MTQA: Text-Based Multitype Question and Answer Reading Comprehension Model

Table 4

Accuracy of key steps in model construction.

Pretrained model	Model	Accuracy		Epoch	Learning rate
		EM	F1
Base	MTMSNBert_Base [21]	68.17	72.81	—	—
	MTMSNElectra_Base	71.49	75.95	8	3e − 5
	MTQABert_Base	69.59	74.87	13	5e − 5
	MTQAElectra_Base	74.62	78.98	8	3e − 5
	MTQApre_Electra_Base	75.67	79.99	8	3e − 5
Large	MTMSNBert_Large [21]	76.68	80.54	—	—
	MTMSNElectra_Large	77.65	82.01	6	3e − 6
	MTQABert_Large	77.96	81.78	8	5e − 6
	MTQAElectra_Large	80.26	84.45	6	9e − 6
	MTQApre_Electra_Large	81.25	85.26	12	5e − 6