Research On Collaborative Filtering Recommendation Method Based On Privacy Protection

The recommendation system is to solve the problem of information overload brought by the network age,in order to provide users with personalized recommendation services,analyze the user's behavior data and basic information,and tap the user's preference characteristics.The collaborative filtering recommendation method is the most widely used,and its main feature is to recommend by calculating the similarity between different users and projects.The new era brings different opportunities and challenges.This paper analyzes the status and shortcomings of existing collaborative filtering recommendation technologies based on privacy protection.Then we start from the different perspectives of privacy protection and collaborative filtering to carry out in-depth research,and propose an improved solution for collaborative filtering methods and corresponding improved privacy protection methods.The main research achievement as follows:First,we study the main methods of the current anonymous privacy protection in view of the problem that the protection level of the traditional anonymous privacy protection method is too high and affects the recommendation effect.The current approach to anonymous privacy protection focuses on enhancing the high level of security provided by protection,making data mining after protection more difficult.In order to solve the characteristics of high data loss and high robustness in these methods,this paper proposes an improved anonymous privacy protection method for collaborative filtering recommendation.Firstly,the method introduces an improved(p,?)-k anonymous privacy protection algorithm before generating the recommendation results.The algorithm can protect sensitive information in preference-aware environment information and provide different levels of protection for different sensitivity levels.We analyze different sensitive attribute values,divide the data set into different categories,and then set different leak rates for them.In addition,the algorithm combines the idea of clustering and generalization to improve the original k-anonymity model,which can reduce the information loss rate.Experimental results show that the model can effectively achieve a better balance between privacy protection and data availability.Second,there is a problem of inefficiency against the traditional cryptographic privacy protection techniques for recommendation.In this paper,a gradient-decreasing data structure for encryption recommendation method is used to completing homomorphic encryption needs to perform fixed-point arithmetic in encrypted form.Then,in order to improve the efficiency of recommendation,this paper introduces the method of machine learning in collaborative filtering,and proposes a collaborative filtering recommendation method based on non-negative matrix factorization.This method makes up for the extreme sparseness of the target-rating matrix in the traditional collaborative filtering recommendation method,which greatly affects the recommendation effect.At the same time,we design a trust-based matrix factorization recommendation method based on Tikhonov regularization and matrix factorization.We also propose a new technique to analyze the rating matrix,retaining the advantages of classical decomposition techniques in this paper.The experimental results show that the method can guarantee the availability of data under the premise of improving data security as much as possible.