Graph Recurrent Neural Network for Emotion Recognition

• 1、Computer and Information Technology School, Beijing JiaoTong University, Beijing 100093

Abstract：Nowadays, Emotion Recognition has been used in Medical and Public Safety field. In this paper, we propose a graph recurrent neural network model based on the temporal and spatial features of EEG signals, using the related theory of graph recurrent neural network. We first use a 1-second Hamming window to divide the EEG signals into samples; after that we calculate the mutual information for the data of the different channels in each sample; next, we combines the mutual information value and the data of different channels to form a graph sequence; finally, the graph sequence is used as the input of the model. Besides, the accuracy and other indicators of the model are measured based on arousal and valance. The experimental results on the public DEAP dataset show that our model is superior to the model that only uses the spatial correlations of EEG signals.

Keywords： EEG, Graph Recurrent Neural Network, Emotion Recognition, Mutual Information