MHNNSL: EEG-Based Emotion Recognized by Multi-task Hypergraph Neural Network combined with Self-supervised Learning

In recent years, the use of electroencephalography (EEG) for emotion recognition has attracted widespread attention in the fields of human-computer interaction and cognitive science. However, how to establish a brain network model to simulate the dynamic processes of the brain during various cognitive activities, thereby revealing the complex interrelationships between brain regions and completing emotion recognition tasks has become a focus of attention. This paper proposes an EEG emotion recognition model based on Multi-task Hypergraph Neural Network combined with Self-supervised Learning (MHNNSL). Firstly, we perform a hypergraph modeling of the brain network according to the structure of brain regions. Then we use a hypergraph neural network to learn the corresponding features. Moreover, by integrating multiple self-supervised tasks (including spatial transformation task, frequency transformation task, and contrastive learning task), more general representations are learned. Finally, emotion recognition is achieved through a multi-layer perceptron (MLP). To verify the effectiveness of the proposed method, comparative experiments were conducted on the publicly available SEED dataset with existing methods. Our proposed model achieved the highest accuracy of 85.17% and successfully reflected the response level of various brain regions to emotional tasks. Meanwhile, ablation experiments have demonstrated the effectiveness of our proposed model.