First page Back Continue Last page Overview Graphics
The four CRM114 configurations tested:
OSBF – OSB with “double extra voodoo” (a very fast approximate TF-IDF probability modifier)
Winnow – an implementation of Nick Littlestone’s Winnow algorithm (basically a wide perceptron with back-propagation learning)
OSB and OSB-Unique – naïve Bayesian classifiers; the only difference is Unique disregards all but the first appearance of any feature.