Table 4.

Experimental results for Google Play application review data.

Method	Principle	Sentiment correct classification rates (normalized)
		Negative sentiment	Positive sentiment	Overall
Logistic regression	TFIDF	0.88	0.90	0.89
Cost sensitive logistic regression	TFIDF	0.89	0.89	0.89
Support vector machine	TFIDF	0.91	0.91	0.91
Naive Bayes	TFIDF	0.85	0.91	0.88
Logistic regression	Count vect	0.89	0.92	0.91
Cost sensitive logistic regression	Count vect	0.90	0.91	0.91
Support vector machine	Count vect	0.89	0.92	0.91
Naive Bayes	Count vect	0.86	0.90	0.88
Vanilla LSTM	Deep learning	0.90	0.84	0.87
Stacked LSTM	Deep learning	0.85	0.88	0.87
Bi-directional LSTM	Deep learning	0.89	0.92	0.91
GRU	Deep learning	0.87	0.91	0.89
Stacked GRU	Deep learning	0.92	0.85	0.88
CNN-LSTM	Deep learning	0.83	0.95	0.89
GRU-CNN	Deep learning	0.86	0.92	0.89
Proposed ensemble fuzzy method	Ensemble	0.91	0.93	0.92