Personalized Recommendation via Suppressing Excessive Diffusion

Figure 1

Illustration of redundant similarities. Objects and users are marked with circles and squares, respectively. The solid line represents the object that has been collected by user and the dotted line denotes the similarity between objects. As we can see that the resources of C obtained through the diffusion following paths Auser 1C and Buser 2C are independent for there is no common user collecting A, B, and C together in Figure 1(a), however, in Figure 1(b), the resources of F are obtained from the same user and are counted twice. If user 4 chooses F just because F has some similarities with D (or E), the similarity E-F (or D-F) is redundant when calculating the resource of F, which degrades the recommendation accuracy.

(a)

(b)