Parallel sequence tagging for concept recognition

BMC Bioinformatics

Table 1 Upper bound of annotation performance (F-score) when using the CoNLL format, comparing different simplification strategies

Unification	First-span		Full-span	Last-span
Unnesting	Keep-shorter	Keep-longer	Keep-longer
CHEBI	0.9979	0.9980	0.9972	0.9974
CL	0.9706	0.9720	0.9574	0.9692
GO_BP	0.9607	0.9626	0.9587	0.9570
GO_CC	0.9811	0.9813	0.9801	0.9785
GO_MF	0.9974	0.9974	0.9974	0.9974
MOP	0.9967	0.9967	0.9967	0.9967
NCBITaxon	0.9996	0.9996	0.9995	0.9996
PR	0.9624	0.9627	0.9619	0.9618
SO	0.9829	0.9831	0.9816	0.9825
UBERON	0.9792	0.9798	0.9776	0.9780

For each annotation type (row), the best value is given in bold
For reasons of clarity, the combinations full-span/keep- shorter and last-span/keep-shorter are omitted; their results are in most cases inferior to those presented in this table

ISSN: 1471-2105