Multimodal Sentiment Analysis Based on Interactive Transformer and Soft Mapping

Table 1

2-class-sentiment of CMU-MOSEI dataset.

Model	2-class-sentiment
	Accuracy (%)	F1-score (%)
Graph-MFN (T + A + V) [14]	76.90	77.00
B2 + B4 w/multimodal fusion (T + A + V) [16]	81.14	78.53
Multilogue-Net (T + A) [18]	80.18	79.88
Multilogue-Net (T + A + V) [18]	82.10	80.01
TBJE (T + A) [19]	82.30	/
TBJE (T + A + V) [19]	81.50	/
The proposed approach (T + A)	82.47	81.23