Graph Convolutional Neural Network and Attention Mechanism based Emotion Classification

Emotion is the subjective cognitive experiences and the corresponding behavioral responses, which plays an important role in interpersonal communication. Emotion classification is an important component of Human Computer Interaction(HCI) that automatically identifies an individual’s emotional state by acquiring physiological or non-physiological signals. Traditional emotion classification methods cannot comprehensively exploit the global and local features contained in EEG signals generated upon stimulation. In this paper, we proposed a graph convolutional neural network and attention mechanism based emotion classification (GCNAEC). First, cross-frequency coupling (CFC) network is leveraged to build adjacency matrices instead of simply combining connections by relative distances and spatial electrode positions of EEG channels. Second, a novel graph attention approach is developed to calculate the attention score of the current graph node and each neighboring node for investigating the differences between single EEG signal and functional connectivity patterns. Third, a novel graph processing method called GAT using PPC is introduced to reduce dimensionality and retain the global information of raw data in each layer. Finally, the experimental results show that the proposed method achieves better performance in emotion classification, in which the classification accuracies are 83.24% for valence dimension and 89.70% for arousal dimension, compared with state-of-the-art methods. In addition, our model outperforms the general GCN model both in terms of recognition accuracy and training time.