Thus, in addition to selecting the items which will accelerate
the model towards the true user model θtrue
u , the active
learning algorithm should also try to select items which have
a very high probability of getting a rating from a user. To
address this issue, Harpale and Yang [19] extended Jin and
Si [18] and introduced a new term P(m|u) into Equation 3,
that is the probability of getting a rating, on the item m from
the user u: