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

Table 7

QUOREF experiments.

Model	Accuracy		Epoch	Learning rate
	EM	F1
QANet [30]	34.41	38.26	—	—
BERT QA [28]	58.44	64.95	—	—
XLNet QA [18]	64.52	71.49	—	—
CorefRoBERTlarge [31]	74.94	81.71	—	—
TASEIO [26]	79.4	84.9	—	—
MTQAElectra_Base	72.56	78.23	10	5e − 5
MTQAElectra_Large	81.34	86.84	9	5e − 6
Human performance [18]	86.75	93.41	—	—