Transformer-GAT Achieves Higher F1 Scores in Dialogue Emotion Recognition

The Transformer-GAT framework combines Transformers for global semantic capture with Graph Attention Networks for modeling fine-grained cross-modal relationships. The model achieved weighted F1 scores of 72.45% on IEMOCAP and 77.37% on MELD.