注意力集中是保证学习者视频有效学习的重要前提，而他们往往难以察觉自身注意状态的变化。虽然学习者在视频学习过程中的注意状态可以通过点击行为、眼动轨迹和脑电信号等方式表征，但是这些单一模态数据在表征时均存在一定的局限性。针对此问题，本项目拟结合心理学实验和混合神经网络技术，探讨视频学习中学习者注意状态表征机制，提出基于多模态数据融合的注意状态表征模型与分类方法。具体内容包括：①研究学习者视频点击行为、眼动轨迹和脑电信号三种模态的表征形式，建立表征视频学习过程中学习者注意状态的框架模型；②采用混合神经网络技术，建构多模态注意状态分类器；③建构面向大规模视频学习的注意状态动态监测的量化方法，并通过教学实践验证。研究成果可用于揭示视频学习中学习者注意状态表征机制，丰富注意状态动态监测的手段，为现代教育科学中学习者的学习状态分析、教学视频质量评价，以及基于视频的有效教学提供理论依据和技术支撑。

Focused attention is a precondition of effective learning from videos. However, learners are hard to detect the changes of their attentional states. Although learners’ attentional states can be detected by clickstream, eye movements, and Electroencephalogram (EEG), the single modal data has limitations to representing their attentional states. To address the above research question, this project aims to investigate the representational mechanism of learners’ attentional states and a classified model to detect the attentional states based on multimodal data by series of psychological experiments and mixed neural networks. It includes three sections: ① It tests the representational mechanism of learners’ attentional states based on their clickstream, eye movements, and EEG signals when learning from videos; ② It tests a classified model of learners’ attentional states based on the multimodal data via mixed neural networks; ③ It builds the detecting approach of attentional states in learning from videos and tests their effectiveness. The results will explain the representational mechanism of learners’ attentional states, extend the dynamic detecting approaches, as well as provide the theoretical bases and technological supports to analyze learners’ learning states, evaluate the quality of videos, and learn effectively.