On Exploiting Trustors in Trust-Based Recommendation

Abstract

In a trust network, two users connected by a trust relationship are assumed to have similar interests. Based on this assumption, existing trust-based recommendation methods predict ratings for a target user on unseen items by utilizing available ratings information of those users who are "similar" to the target user in the trust network. Here, the concept of similarity is defined as the reachability via the trust relationship in the network. In the process, however, these methods usually follow the trustor-trustee relationship only in the forward direction, i.e., from trustor to trustee, but not in the backward direction. That is, they have overlooked the possibility of utilizing the ratings information obtainable from those users reachable in the backward direction who, we believe, are also deemed to have similar interests to the target user. In this paper, we investigate this possibility of identifying and adding these users to the trustable user group when predicting ratings by collaborative filtering. Experiments show that our approach can improve the coverage of prediction while preserving the accuracy.