Multimodal Sentiment Analysis Based on Cross-Modal Attention and Gated Cyclic Hierarchical Fusion Networks

Table 4

Performance tables for different cross-modal notes on MOSI and MOSEI datasets.

Task	MOSI				MOSEI
	MAE (↓)	Corr (↑)	Acc-2 (↑)	F1-score (↑)	MAE (↓)	Corr (↑)	Acc-2 (↑)	F1-score (↑)
	1.442	0.210	53.81	46.48/	1.315	0.197	60.13/61.25	60.48/59.38
	1.321	0.233	62.33	57.85/58.94	1.244	0.182	58.48/59.47	61.63/61.48
	0.896	0.393	68.52	64.44/65.38	0.815	0.213	64.15/63.18	64.85/64.25
	0.901	0.384	71.49	69.12/67.20	0.843	0.241	63.48/63.14	63.54/63.89
	0.976	0.223	73.62/71.40	71.04/64.31	0.821	0.213	61.84/62.37	61.23/61.66
	0.957	0.381	74.22/72.67	71.04/65.86	0.784	0.230	63.24/62.56	61.05/60.72
	0.819	0.486	76.80/76.01	75.72/74.84	0.763	0.361	72.18/72.56	74.37/74.03