The Study On Collaborative Filtering Recommendation Technology Based On Differential-Private

Recommender Systems can help users to discover information by their interest, and solve the information overload problem efficiently. At the same time, the recommended system privacy issues are gaining attention. For Recommender Systems, pursue for the recommendation accuracy and the user's privacy are both important. On the one hand, a more accurate recommendation will be greatly raised the user experience. On the other hand, the more powerful privacy guarantee will reduce the user's concerns about sharing personal privacy information, and thus increase the user's trust and participation degree.Collaborative filtering(CF) is the most commonly used algorithm in the recommendation system. However, the collaborative filtering recommendation system can not ensure the security of the user's privacy. Recent research revealed that by observing the public output of the CF, the adversary could infer the historical ratings of the particular user, which is known as the KNN attack and is considered a serious privacy violation for recommender systems. Differential privacy protection technology has the characteristics of strict mathematical definition and the maximum background knowledge hypothesis, which can ensure the security of user's privacy very well.The main contribution of this paper to includes the following aspects:First, we have studied the traditional neighborhood-based collaborative filtering algorithm deeply, including three main steps: collecting user preferences, finding similar users or items and calculating for recommendation. In order to pave the way for the subsequent application of privacy protection algorithm design, we have analysis the possible hidden private leak point in collaborative filtering recommendation algorithm.Secondly, we have studied the common privacy protection technologies, including k-Anonymity, data encryption, confusion and disturbance, and focus on the implementation of differential privacy protection algorithm in collaborative filtering recommendation algorithm.Then, we have studied the privacy attack model and its extension K-nearest neighbor attack, which the adversary could infer the historical ratings of the particular user by observing the public output of the CF.In this paper, the attack mode of the user's privacy information is reduced by using the high probability of the recommendation result, and the advantage of the differential privacy protection technology is compared with the traditional privacy protection method.Finally, based on the existing work, we proposed a privacy preserving algorithm, which has a higher degree of accuracy under the guarantee of privacy.