In order to develop an active learning algorithm for aspect
model, one should consider the rating prediction method in
aspect model and then optimize it. In aspect model, the
unknown ratings are estimated according to equ. 1. This
equation consists of user latent parameters and item latent
parameters. After getting a new rating, these parameters are
retrained to improve them toward true latent parameters. Better
estimation of user and item parameters provides a more
accurate prediction for unknown ratings. According to [17]
item parameters have a normal distribution with mean and
variance that are computed as following :