TY  - JOUR
AU  - YE, Xin-Dong
AU  - REN, Ji-Jian
AU  - XU, Xuan-Wu
PY  - 2026
TI  - Educational Virtual Samples Dirven by Language Models: Theoretical Logic, Operational Models, and Methodological Value
JO  - Modern Educational Technology
SN  - 1009-8097
SP  - 25
EP  - 35
VL  - 36
IS  - 7
AB  - With the advancement of large language models’ capabilities in generation, reasoning, and behavioral simulation, the constraints that have long hindered traditional education empirical research centered on real individuals, including difficulty in sample acquisition, high experimental costs, and significant ethical risks, are encountering new breakthroughs. As quasi-real, annotatable, and controllable synthetic samples, educational virtual samples driven by large language models are emerging as a new methodological for empowering educational research. Building upon a review of the research trajectory of virtual samples, this paper systematically analyzed the theoretical logic of virtual samples in education by integrating the perspective of human-machine symbiosis, actor-network theory, and information processing theory. It further constructed a closed-loop pathway encompassing five stages of setting, memory, planning, action, and regulation-feedback, and explored its boundary expansion in survey research and experimental research. The study concluded that educational virtual samples can effectively simulate learners’ cognitive responses and behavioral processes, support pre-verification of research, sample supplementation, and scenario extapolation. However, their application should adhere to the principles of authenticity, ethics, and boundary, so as to avoid substituting technological verisimilitude for educational reality.
UR  - https://doi.org/10.3969/j.issn.1009-8097.2026.07.003
DO  - 10.3969/j.issn.1009-8097.2026.07.003