An Approach Based on Multilevel Convolution for Sentence-Level Element Extraction of Legal Text

Table 6

The comparison results on legal dataset of single-layer CNN and multilevel CNN.

Model	Divorce				Loan				Labor
	HL (-)	Micro-P (+)	Micro-R (+)	Micro-F1 (+)	HL (-)	Micro-P (+)	Micro-R (+)	Micro-F1 (+)	HL (-)	Micro-P (+)	Micro-R (+)	Micro-F1 (+)
Single CNN ()	0.02337	0.8776	0.8612	0.8693	0.02814	0.8468	0.8416	0.8441	0.01785	0.8757	0.8386	0.8567
Single CNN ()	0.02280	0.8703	0.8705	0.8704	0.02892	0.8482	0.8386	0.8434	0.01904	0.8617	0.8461	0.8538
Single CNN ()	0.02385	0.8652	0.8719	0.8685	0.02762	0.8512	0.8296	0.8402	0.02131	0.8576	0.8432	0.8503
Our model	0.02126	0.8852	0.8728	0.8790	0.02772	0.8633	0.8573	0.8603	0.01688	0.8812	0.8589	0.8699