. Author manuscript; available in PMC: 2019 Jul 1.

Published in final edited form as: Proc Conf Assoc Comput Linguist Meet. 2018 Jul;2018:2225–2235.

Table 1:

Comparison of models.

Sentiment Analysis (MOSI)					Emotion Recognition (IEMOCAP)
Approach	Category	WA(%)	UA(%)	Weighted-Fl	Approach	Category	WA(%)	UA(%)	Weighted-Fl
BL-SVM^*	2-class	70.4	70.6	0.668	SVM Trees	4-class	67.4	67.4	-
LSTM-SVM^*	2-class	72.1	72.1	0.674	GSV-e Vector	4-class	63.2	62.3	-
C-MKL₁	2-class	73.6	-	0.752	C-MKL₂	4-class	65.5	65.0	-
TFN	2-class	75.2	-	0.760	H-DMS	5-class	60.4	60.2	0.594
LSTM(A)	2-class	73.5	-	0.703	UL-Fusion^*	4-class	66.5	66.8	0.663
UL-Fusion^*	2-class	72.5	72.5	0.730	DL-Fusion^*	4-class	65.8	65.7	0.665
DL-Fusion^*	2-class	71.8	71.8	0.720	Ours-HF	4-class	70.0	69.7	0.695
Ours-HF	2-class	74.1	74.4	0.744	Ours-VF	4-class	71.8	71.8	0.713
Ours-VF	2-class	75.3	75.3	0.755	Ours-FAF	4-class	72.7	72.7	0.726
Ours-FAF	2-class	76.4	76.5	0.768	Ours-FAF	5-class	64.6	63.4	0.644

WA = weighted accuracy. UA = unweighted accuracy.

denotes that we duplicated the method from cited research with the corresponding dataset in our experiment.