Vietnamese Sentiment Analysis under Limited Training Data Based on Deep Neural Networks

Table 5

The mean accuracy results (with standard deviation) of various data augmentation techniques for Vietnamese sentiment analysis in 10 runs based on the deep learning model.

Datasets	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Dataset 1	0.764 0.05	0.838 0.01	0.818 0.03	0.843 0.02	0.844 0.01	0.840 0.01	0.8470.0003
Dataset 2	0.415 0.06	0.518 0.03	0.438 0.10	0.433 0.07	0.5200.01	0.449 0.05	0.367 0.0003
Dataset 3	0.742 0.05	0.797 0.01	0.683 0.10	0.742 0.07	0.780 0.01	0.782 0.01	0.8010.0001
Dataset 4	0.757 0.02	0.8020.01	0.710 0.07	0.738 0.09	0.782 0.01	0.786 0.01	0.746 0.0001

(1) With preprocessing techniques; (2) EDA; (3) sentence shuffling; (4) back translation; (5) contextual substitution (); (6) contextual substitution ( +); (7) sentence embedding mixup (the standard deviations are very minimum).