without further adaptation.
• The effect size is imprecise: with Cohen's d = 0.62 at p = 0.012, the 95% confidence interval for d ranged from 0.12 to 1.12, which the authors attribute to the small sample.
• The sample is small and strongly concentrated by region and institution, so it may not capture the diversity of mathematics M.Ed. students more broadly.
• The comparison was between a smart-classroom experimental group (M = 84.15, SD = 6.48) and a control group (M = 80.09, SD = 6.47) drawn from the same program, so findings describe instructional-objective design scores rather than transfer to classroom teaching.

Connected Concepts

• Learning Design
• Math Education
• Professional Development
• Educational Development
• Intelligent Tutoring
• Higher Education

Connected Articles

• AI literacy-related domains and AI-TPACK readiness among preservice mathematics teachers: A factor-informed structural equation modelling study — AI literacy domains and AI-TPACK readiness among preservice mathematics teachers
• Preparing Pre-Service Teachers for Responsible Generative AI Use: Curriculum Implications for Ethics, Privacy, and AI Literacy — curriculum implications for preparing pre-service teachers to use generative AI responsibly
• Design of a Science-Integrated Secondary School AI Literacy Curriculum: A Youth & AI Expert Guided Design-Based Research Approach — design-based research on building an AI literacy curriculum
• A Framework for Evaluation of Large Language Models in Essay Assessment: Reliability, Alignment, and Causal Reasoning — a framework for evaluating LLMs in essay assessment (reliability, alignment, causal reasoning)

Citation

Improving instructional design proficiency of master's students in mathematics education through intelligent educational technologies — Zhu, F., Liang, Q., Mao, Z., & Wang, Y. (2026). Computers and Education: Artificial Intelligence, 10, 100597.