Imputing trust network information in NMF-based collaborative filtering

Abstract

We propose an NMF (Nonnegative Matrix Factorization)-based approach in collaborative filtering based recommendation systems to handle the cold-start users issue, especially for the New-Users who did not rate any items. The proposed approach utilizes the trust network information to impute missing ratings before NMF is applied. We do two cases of imputation: (1) when all users are imputed, and (2) when only New-Users are imputed. To study the impact of the imputation, we divide users into three groups and calculate their recommendation errors. Experiments on four different datasets are conducted to examine the proposed approach. The results show that our approach can handle the New-Users issue and reduce the recommendation errors for the whole dataset especially in the second imputation case.