More reputable recommenders give more accurate recommendations?

Existing models of the Trust-Aware Recommender System (TARS) build personalized trust networks for the active users to predict ratings. These models have reasonable rating prediction performances, while suffer from high computational complexity. One solution is to utilize the global rating prediction mechanism for TARS, in which an intuitive assumption is that more reputable recommenders give more accurate recommendations. In addition, due to the scale-freeness of the trust network, some users have and continuously have superior reputations than others. However, we show via comprehensive experiments on the real TARS data that the recommendations given by recommenders with higher reputations do not tend to be more accurate. Furthermore, even the recommendations given by the recommenders with superior high reputations do not tend to more accurate. Our experimental study provides promising directions for the future research on the rating prediction mechanism of TARS.