Short Text Paraphrase Identification Model Based on RDN-MESIM

Table 4

The results of different networks on PI task using tax intelligence consulting question pairs’ data based on original ESIM.

Structure	Precision	Recall	Fmeasure	Loss	Accuracy
ESIM [19]	0.9666	0.9577	0.9621	0.1036	0.9649
+BiGRU	0.9739	0.9310	0.9519	0.1037	0.9610
+BiGRU + Dense + L2	0.9700	0.9311	0.9501	0.1327	0.9575
+BiGRU + CNN	0.9939	0.9505	0.9717	0.0993	0.9738
+K-Maxpooling + Dense	0.9811	0.9404	0.9603	0.1121	0.9689
+CNN + K-Maxpooling + Dense + L2	0.9807	0.9414	0.9606	0.1111	0.9666
+CNN + attention + L2	0.9796	0.9542	0.9667	0.0945	0.9699
+Xception [25]	0.9896	0.9453	0.9669	0.1104	0.9689
RDN-MESIM	0.9889	0.9612	0.9748	0.0808	0.9763