【摘要】 教育智能体作为连接大模型与教学场景的重要桥梁，已成为推动智能化、个性化与精准化教学的关键工具。已有研究从不同角度探讨了教育智能体的应用效果，但其对学生学习表现的影响及内在机制仍有待明确。基于对87篇实证文献的元分析结果表明：教育智能体整体上对学生学习表现具有中等偏小的正向促进作用，在认知能力层面的数字素养维度促进效果最强；学段与学科对学生学习表现无显著调节作用，但教育智能体是否集成GAI、角色设定、社会特征及应用形式对其具有显著调节效应。进一步分析发现，具备集成GAI、扮演专家角色、强调交互风格等特征的教育智能体对于提升学生的学习表现更具成效；具有“师—生—机”协同、知识生成导向教学模式等应用形式的教育智能体亦能显著促进学生的学习表现。为提高教育智能体应用效果，未来可基于最近发展区构建认知支架，强化协同支持以促进知识生成；融合多媒体认知学习理论，设计多元教学模式组合；警惕“快餐式知识”与幻觉输出，结合生成学习策略协同引导深度学习的发生。更多还原

【Abstract】 As a vital bridge between large language models and teaching scenarios, educational agents have emerged as key tools for advancing intelligent, personalized, and precise teaching. Although previous studies have examined their effects from various perspectives, the overall impact of educational agents on student learning outcomes and the underlying mechanisms remain unclear. This meta-analysis of 87 empirical studies reveals that educational agents have a small to moderate positive effect on student learning outcomes, with the strongest impact observed on the digital literacy dimension of cognitive abilities. While neither educational level nor subject area significantly moderates the effect, the integration of GAI, role design, social features and application forms show significant moderating effects. Further analysis indicates that educational agents integrated with GAI, adopting expert roles, and emphasizing interactive styles are more effective in enhancing student performance. Application forms featuring teacher-student-agent collaboration and knowledge-generative instructional models also significantly promote learning outcomes. To further improve the application effectiveness of educational agents, it is necessary to build cognitive scaffolds aligned with the zone of proximal development to strengthen collaborative support for knowledge construction, incorporate multimedia learning theory to design diverse instructional strategies, and avoid“fast-food knowledge”and hallucinated outputs by integrating generative learning strategies to guide deep learning processes.更多还原