Exploiting sequence labeling framework to extract document-level relations from biomedical texts

BMC Bioinformatics

Table 5 The overall and intra- and inter-sentential level results of different component evaluated on CDR corpus. ∆ denotes the corresponding F-score decrease percentage when a component is removed

Component(s) removed	Overall				Intra-sentential level				Inter-sentential level
Component(s) removed	P	R	F1	∆(%)	P	R	F1	∆(%)	P	R	F1	∆(%)
None	0.600	0.675	0.635	–	0.634	0.720	0.674	–	0.518	0.573	0.544	–
-DE	0.554	0.603	0.577	−5.8	0.592	0.683	0.634	−4.0	0.444	0.417	0.430	−11.4
- EnDet	0.599	0.647	0.622	−1.3	0.637	0.696	0.665	−0.9	0.507	0.533	0.520	−2.4
-DE & EnDet	0.551	0.611	0.579	−5.6	0.587	0.692	0.635	−3.9	0.446	0.420	0.433	−11.1
-HE & EnDet	0.609	0.633	0.621	−1.4	0.648	0.684	0.666	−0.8	0.513	0.515	0.514	−3.0

ISSN: 1471-2105