Attention-Based Sentiment Region Importance and Relationship Analysis for Image Sentiment Recognition

Table 4

The results of recognition accuracy by using the proposed network, and the comparison with previous work. The results of previous work employ the estimation in related literature. The symbol ‘—’ expresses that there are no appropriate results.

Model	Abstract (%)	IAPSa (%)	Artphoto (%)	TwitterLDL (%)	FlickrLDL (%)	FI (%)
AlexNet [49]	65.49	84.58	69.27	85.10	76.49	72.43
VGGNet16 [50]	72.48	88.51	70.09	90.77	80.53	83.05
PCNN [51]	70.84	88.84	70.96	91.87	82.07	75.34
ResNet50 [52]	73.07	89.95	70.93	91.48	81.61	85.43
ResNet101 [52]	73.36	90.13	71.08	91.68	82.08	85.92
ViT [53]	75.34	89.69	76.52	91.55	83.12	86.33
Zhu [21]	73.88	91.38	75.50	—	—	84.26
Yang [35]	76.03	92.39	74.80	—	—	86.35
Rao [39]	77.77	93.66	77.28	—	—	87.51
Zhang [4]	83.02	95.84	79.24	—	—	90.97
Ours	83.43	94.85	80.62	93.92	85.66	91.52