A hybrid graph attention network with multi-dimensional features for enhanced EEG-based emotion recognition.

Area of Science:

• Affective computing
• Neuroscience
• Machine learning

Background:

• Emotion recognition from electroencephalogram (EEG) signals is crucial for human-computer interaction.
• Existing methods often treat EEG signals as 1D time series, neglecting multidimensional structures and segment dynamics.
• There is a need for advanced models to capture complex relationships within EEG data for improved emotion recognition.

Purpose of the Study:

• To propose a novel Hybrid Graph Attention Network (H-GAT) model for enhanced emotion recognition using EEG signals.
• To effectively model multidimensional dependencies and dynamic segment relationships in EEG data.
• To improve the accuracy and interpretability of emotion recognition systems.

Main Methods:

• Developed a Hybrid Graph Attention Network (H-GAT) integrating Graph Attention Networks (GAT), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM).
• Transformed EEG signals into a graph representation using dynamic similarity measures (Pearson correlation) for the GAT module.
• Employed CNN for local temporal pattern extraction and LSTM for overall temporal dynamics, followed by feature fusion and classification.

Main Results:

• Achieved superior performance on SEED and DEAP datasets.
• Attained an average accuracy of 97.89% (binary) and 96.67% (ternary) on the SEED dataset.
• Reached an average accuracy of 93.09% for valence and 93.93% for arousal on the DEAP dataset.

Conclusions:

• The H-GAT model demonstrates significant potential for advancing emotion recognition through hybrid neural architectures.
• The model effectively captures multidimensional dependencies and temporal dynamics in EEG signals.
• The proposed approach offers a robust and interpretable solution for emotion recognition in affective computing.