Research Article
Multimodal Fusion Method Based on Self-Attention Mechanism
Table 1
Results for emotion recognition on IEMPCAP, personality trait recognition on POM, and sentiment analysis on CMU-MOSI.
| Data set | IEMOCAP | POM | CMU-MOSI | Metric | F1-happy | F1-sad | F1-angry | F1-neutral | MAE | Corr | Acc | MAE | Corr | Acc-2 | F1 | Acc-7 |
| SVM | 81.5 | 78.8 | 82.4 | 64.9 | 0.887 | 10.4 | 33.9 | 1.864 | 0.057 | 50.2 | 50.1 | 17.5 | DF | 81.0 | 81.2 | 65.4 | 44.0 | 0.869 | 14.4 | 34.1 | 1.143 | 0.518 | 72.3 | 72.1 | 26.8 | BC-LSTM | 81.7 | 81.7 | 84.2 | 64.1 | 0.840 | 27.8 | 34.8 | 1.079 | 0.581 | 73.9 | 73.9 | 28.7 | MV-LSTM | 81.3 | 74.0 | 84.3 | 66.7 | 0.891 | 27.0 | 34.6 | 1.019 | 0.601 | 73.9 | 74.0 | 33.2 | LMF | 85.2 | 85.8 | 87.4 | 71.7 | 0.837 | 32.3 | 36.8 | 1.071 | 0.571 | 73.3 | 73.3 | 30.0 | HPFN | 85.7 | 86.2 | 87.8 | 71.9 | 0.840 | 35.6 | 36.7 | 0.975 | 0.601 | 73.0 | 73.1 | 35.1 | Our methods | 85.7 | 86.3 | 87.6 | 71.9 | 0.834 | 52.3 | 36.7 | 1.032 | 0.610 | 74.1 | 73.0 | 30.5 |
|
|
Best results are italicized.
|