Improving Medical Q&A Matching by Augmenting Dual-Channel Attention with Global Similarity

Table 3

The top-1 accuracy results of the models.

Index	Pretrained embedding	Model	Dev (%)	Test (%)
1	GloVe	SingleCNN	64.5	64.1
2		MultiCNN	65.4	64.8
3		Multiscale attentive interaction networks	66.1	67.1
4		BiGRU-CNN	68.5	68.5
5		DCAG	71.9	72.5
6	BERT	SingleCNN	66.3	66.8
7		MultiCNN	67.5	67.2
8		Multiscale attentive interaction networks	70.1	70.5
9		BiGRU-CNN	72.4	71.9
10		DCAG	75.1	75.6

The best performance is boldfaced, and the state-of-the-art performance is italicized.