IARNN-Based Semantic-Containing Double-Level Embedding Bi-LSTM for Question-and-Answer Matching

Table 5

Top-1 accuracy and loss of each model.

	Model	Train acc (%)	Train loss	Test acc (%)
1	BM25	44.80	∗∗∗	45.40
2	Multiscale CNN	66.53	1.95	64.67
3	Attentive pooling	85.33	1.7851	72.5256
4	DMN	75.24	0.92	74.38
5	ESIM + ELMo	80.37	1.15	77.15
6	Multiview	79.61	1.25	75.37
7	CapsNet	82.63	1.37	77.53
8	SCDE-Bi-LSTM	84.23	0.95	79.15