Table 5. Validation results for deep learning models on the 21 K review dataset.

Method	Accuracy	Std. dev.	Class	Precision	Recall	F1-score
RNN	0.946	0.0017	Positive	0.970	0.981	0.975
			Neutral/Mixed	0.688	0.621	0.652
			Negative	0.930	0.914	0.922
GRU	0.950	0.0011	Positive	0.971	0.983	0.977
			Neutral/Mixed	0.710	0.627	0.665
			Negative	0.935	0.924	0.930
LSTM	0.945	0.0007	Positive	0.971	0.976	0.974
			Neutral/Mixed	0.676	0.621	0.646
			Negative	0.925	0.936	0.931
BERT	0.964	0.0027	Positive	0.975	0.993	0.984
			Neutral/Mixed	0.834	0.679	0.748
			Negative	0.955	0.946	0.951