Semi-supervised Graph Convolutional Network for Emotion Recognition using Multimodal Biosignals Fusion
Oral presentation
Research approach
F-MSGCN builds similarity graphs for EEG and eye features, then fuses relationships within and across modalities. A GCN classifier learns from the fused graph with limited labels.
View full-size figure
Results
Subject-wise transductive evaluation on SEED-IV: 15 participants and four emotions. With 120 labels per class, accuracy was 99.71 ± 0.31%, compared with 77.18 ± 8.78% for EGFG.
| Model | Accuracy |
|---|---|
| EGFG | 77.18 ± 8.78 |
| F-MSGCN | 99.71 ± 0.31 |