Scholarlink

<p>Attributing student engagement and involvement in ideological and political education (IPE) to learning value-oriented outcomes is an utmost importance. However, traditional classroom observations do not pick up on subtle signals of behavior-under-the-radar classes-student signs of being tired, inattentive, or passive. Analyzing behavior of a big bunch of students or online events might become impossible with observation interference such as engagement. Therefore, a multi-modal, interpretable, and domain-adaptive system for real-time behavior recognition and fatigue detection in IPE class has been put forward. In short, the system allows us to observe facial dynamics (including eyeblinks via eye aspect ratio, yawns via mouth aspect ratio, and the PERCLOS metric), skeletal posture features (various distances and joint angles), and apply temporal modelling with Temporal Shift Modules (TSM) to grab cues of cognitive and physical engagement. Meta-learning framework (MAML) allows fast domain adaptation to new classroom environments with only a few labelled data, thereby increasing generalizability. Experimental results of various classroom scenes show the system is capable of generating very accurate classification of behavior such as asking, looking, and boredom (F1-score ≥ 0.90), fatigue detection with up to 94.5% accuracy; the system also quantifies Q&amp;A participation through an XP model and uncovers inequalities in student engagement, showing substantial agreement with teacher ratings (Cohen’s κ = 0.81). Transparency to models is ensured by visual explanations through heatmaps and time-series plots, thus enabling the ethical deployment of the system in the educational environment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Domain-adaptive multi-modal deep learning for monitoring student fatigue and engagement in remote ideological and political education

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月28日

查看全文
  • Zeng Wang,
  • Xu Jiang,
  • Pingping Long

摘要

Attributing student engagement and involvement in ideological and political education (IPE) to learning value-oriented outcomes is an utmost importance. However, traditional classroom observations do not pick up on subtle signals of behavior-under-the-radar classes-student signs of being tired, inattentive, or passive. Analyzing behavior of a big bunch of students or online events might become impossible with observation interference such as engagement. Therefore, a multi-modal, interpretable, and domain-adaptive system for real-time behavior recognition and fatigue detection in IPE class has been put forward. In short, the system allows us to observe facial dynamics (including eyeblinks via eye aspect ratio, yawns via mouth aspect ratio, and the PERCLOS metric), skeletal posture features (various distances and joint angles), and apply temporal modelling with Temporal Shift Modules (TSM) to grab cues of cognitive and physical engagement. Meta-learning framework (MAML) allows fast domain adaptation to new classroom environments with only a few labelled data, thereby increasing generalizability. Experimental results of various classroom scenes show the system is capable of generating very accurate classification of behavior such as asking, looking, and boredom (F1-score ≥ 0.90), fatigue detection with up to 94.5% accuracy; the system also quantifies Q&A participation through an XP model and uncovers inequalities in student engagement, showing substantial agreement with teacher ratings (Cohen’s κ = 0.81). Transparency to models is ensured by visual explanations through heatmaps and time-series plots, thus enabling the ethical deployment of the system in the educational environment.