Table 1: Classification approaches used: Naïve Bayes classifier (NBC), NBC + Expectation Maximization (EM), Maximum Entropy (MaxEnt), Conditional Random Fields (CRFs).

Granularity Features Classifier

Single label, Bag-of-words (BOW) NBC
Sentence level BOW + gene names boosted
BOW + trigger words boosted
BOW + gene names and trigger
words boosted

Multiple labels BOW NBC +
Sentence level MaxEnt

Event trigger BOW + CRFs
phrase labeling3-gram and 4-gram
prefixes and suffixes +
orthographic features +
trigger phrase dictionary