TABLE 2.

Cross validation performance comparison between different deep structures and feature encondings.

Model tuning	Accuracy	Precision	Recall	F-score
One-hot encoding + 2D convolutional neural network (CNN)	0.623 ± 0.037	0.662 ± 0.028	0.636 ± 0.019	0.647 ± 0.021
Embedding layer + CNN	0.732 ± 0.006	0.745 ± 0.011	0.692 ± 0.024	0.716 ± 0.029
Fixed word2vec + CNN	0.685 ± 0.024	0.701 ± 0.019	0.653 ± 0.015	0.677 ± 0.022
Transfer embedding + recurrent neural network (RNN)	0.743 ± 0.012	0.749 ± 0.004	0.716 ± 0.017	0.729 ± 0.015
Proposed method	0.782 ± 0.008	0.791 ± 0.013	0.785 ± 0.011	0.782 ± 0.016