A review: preprocessing techniques and data augmentation for sentiment analysis

Computational Social Networks

Table 3 The F1 scores of various experiments

Datasets	Classifiers	(1)	(2)	(3)	(4)	(5)	(2)+(3)	(2)+(4)	(2)+(5)
Dataset 1	LR	0.828	0.847	0.834	0.835	0.830	0.861	0.862	0.861
	SVM	0.829	0.856	0.829	0.837	0.812	0.854	0.861	0.838
	OVO	0.829	0.850	0.838	0.839	0.818	0.863	0.862	0.852
	OVR	0.829	0.850	0.838	0.839	0.818	0.863	0.862	0.852
Dataset 2	LR	0.700	0.739	0.706	0.704	0.695	0.743	0.743	0.740
	SVM	0.662	0.705	0.698	0.696	0.669	0.739	0.736	0.731
	OVO	0.706	0.744	0.713	0.717	0.681	0.750	0.750	0.736
	OVR	0.693	0.726	0.699	0.702	0.675	0.730	0.729	0.725
Dataset 3	LR	0.790	0.820	0.793	0.791	0.786	0.827	0.824	0.824
	SVM	0.789	0.818	0.793	0.791	0.782	0.824	0.821	0.819
	OVO	0.793	0.818	0.794	0.793	0.783	0.825	0.821	0.820
	OVR	0.793	0.818	0.794	0.793	0.783	0.825	0.821	0.820
Dataset 4	LR	0.779	0.820	0.790	0.791	0.772	0.826	0.822	0.818
	SVM	0.784	0.817	0.788	0.786	0.766	0.819	0.822	0.808
	OVO	0.785	0.822	0.789	0.787	0.766	0.824	0.826	0.808
	OVR	0.785	0.822	0.789	0.787	0.766	0.824	0.826	0.808

(1) Without preprocessing techniques (baseline results), (2) with preprocessing techniques, (3) back translation, (4) Syntax-Tree transformation, (5) EDA