User preference with trust and rating in collaborative filtering

Collaborative filtering as the most successful recommender technique is widely used for e-commerce to solve the problem of information overload. In this paper, focusing on the poor scalability of existing user preference in collaborative filtering, we present a new user preference with trust and rating. According to the sparsity of rating matrix, the threshold value of co-ratings is set to selectively use the rating and the trust value and compute the similarity weight. Then, based on user preference with trust and rating, an algorithmic framework is proposed for performing recommendation. Comparing the above algorithm to traditional collaborative filtering and trust-based collaborative filtering on different sparsity rating matrix, Epinions, Flixster and ER, we run experiments that show the presented algorithm has is greatly improved on coverage and attains a good accuracy, and the presented user preference has the good scalability.