ENHANCE NMF-BASED RECOMMENDATION SYSTEMS WITH SOCIAL INFORMATION IMPUTATION

We propose an NMF (Nonnegative Matrix Factorization)-based approach in collaborative filtering based recommendation systems to improve the Cold-Start-Users predictions since Cold-Start-Users suffer from high error in the results. The proposed method utilizes the trust network information to impute a subset of the missing ratings before NMF is applied. We proposed three strategies to select the subset of missing ratings to impute in order to examine the influence of the imputation with both item groups: Cold-Start-Items and Heavy-Rated-Items; and survey if the trustees' ratings could improve the results more than the other users. We analyze two factors that may affect results of the imputation: (1) the total number of imputed ratings, and (2) the average of imputed rating values. Experiments on four different datasets are conducted to examine the proposed approach. The results show that our approach improves the predicted rating of the cold-start users and alleviates the impact of imputed ratings