Automatic text classification of actionable radiology reports of tinnitus patients using bidirectional encoder representations from transformer (BERT) and in-domain pre-training (IDPT)

Table 7 Comparison of model performance metrics

Embedding	Classifier	Accuracy	Precision	Recall	AUC	F1-score
Word2Vec	CNN	0.729	0.744	0.729	0.767	0.733
	MLP	0.644	0.643	0.644	0.711	0.644
	Bi-LSTM	0.737	0.740	0.737	0.677	0.738
	Bi-LSTM-CNN	0.728	0.729	0.728	0.692	0.728
BERT	CNN	0.770	0.788	0.777	0.908	0.781
	MLP	0.719	0.714	0.719	0.874	0.712
	Bi-LSTM	0.777	0.792	0.780	0.888	0.774
	Bi-LSTM-CNN	0.698	0.696	0.698	0.861	0.690
	Fine-tune	0.760	0.761	0.759	0.868	0.760
	IDPT	0.842	0.843	0.842	0.948	0.841
BERT-wmm-ext	Fine-tune	0.756	0.756	0.756	0.883	0.754
Mengzi	Fine-tune	0.751	0.751	0.751	0.846	0.750
Roberta	Fine-tune	0.767	0.767	0.767	0.878	0.764

ISSN: 1472-6947