A Graph Neural Network with Spatial Attention for Emotion Analysis

Emotion recognition plays a crucial role in the diagnosis and treatment of various mental disorders. Research studies revealed the close relationship between brain regions and their functional roles in emotions. Propose a learning method that extends graph neural networks and takes into account the spatial relationship between EEG channels and their contributions of different regions of the brain to human emotions. Our method uses the adjacency matrix to model the spatial topological relationships in multi-channel EEG signals and learns weights to adjust their contributions to the classification. Extensive evaluation is conducted using public data sets, including comparison studies with state-of-the-art methods and performance analysis. In our comparison studies, our method demonstrates superior performance in terms of average accuracy. It is demonstrated that the proposed method improves the accuracy of emotion recognition and analyzes the brain at a fine granularity to decide the part that is most related to the triggering of the emotion.