Research Article
Attention-Based Sentiment Region Importance and Relationship Analysis for Image Sentiment Recognition
Table 3
The result of binary sentiment recognition on the FI dataset.
| Model | Positive | Negative | Accuracy (%) | Precision (%) | Recall (%) | F1 (%) | Precision (%) | Recall (%) | F1 (%) |
| VGGNet16 | 87.11 | 79.69 | 83.23 | 79.41 | 86.34 | 82.73 | 82.98 | ResNet50 | 88.32 | 81.43 | 84.74 | 83.61 | 90.25 | 86.80 | 85.77 | ResNet101 | 89.48 | 80.79 | 84.91 | 84.04 | 90.46 | 87.13 | 86.02 | Ours + important attention | 95.27 | 86.07 | 90.44 | 86.22 | 95.38 | 90.57 | 90.50 | Ours + relation attention | 95.11 | 87.16 | 90.96 | 85.32 | 95.82 | 90.27 | 90.61 | Ours + all | 96.51 | 87.47 | 91.77 | 86.66 | 96.26 | 91.21 | 91.52 |
|
|