Classifying protein-protein interaction articles using word and syntactic features

BMC Bioinformatics

Table 1 The corpus information used in our experiments.

Corpus Name	Positive Examples	Negative Examples	Total Examples
BioCreative II	3874	2298	6172
BioCreative II.5	124	1066	1190
BioCreative III Training Set	1140	1140	2280
BioCreative III Development Set	682	3318	4000
Total Training Set	5820	7822	13642
BioCreative III Test Set	910	5090	6000

BioCreative II, BioCreative II.5, BioCreative III training, and development sets were used as the training corpus for the ACT competition. While the training corpus is balanced, the BioCreative III test set is an imbalanced set with the number of negative examples about six times higher than the number of positive examples. Hence, for the official submission, system parameters were tuned for the BioCreative III development set.

ISSN: 1471-2105