A detailed error analysis of 13 kernel methods for protein-protein interaction extraction

BMC Bioinformatics

Table 8 Classification results on the 1510 NE pairs with CV evaluation

Kernel	r	TN	FN	e
APG	25.3	1510	0	1129
cosine	24.9	1510	0	1134
edit	18.1	1510	0	1237
combined	24.8	1510	0	1135
shallow	24.6	1510	0	1138
syntactic	24.4	1510	0	1142
kBSPS	36.6	1509	1	957
SL	30.4	1508	2	1051
lexical	25.0	1506	4	1133
PT	27.9	1505	5	1089
ST	28.0	1502	8	1088
SST	26.6	1501	9	1108
SpT	46.4	1484	26	810

Classification results on the 1510 NE pairs with CV evaluation (in decreasing order according to the successfully classified pairs). Ratio (r) refers to the distribution of positive classes predicted by the kernel measured across the 5 corpora; TN/FN is the number of correctly/incorrectly classified NE pairs; e is 1510·(1−r), the expected number of negative class prediction projected onto the 1510 NE pairs.

ISSN: 1471-2105