. 2022 Dec 19;16:1086380. doi: 10.3389/fnins.2022.1086380

Table 2.

Binary classification results of each emotion category on IEMOCAP and CMU-MOSEI datasets.

	IEMOCAP
Models	Happiness		Anger		Sadness		Excited		Frustrated		Neutral
	Acc	F ₁	Acc	F ₁	Acc	F ₁	Acc	F ₁	Acc	F ₁	Acc	F ₁
LF-LSTM^†	0.672	0.376	0.712	0.494	0.782	0.540	0.793	0.572	0.682	0.515	0.665	0.470
LF-TRANS^†	0.852	0.376	0.819	0.507	0.874	0.574	0.853	0.573	0.605	0.493	0.724	0.497
EmoEmbs (Dai et al., 2020)^†	0.696	0.383	0.659	0.489	0.808	0.530	0.735	0.583	0.685	0.520	0.736	0.487
MulT (Tsai et al., 2019)^†	0.800	0.468	0.779	0.607	0.835	0.654	0.769	0.580	0.724	0.570	0.749	0.537
BIMHA (Wu et al., 2022)^††	0.834	0.432	0.772	0.576	0.838	0.637	0.783	0.561	0.739	0.542	0.764	0.509
CMHA (Zheng et al., 2022)^††	0.890	0.458	0.886	0.611	0.883	0.616	0.879	0.605	0.751	0.563	0.765	0.512
MESM (Dai et al., 2021)^†	0.895	0.473	0.882	0.628	0.886	0.622	0.883	0.612	0.749	0.584	0.770	0.520
FE2E (Dai et al., 2021)^†	0.900	0.448	0.887	0.639	0.891	0.657	0.891	0.619	0.712	0.578	0.791	0.584
MER-SEM-MBT (Our full model)	0.891	0.577	0.894	0.665	0.924	0.721	0.905	0.677	0.797	0.613	0.832	0.623
MER-SEM-MBT (Ours w/o textual decision)	0.889	0.546	0.893	0.662	0.918	0.701	0.892	0.643	0.794	0.602	0.827	0.613
	CMU-MOSEI
Models	Happiness		Sadness		Anger		Surprise		Fear		Disgust
	WA	F ₁	WA	F ₁	WA	F ₁	WA	F ₁	WA	F ₁	WA	F ₁
LF-LSTM^†	0.613	0.732	0.634	0.472	0.645	0.471	0.571	0.206	0.617	0.222	0.705	0.498
LF-TRANS^†	0.606	0.729	0.601	0.455	0.653	0.477	0.621	0.242	0.621	0.240	0.744	0.519
EmoEmbs (Dai et al., 2020)^†	0.612	0.719	0.605	0.475	0.668	0.494	0.633	0.240	0.638	0.234	0.696	0.487
MulT (Tsai et al., 2019)^†	0.672	0.754	0.640	0.483	0.649	0.475	0.614	0.256	0.629	0.253	0.716	0.493
BIMHA (Wu et al., 2022)^††	0.658	0.721	0.626	0.479	0.653	0.474	0.625	0.249	0.618	0.247	0.705	0.489
CMHA (Zheng et al., 2022)^††	0.652	0.721	0.642	0.467	0.659	0.491	0.645	0.266	0.634	0.273	0.736	0.532
MESM (Dai et al., 2021)^†	0.641	0.723	0.630	0.466	0.668	0.493	0.657	0.272	0.658	0.289	0.756	0.564
FE2E (Dai et al., 2021)^†	0.654	0.726	0.652	0.490	0.670	0.496	0.667	0.291	0.638	0.268	0.777	0.571
MER-SEM-MBT (Our full model)	0.673	0.753	0.668	0.538	0.687	0.495	0.676	0.330	0.672	0.319	0.802	0.616
MER-SEM-MBT (Ours w/o textual decision)	0.672	0.749	0.655	0.531	0.673	0.491	0.660	0.328	0.659	0.312	0.787	0.612

P < 0.05 for paired t-test.

^†

denotes the results are from Dai et al. (2021), and

^††

means our reproduction using the same data split as other experiments. The bold values are indicated to highlight the best results.