GC-STCL: A Granger Causality-Based Spatial-Temporal Contrastive Learning Framework for EEG Emotion Recognition.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Electroencephalogram (EEG) signals offer potential for human emotion recognition.
• High noise levels and signal diversity in EEG present challenges like overfitting.
• Extracting meaningful information from EEG for emotion recognition remains difficult.

Purpose of the Study:

• To propose a Granger causal-based spatial-temporal contrastive learning framework to enhance EEG signal information capture.
• To model rich spatial-temporal relationships for improved emotion recognition.
• To address noise and overfitting issues in EEG-based sentiment analysis.

Main Methods:

• Spatial dimension: Sampling strategy for positive pairs, Granger causality test for graph enhancement, and residual graph convolutional neural network for feature extraction.
• Temporal dimension: Frequency domain noise reduction and Granger-Former model for time domain representation.
• Contrastive learning framework incorporating spatial and temporal contrast loss.

Main Results:

• Achieved 1.65% improvement on the DEAP dataset and 1.55% improvement on the SEED dataset compared to state-of-the-art unsupervised models.
• Demonstrated superior prediction accuracy over benchmark methods.
• Showcased enhanced interpretability of the results.

Conclusions:

• The proposed Granger causal-based spatial-temporal contrastive learning framework effectively enhances EEG signal information capture.
• The method significantly improves emotion recognition accuracy and interpretability.
• This approach offers a promising solution for overcoming challenges in EEG-based sentiment analysis.