Table 3.

Combined performance results

Exp	Method	P	r	F	signif	% alt
A	CRF noalt, nom and word	0.9255	0.8885	0.9066	1-19, C-F	13.62
B	BDT nom and word	0.9221	0.8885	0.9050	1-19, C-F	25.67
C	BDT nom and word, top 10 teams	0.9118	0.8768	0.8940	1-19, E, F	23.37
D	BDT nom only	0.9092	0.8773	0.8929	1-19, E, F	25.42
E	BDT noalt, nom and word	0.9242	0.8165	0.8670	7-19, F	9.58
F	BDT word only	0.7165	0.6187	0.6640	18-19	37.07

The precision, recall, and F score of machine learning experiments to learn gene mentions using the data extracted from all submitted runs as features. Method column: BDT, boosted decision trees; CRF, conditional random fields; nom, all nomination features; word, words of candidate; noalt, alternate gene data not used. The column signif indicates the ranks of runs for which there was a significant difference, and the letters indicate the machine learning experiments for which there was a significant difference. The column % alt gives the percentage of alternate gene mentions among the resulting true positives.