Research on Privacy-preserving and Recommendation Accuracy of Collaborative Filtering System

On the basis of the users' profile-exposing in the prediction generation process,a data obfuscation policy without affecting the accuracy of recommender system is proposed,in which the ratings of the responding user are substituted by false data before computing the similarity degree of users.The process does not(or a little) expose the ratings of user and preserves users' privacy data. Experimental results demonstrate the impact of obfuscation policies on the accuracy of the generated predictions,and show the improvement is effective.