An Improved Trust Metric for Trust-Aware Recommender Systems

Abstract

Collaborative filtering (CF) is the most widely used technique for recommender systems. However, user similarity alone is not enough for recommendation. We propose that trust is another important issue in recommender systems. Due to data sparsity of the item ratings matrix, we may not find the similar neighbors of the active user and thus CF recommender systems often fails in this condition. Taking trust into consideration can alleviate those problems. We consider replacing similarity weight with trust weight by trust propagation over the trust network. And we propose that trust decreases along propagation. A comparison between MoleTrust and our trust metric-DecTrust based on Epinions.com dataset shows that our trust metric can improve the accuracy while keeping coverage.