Medical Specialty Classification Based on Semiadversarial Data Augmentation

Table 2

The performance of different models trained in a plain way and the proposed method.

Models	Mode	Acc	Micro-R	Micro-	Micro-F1
CNN	Plain	65.2	63.8	65.9	64.8
	SemiADA + PI	79.1	79.3	78.1	78.7
LSTM	Plain	70.5	70.6	71.9	71.2
	SemiADA + PI	81.6	84.1	81.2	82.6
BERT	Plain	69.4	71.5	69.6	70.5
	SemiADA + PI	83.9	83.9	83.0	83.4
BioBERT	Plain	73.0	71.5	74.7	73.1
	SemiADA + PI	87.7	88.6	87.9	88.2

SemiADA + PI is our proposed method, where SemiADA represents the data augmentation mechanism and PI represents the classification mechanism incorporating probabilistic information.