Scholarlink

<p>The rapid expansion of e-learning platforms has increased the demand for accurate and robust emotion recognition systems that can dynamically adapt to learners’ affective states. Existing emotion recognition methods predominantly rely on unimodal physiological signals or simplistic fusion strategies, resulting in limited generalizability due to minimal correlation between features and labels. These approaches often suffer from poor interpretability and inadequate handling of temporal variations in emotional intensity, which lead to unstable and inaccurate recognition in dynamic real-world learning environments. To overcome these challenges, this study proposes a novel multimodal physiological emotion recognition framework integrating Electroencephalogram (EEG), Electrocardiogram (ECG), and Electrodermal Activity (EDA) signals through an advanced Bayes-optimal cue integration model. The framework employs Graph Convolutional Networks (GCNs) to capture complex spatial dependencies inherent in physiological signals’ non-Euclidean structure, combined with Spatio-Temporal Long Short-Term Memory (ST-LSTM) networks to model temporal dynamics robustly. Synchronization of physiological signals at the beat level ensures precise temporal alignment, further enhancing reliability. The final classification employs a Deep Belief Network (DBN) optimized via hierarchical layer-wise training and fine-tuning. Experimental validation on benchmark datasets including SenseCobot, AMIGOS, and ASCERTAIN demonstrates state-of-the-art performance with an accuracy of 99.5%, significantly surpassing existing models.</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

E-learning emotion recognition by fusing ECG, EEG and EDA signals based on Bayes optimal cue integration model

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Sherimon Puliprathu Cherian

摘要

The rapid expansion of e-learning platforms has increased the demand for accurate and robust emotion recognition systems that can dynamically adapt to learners’ affective states. Existing emotion recognition methods predominantly rely on unimodal physiological signals or simplistic fusion strategies, resulting in limited generalizability due to minimal correlation between features and labels. These approaches often suffer from poor interpretability and inadequate handling of temporal variations in emotional intensity, which lead to unstable and inaccurate recognition in dynamic real-world learning environments. To overcome these challenges, this study proposes a novel multimodal physiological emotion recognition framework integrating Electroencephalogram (EEG), Electrocardiogram (ECG), and Electrodermal Activity (EDA) signals through an advanced Bayes-optimal cue integration model. The framework employs Graph Convolutional Networks (GCNs) to capture complex spatial dependencies inherent in physiological signals’ non-Euclidean structure, combined with Spatio-Temporal Long Short-Term Memory (ST-LSTM) networks to model temporal dynamics robustly. Synchronization of physiological signals at the beat level ensures precise temporal alignment, further enhancing reliability. The final classification employs a Deep Belief Network (DBN) optimized via hierarchical layer-wise training and fine-tuning. Experimental validation on benchmark datasets including SenseCobot, AMIGOS, and ASCERTAIN demonstrates state-of-the-art performance with an accuracy of 99.5%, significantly surpassing existing models.