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

Table 5

Comparison with existing models.

Models	Pretraining	Accuracy		Epoch	Learning rate
		EM	F1
QANet + ELMo [27]	—	27.71	30.33	—	—
BERT [28]	Base	30.10	33.36	—	—
NAQANet [19]	—	46.20	49.24	—	—
MTMSN [21]	Base	68.17	72.81	—	—
	Large	76.68	80.54	—	—
TbMS [22]	Base	66.91	70.55	—	—
	Large_SQuAD	76.91	79.92	—	—
MTQA	pre_Electra_Base	75.67	79.99	8	3e − 5
	pre_Electra_Large	81.25	85.26	12	5e − 6
Human performance [19]	—	92.38	95.98	—	—