Table 2. Algorithm of AHD.

Input:

data , (one positive and one negative)

number of training data

number of iterations

number of active choosing data for every iteration

parameter space

Output:

deep architecture with parameter space

for ; ; do

Train HDBN with labeled dataset and all unlabeled data in .

Choose reviews which near the separating line from train dataset through Eq. 17.

Add reviews into the labeled data set .

end for

Train HDBN with labeled dataset and all unlabeled data in .