Incorporating Auxiliary Information in Collaborative Filtering Data Update with Privacy Preservation

Online shopping has become increasingly popular in recent years. More and more people are willing to buy products through Internet instead of physical stores. For promotional purposes, almost all online merchants provide product recommendations to their returning customers. Some of them ask professional recommendation service providers to help develop and maintain recommender systems while others need to share their data with similar shops for better product recommendations. There are two issues, (1) how to protect customers’ privacy while retaining data utility before they release the data to the third parties; (2) based on (1), how to handle data growth efficiently. In this paper, we propose a NMF (Nonnegative Matrix Factorization)-based data update approach in collaborative filtering (CF) that solves the problems. The proposed approach utilizes the intrinsic property of NMF to distort the data for protecting user’s privacy. In addition, the user and item auxiliary information is taken into account in incremental nonnegative matrix tri-factorization to help improve the data utility. Experiments on three different datasets (MovieLens, Sushi and LibimSeTi) are conducted to examine the proposed approach. The results show that our approach can quickly update the new data and provide both high level privacy protection and good data utility. Keywords—auxiliary information; collaborative filtering; data growth; nonnegative matrix factorization; privacy