13 Nov 2004
WIDM 04: Lee et al. Co-training Web Block
Classification
8
Architecture
•B&M co-training handles only binary
classification
•Handles distribution skewing
L1 learner
L1 training
data
data
L1 model
L2 learner
L2 training
data
data
L2 model
training
data
data
unlabeled
data
data
sample
unlabeled
data
unlabeled
data
L1 labeled
data
data
L2 labeled
data
data
Conserve
p/n ratio
Conserve
p/n ratio
Add k new examples