Elevating Prediction Accuracy in Trust-aware Collaborative Filtering Recommenders through T-index Metric and TopTrustee lists

摘要

The growing popularity of Social Networks raises theudimportant issue of trust. Among many systems which haveudrealized the impact of trust, Recommender Systems haveudbeen the most influential ones. Collaborative Filtering Recommenders take advantage of trust relations between usersudfor generating more accurate predictions. In this paper, weudpropose a semantic recommendation framework for creatingudtrust relationships among all types of users with respectudto different types of items, which are accessed by uniqueudURI across heterogeneous networks and environments. Weudgradually build up the trust relationships between usersudbased on the rating information from user profiles and itemudprofiles to generate trust networks of users. For analyzingudthe formation of trust networks, we employ T-index as anudestimate of a user’s trustworthiness to identify and selectudneighbors in an effective manner. In this work, we utilizeudT-index to form the list of an item’s raters, called Top-udTrustee list for keeping the most reliable users who haveudalready shown interest in the respective item. Thus, when auduser rates an item, he/she is able to find users who canudbe trustworthy neighbors even though they might not beudaccessible within an upper bound of traversal path length.udAn empirical evaluation demonstrates how T-index improvesudthe Trust Network structure by generating connections toudmore trustworthy users. We also show that exploiting Tindexudresults in better prediction accuracy and coverageudof recommendations collected along few edges that connectudusers on a Social Network.