A Hybrid Neural Network BERT-Cap Based on Pre-Trained Language Model and Capsule Network for User Intent Classification

Table 4

The performance of different methods on two Chinese datasets.

Models\datasets	ECDT				FDQuestion
		R	F1	Acc		R	F1	Acc
BERT	0.962	0.952	0.956	0.955	0.791	0.819	0.801	0.843
BERT + CNN	0.968	0.956	0.96	0.960	0.798	0.824	0.808	0.849
BERT-Cap	0.969	0.960	0.963	0.963	0.808	0.831	0.814	0.856
BERT-Cap + focal loss	0.972	0.972	0.971	0.967	0.778	0.847	0.803	0.845