Scholarlink

<p>Educational Data Mining (EDM) leverages artificial intelligence techniques to optimize learning processes and enhance the effectiveness of Massive Open Online Courses (MOOCs). Among the factors influencing learner behavior, emotions play a central role, as they impact engagement, motivation, and dropout rates. While emotions are traditionally inferred from text and facial expressions, predicting emotional states from navigation behavior remains largely unexplored. This study addresses this gap by predicting learners positive emotional patterns using clickstream data from the <i>Gestion de Projet</i> MOOC. The research design involved a case study with learners enrolled during the first three weeks of the course, where emotional states were measured using the Achievement Emotions Questionnaire (AEQ) and combined with log-trace data capturing weekly navigation activities. Behavioral patterns associated with positive emotions were previously identified through sequential pattern mining. To support the prediction task, we designed a Long Short-Term Memory (LSTM) architecture and evaluated two data representations: (1) a sequential format using a tailored embedding matrix that encodes click order and (2) a time series format capturing weekly temporal dynamics of learners activity. Experimental results demonstrate that the time series-based LSTM significantly outperforms the sequence-based model, achieving an f1-score of 93%. Moreover, augmenting the architecture with a multi-head attention mechanism further improved generalization and reduced prediction errors. These findings highlight the importance of temporal modeling for accurate emotion prediction and provide valuable insights for developing emotionally responsive and adaptive learning environments in MOOCs.</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

LSTM-based prediction of emotional patterns in MOOCs from behavioral data

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • Nihel Fatima Baarir,
  • Samir Bourekkache,
  • Asma Ammari,
  • Ahmed Aloui

摘要

Educational Data Mining (EDM) leverages artificial intelligence techniques to optimize learning processes and enhance the effectiveness of Massive Open Online Courses (MOOCs). Among the factors influencing learner behavior, emotions play a central role, as they impact engagement, motivation, and dropout rates. While emotions are traditionally inferred from text and facial expressions, predicting emotional states from navigation behavior remains largely unexplored. This study addresses this gap by predicting learners positive emotional patterns using clickstream data from the Gestion de Projet MOOC. The research design involved a case study with learners enrolled during the first three weeks of the course, where emotional states were measured using the Achievement Emotions Questionnaire (AEQ) and combined with log-trace data capturing weekly navigation activities. Behavioral patterns associated with positive emotions were previously identified through sequential pattern mining. To support the prediction task, we designed a Long Short-Term Memory (LSTM) architecture and evaluated two data representations: (1) a sequential format using a tailored embedding matrix that encodes click order and (2) a time series format capturing weekly temporal dynamics of learners activity. Experimental results demonstrate that the time series-based LSTM significantly outperforms the sequence-based model, achieving an f1-score of 93%. Moreover, augmenting the architecture with a multi-head attention mechanism further improved generalization and reduced prediction errors. These findings highlight the importance of temporal modeling for accurate emotion prediction and provide valuable insights for developing emotionally responsive and adaptive learning environments in MOOCs.