Variable Weighted BSVD-Based Privacy-Preserving Collaborative Filtering

Recommender systems typically use collaborative filtering to make sense of huge and growing volumes of data. However, sharing user-item preferential data for use in collaborative filtering poses significant privacy and security challenges. In recent years, privacy has attracted a lot of attention. There are many existing works on privacy-preserving collaborative filtering. However, while these schemes are theoretically feasible, there are many practical implementation difficulties on real world. In this paper, a privacy-preserving collaborative filtering algorithm based on weighted singular value decomposition is proposed. The users´ needs are considered in the algorithm, and the user can disturb their original data with different weights according to their needs. At the privacy-preserving stage, the variable weighted-based BSVD scheme is used to protect the data privacy. At the prediction stage, the improved Slope One algorithm is used to get the prediction. Some experiments are performed using the proposed algorithm. The results indicate a good performance of the scheme in comparison with the Slope One algorithm. Meanwhile, it is shown that the algorithm can preserve the data privacy efficiently with high data usability.