用大型语言模型模拟学生：生成式AI在教育中的架构、机制与角色建模综述
Simulating Students with Large Language Models: A Review of Architecture, Mechanisms, and Role Modelling in Education with Generative AI
AI总结：
本文综述了利用大型语言模型模拟学生行为在教育中的应用，探讨了其在学习者建模、教学评估和教师培训中的潜力与挑战。
摘要：
模拟学生为评估教学方法和建模多样化学习者提供了有价值的框架，但现实环境中系统性地进行这些任务具有挑战性。近期研究日益关注开发此类模拟代理，以捕捉多种学习风格、认知发展路径和社会行为。在当代模拟技术中，将大型语言模型（LLMs）整合到教育研究中已成为特别灵活且可扩展的范式。LLMs提供了高度的语言真实性和行为适应性，使代理能够近似认知过程并参与上下文适当的教学对话。本文对利用LLMs模拟学生行为的实证和方法学研究进行了主题综述。我们综合了当前关于LLM代理模拟学习者原型、响应教学输入以及在多代理课堂场景中互动的证据。此外，我们还探讨了此类系统对课程开发、教学评估和教师培训的影响。尽管LLMs在自然语言生成和情境灵活性上优于基于规则的系统，但仍然存在关于算法偏见、评估可靠性和与教育目标的一致性等持续关注的问题。综述指出了现有技术和方法学的空白，并提出了未来研究方向，以将生成式AI整合到自适应学习系统和教学设计中。
英文摘要：
Simulated Students offer a valuable methodological framework for evaluating pedagogical approaches and modelling diverse learner profiles, tasks which are otherwise challenging to undertake systematically in real-world settings. Recent research has increasingly focused on developing such simulated agents to capture a range of learning styles, cognitive development pathways, and social behaviours. Among contemporary simulation techniques, the integration of large language models (LLMs) into educational research has emerged as a particularly versatile and scalable paradigm. LLMs afford a high degree of linguistic realism and behavioural adaptability, enabling agents to approximate cognitive processes and engage in contextually appropriate pedagogical dialogues. This paper presents a thematic review of empirical and methodological studies utilising LLMs to simulate student behaviour across educational environments. We synthesise current evidence on the capacity of LLM-based agents to emulate learner archetypes, respond to instructional inputs, and interact within multi-agent classroom scenarios. Furthermore, we examine the implications of such systems for curriculum development, instructional evaluation, and teacher training. While LLMs surpass rule-based systems in natural language generation and situational flexibility, ongoing concerns persist regarding algorithmic bias, evaluation reliability, and alignment with educational objectives. The review identifies existing technological and methodological gaps and proposes future research directions for integrating generative AI into adaptive learning systems and instructional design.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
