Table 4.

Binary sentiment analysis results of the multi-modal model using the recurrent EEG decoding component.

	Randomly initialized			GloVe			BERT
Model	P	R	F1 (std)	P	R	F1 (std)	P	R	F1 (std)
Baseline	0.572	0.573	0.552 (0.07)	0.751	0.738	0.728 (0.08)	0.900	0.899	0.893 (0.04)
+ noise	0.599	0.574	0.541 (0.08)	0.721	0.715	0.709 (0.09)	0.914	0.916	0.913 (0.03)
+ ET	0.615	0.605	0.586 (0.06)	0.795	0.786	0.781 (0.06)	0.913	0.907	0.904 (0.05)
+ EEG full	0.540	0.538	0.525 (0.06)	0.738	0.729	0.725 (0.07)	0.913	0.909	0.906 (0.04)
+ EEG θ	0.602	0.599	0.584* (0.08)	0.789	0.785	0.783+ (0.05)	0.917	0.916	0.913* (0.04)
+ EEG α	0.610	0.590	0.565 (0.05)	0.763	0.758	0.753 (0.05)	0.912	0.908	0.906 (0.03)
+ EEG β	0.587	0.578	0.555 (0.07)	0.781	0.777	0.774+ (0.06)	0.911	0.911	0.907* (0.04)
+ EEG γ	0.614	0.591	0.553 (0.08)	0.777	0.773	0.769* (0.07)	0.917	0.917	0.915* (0.04)
+θ+α+β+γ	0.597	0.597	0.569 (0.08)	0.766	0.764	0.760* (0.07)	0.913	0.913	0.911* (0.04)

We report precision (P), recall (R), F₁-score and the standard deviation (std) between five runs. The best results per column are marked in bold. Significance is indicated on the F₁-score with asterisks:

^*denotes p < 0.05 (uncorrected), + denotes p < 0.003 (Bonferroni corrected p-value).