Efficient, secure and verifiable outsourcing scheme for SVD-based collaborative filtering recommender system

Abstract

With the explosive growth of e-commerce, recommender systems are widely used to generate personalized recommendation for customers. SVD-based collaborative filtering and its variants are highly accurate and scalable approaches to recommender systems. Due to the heavy computation of SVD-based collaborative filtering, outsourcing the computation is an efficient solution to reduce computational complexity. In this paper, we propose an efficient, secure and verifiable outsourcing scheme for SVD-based collaborative filtering recommender system. We use symmetric block diagonal matrices as seeds to generate secret keys, which are novel orthogonal sparse matrices to blind the target matrices of SVD. Security analysis shows that our scheme can protect the privacy of both the input and output and efficiency analysis shows that our scheme is (3m2+3n2+5mn)/(m3+n3) efficient compared to fully local algorithm. In our scheme, we also create a verification approach that is capable of detecting misbehavior from a cloud server with probability (1−12n). The experiment shows that the client achieves significant computational savings and the recommendation accuracy of the scheme is nearly as good as that of the fully local algorithm.