Biomedical relation extraction via knowledge-enhanced reading comprehension

BMC Bioinformatics

Table 6 Ablation study over our proposed KRC model on the CDR dataset

Model	Intra sentential level\(^{\mathrm{1}}\)			Inter sentential level\(^{\mathrm{2}}\)			Document level
Model	P (%)	R (%)	F1 (%)	P (%)	R (%)	F1 (%)	P (%)	R (%)	F1 (%)
Only KB	58.61	62.29	60.39	39.56	18.67	25.37	52.86	81.61	64.16
RC	67.09	54.50	60.14	60.10	11.16	18.83	65.83	66.32	66.07
RC+KB(atten2)	67.94	56.85	61.90	62.00	11.63	19.59	66.88	69.14	67.99
RC+KB(atten1+atten2)	72.74	58.07	64.58	68.31	11.73	20.02	71.93	70.45	71.18

\(^{\mathrm{1,2}}\)metrics on intra and inter sentential levels, we used the calculation methods in [1]. If using the calculation methods in [24], F1 measures of our model without KBs (RC in the table) are 73.82% and 57.07% respectively on intra and inter sentential levels

ISSN: 1471-2105