Scholarlink

<p>In the context of smart classrooms, monitoring students’ engagement is crucial for ensuring sustainable motivation, successful attainment of learning outcomes, ultimately, better academic achievement regardless of learning mode or environment. The use of IoT sensors in such contexts becomes a very important source of multi-modal data that can be used in the prediction of students’ engagement levels in real-time, subsequently, the provision of proper interventions. Thus, our study developed a framework for real-time students’ engagement prediction that utilizes camera, audio, physiological and environmental sensors to collect data from 36 female students enrolled in a master program. After combining the data and pre-processing the resultant dataset with additional steps of feature engineering and augmentation using WGAN, we trained a Bi-LSTM with Self-Attention and evaluated it on the augmented dataset and compared the results to four baseline models to output an accuracy per engagement level. The BiLSTM outperformed the baselines with an accuracy of 91.8%. The research reports also on the implementation of this pre-trained model in a Real-Time Student Engagement Monitoring Dashboard that provides a real-time view of individual and class engagement levels and is enhanced with an LLM’s (ChatGPT) generative capabilities to generate faculty targeted feedback that will assist in re-engaging the students.</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

Emotional Engagement Prediction and LLM Feedback in Smart Learning Environments

  • 期刊论文

  • 发表日期：

    2025年7月28日

查看全文
  • Miada Almasre,
  • Wafaa Alsaggaf,
  • Norah Al-Malki,
  • Hani Brdesee

摘要

In the context of smart classrooms, monitoring students’ engagement is crucial for ensuring sustainable motivation, successful attainment of learning outcomes, ultimately, better academic achievement regardless of learning mode or environment. The use of IoT sensors in such contexts becomes a very important source of multi-modal data that can be used in the prediction of students’ engagement levels in real-time, subsequently, the provision of proper interventions. Thus, our study developed a framework for real-time students’ engagement prediction that utilizes camera, audio, physiological and environmental sensors to collect data from 36 female students enrolled in a master program. After combining the data and pre-processing the resultant dataset with additional steps of feature engineering and augmentation using WGAN, we trained a Bi-LSTM with Self-Attention and evaluated it on the augmented dataset and compared the results to four baseline models to output an accuracy per engagement level. The BiLSTM outperformed the baselines with an accuracy of 91.8%. The research reports also on the implementation of this pre-trained model in a Real-Time Student Engagement Monitoring Dashboard that provides a real-time view of individual and class engagement levels and is enhanced with an LLM’s (ChatGPT) generative capabilities to generate faculty targeted feedback that will assist in re-engaging the students.