ECL-tEEG: MÔ HÌNH ƯỚC LƯỢNG CẢM XÚC CỦA NGƯỜI HỌC DỰA TRÊN SÓNG NÃO

Abstract

Emotion recognition and student engagement assessment play an important role in improving teaching efficiency and optimizing personalized teaching methods. Previous studies mainly use videos or images combined with deep learning models to analyze behavior and emotions. However, this method is easily affected by camera angles, lighting conditions, and unclear facial expressions, leading to errors in emotion assessment. To overcome these limitations, this study proposes ECL-tEEG, a model that combines CNN, LSTM-GRU, and Transformer to analyze electroencephalogram (EEG) signals to recognize emotions and assess student engagement. The proposed model utilizes CNN to extract spatial features, LSTM-GRU to learn temporal sequential features, and Transformer to exploit long-term relationships using a parallel learning mechanism. To evaluate the performance, the study conducted experiments on three models: CNN-LSTM-GRU, CNN-Transformer and ECL-tEEG. The experimental results showed that ECL-tEEG achieved 71% accuracy, outperforming the other two models in classifying positive, neutral and negative emotions. This study not only affirms the potential application of EEG in recognizing learners' emotions, but also lays the foundation for an intelligent teaching support system, helping to personalize the learning experience and support teachers in adjusting teaching methods based on brain signals instead of just relying on behavioral observations.