DCMR: A Method for Combining User-Based and Trust-Based Recommendation

Trust-based methods can recommend for cold-start users while collaborative filtering cannot, but collaborative filtering methods outperform in precision when recommending for the users with many ratings. In our approach, we combine these two kinds of methods in a novel way that exerts both of their advantages. Our combination method is a procedure of weight distribution and collection on predictors. It finds predictors by the breadth first search through the trust network. We present prediction confidence and trust attenuation as the two factors that affect weight distribution. Our experimental evaluation on the Epinions data set indicates that our method has a good performance.