where p(z|u) stands for the likelihood for user u to be in
class z and p(r|z,m) stands for the likelihood of assigning
item m with rating r by users in class z. In order to achieve
better performance, the ratings of each user are normalized
to be a normal distribution with zero mean and variance
as 1 [17]. The parameter p(r|z,m) is approximated as a
Gaussian distribution N(mz, sz) and p(z|u) as a multinomial
distribution.