Research On Collaborative Filtering Recommendation Algorithm Based On Differential Privacy

In recent years,with the sustained and rapid development of Internet and mobile communication technologies,various types of information on the Internet is increasing explosively.The recommender system can find the information that people need from the massive data and has been widely used.The collaborative filtering recommendation algorithm is one of the most widely used recommendation algorithms in the current recommendation field,but there is a big risk of user privacy leakage in the collaborative filtering recommendation,which has become a recognized fact.More and more users do not want to provide more personal information because of the privacy disclosure concerns,which is harmful to the further development of the recommender system.Establishing the privacy protection means for the recommender system is becoming a research hotspot now.In the field of data privacy protection,differential privacy protection technology has been researched and approved by researchers both at home and abroad due to its strict privacy protection effect.However,the research on differential privacy protection mostly focuses on the theory and lacks the research on practical application.In order to improve the privacy protection level of the recommender system,this paper combines the differential privacy protection and the collaborative filtering recommendation algorithm together.Based on the characteristics of the two different types of collaborative filtering,this paper constructs the corresponding differential privacy protection models.The main research work and achievements are as follows:First of all,aiming at the privacy protection of the collaborative filtering recommendation algorithm based on matrix factorization,this paper adopts a matrix factorization collaborative filtering recommendation algorithm model as the research foundation of privacy protection.This recommendation algorithm adds biases to the basic matrix factorization model to improve the accuracy of the recommendation.Then,we analyze that how to apply the differential privacy protection to the recommendation algorithm.By adopting the differential privacy protection to the calculation of average and stochastic gradient descent,a differential privacy matrix factorization recommendation algorithm is designed and implemented.Secondly,aiming at the privacy protection of neighborhood-based collaborative filtering recommendation algorithm,this paper also uses an optimized neighborhood-based collaborative filtering model which add biases to the basic model as the research foundation.By applying the corresponding differential privacy protection method to the process of the recommendation algorithm,which contains calculating the average,calculating the biases,neighbor selection,similarity calculation,a differential privacy neighborhood-based recommendation algorithm is constructed.For the two algorithms proposed in this paper,we not only analyze and prove that the algorithm satisfies the e-differential privacy theoretically,but also evaluate privacy-personalization trade-off on the data sets of two different scales,MovieLens and Netflix.The experimental results show that the proposed differential privacy protection collaborative filtering algorithm can achieve better recommendation accuracy based on the guaranteed differential privacy protection.Compared with the existing differential privacy protection recommendation algorithm,with the slight sacrifice of the privacy protection effect,the proposed algorithms can get better recommended results,which make them have better practical value.