Table 6 The influence of number of MSB in MAT.

From: A text guided multimodal scale path fusion network for multimodal sentiment analysis

Nums of MSB

MAE\(\downarrow\)

Corr\(\uparrow\)

ACC-2\(\uparrow\)

F1\(\uparrow\)

ACC-7\(\uparrow\)

Training time\(\uparrow\)

MSB layers = 1

0.586

0.864

88.72

88.68

49.7

497

MSB layers = 2

0.589

0.874

90.02

90.01

50.5

624

MSB layers = 3

0.598

0.882

90.41

90.31

50.1

4519

MSB layers = 4

0.594

0.859

89.63

89.62

49.125

8770

  1. The best results are indicated in bold.