Research On Privacy-Preserving Collaborative Filtering Recommendation Algorithm

With the gradual increase of the information resources on the internet, users'demands for personalized services are increasingly heightend. Collaborative Filtering Recommendation Technique is one of the most successfully and widely used recommendation techniques, however, with privacy considerations, many users are reluctant to divulge personal information. Therefore, how to make users provide necessary information that recommendation systems use without revealing their privacy becomes the main issue in the development of Collaborative Filtering Recommendation.In this paper, we mainly research into privacy protection problem in Collaborative Filtering Recommendation.Firstly, aimed at the problem that Collaborative Filtering Recommendation using P2P network model can not properly protect users'privacy, we deeply analysis the existing network models and propose a hierarchical hypercube P2P network structure. And we also propose a hierarchical hypercube P2P structure users'model stored Collaborative Filtering Algorithm. In it, users are classified as super users and regular users. Local recommendation and local average similarity are transmitted among users. Therefore, the algorithm can protect users'privacy better without affecting the accuracy of the recommendation.Secondly, aimed at the problem that Collaborative Filtering Recommendation Algorithm based on Randomized Perturbation Technique brings accuracy decrease when protecting users'privacy, we deeply analysis the principle of the Randomized Perturbation Technique. The concept of perturbation intensity weight, its measurement and an improved method for similarity calculation are introduced. Based on these, we propose an improved privacy preserving collaborative fitering algorithm. In it, Users'perturbation intensity weight are calculated according to its perturbation intensity. Both users'rating similarity and its perturbation intensity weight are considered when similarity is calculated. Experimental results show that the algorithm outperforms the initial one in accuracy without affecting the effectiveness of privacy protection.Finally, experimental schemes for hypercube P2P model structure stored collaborative filtering and privacy preserving collaborative fitering improved algorithm based on Randomized Perturbation Technique are given, and the two algorithms are verified.