Trust-enhanced matrix factorization using PageRank for recommender system

As the most widely used recommendation algorithm, collaborative filtering (CF) has been studied for many years due to its simplicity and effectiveness. The two main categories of CF have their own shortcomings. Memory-based CF can't generate accurate results when faced with data sparsity; and model-based CF always loses the information between users or items. To alleviate this problem, we propose an algorithm that integrate user trust into the traditional matrix factorization (MF). Trust network is introduced to utilize all the trusted users to help make prediction. Experiments are performed on Epinions dataset and FilmTrust dataset to compare proposed approach with traditional ones. The reported results indicate that the idea of employing user trust into MF is valid and can improve the recommendation quality.