<![CDATA[In most B2Cs, such as 5-star Amazon, users’ rating capability is nonhomogeneous, which make it necessary to distinguish different users' profiles. Besides, recommender systems sometimes overlook differences among users’ judging power and simply recommend for the target users based on the preferences of his/her neighbors, leading to unsatisfactory recommendation performances. In this paper, we firstly propose a natural extension of the YZLM algorithm to get users’ judging power. Then we equally divide the users into some groups so that the intra-group users have close judging power. Through experiments on three benchmark datasets, namely MovieLens, Netflix and Amazon, we can find the interesting phenomenon which indicates a positive correlation between groups’ judging power and group’s recommendation performance. By analyzing this phenomenon in detail, we put forward some guidance to improve recommendation algorithm according to different users' judging power. DOI: http://dx.doi.org/10.11591/telkomnika.v11i7.2277]]>
