Enhancing emotion recognition with attention-augmented EEG models

The analysis of human reliability in the context of nuclear power is profoundly influenced by the role of emotions. In this paper, we utilize electroencephalogram (EEG) signals for emotion recognition, employing the SJTU Emotion EEG Dataset (SEED) alongside a Graph-based Multi-task Self-Supervised Learning model (GMSS) enhanced with attention-activated convolutional networks. Experimental validation revealed that the accuracy of the model improved fro m 91.49% before the introduction of the networks to 92.51% afterward, marking an increase of 1.02%. These results demonstrate that the enhanced framework proposed in this paper is capable of effectively recognizing emotions, and the incorporation of attent ion-augmented convolutional networks has the potential to further enhance the model's recognition efficacy.