Research on the Classification of Policy Instruments Based on BERT Model

Table 5

Comparative experimental results.

Model	Pretrained model	Loss function	Accuracy	F1	Precision	Recall
TF-IDF + SVM	—	CE	0.75	0.74	0.75	0.75
TextCNN	—	CE	0.67	0.66	0.68	0.67
LSTM	—	CE	0.67	0.67	0.71	0.67
PLM + MLP	BERT-base-Chinese	CE	0.82	0.82	0.83	0.83
PLM + MLP	BERT-base-Chinese	Focal	0.83	0.84	0.85	0.83
PLM + MLP	Chinese-RoBERTa-wwm-ext	CE	0.85	0.86	0.86	0.85