Research On Privacy-preserving Personalized Recommendation In Mobile Internet

With the popularity and development of mobile internet,Location-Based Services(LBSs)are becoming more and more popular.Location based personalized recommendation are one of the typical applications.It is difficult to find a place to meet their preference quickly for mobile users in the relatively unfamiliar geographical environment,while location based personalized recommendation can mine users' interests and provide users with accurate and effective service information,and avoid users to be engaged into the "Information Overload" dilemma.However,the location privacy safety problem can't be ignored caused by the disclosure of users' location information.The leakage of users' location information may expose more sensitive personal information,such as activity pattern,working environment et al.,which leads that users have concerns on their information privacy and hinders the further healthy development of the mobile internet.However,if users' location privacy is under inappropriate "excessive" protection,it is difficult to guarantee the recommendation effect.Therefore,how to improve personalized recommendation result as well as protect users' privacy has become an important research topic in location based personalized recommendation service.Based on the above considerations,this thesis proposes a users' query-based privacy-preserving personalized recommendation scheme.The scheme can protect the users' location privacy and query privacy,and based on the historical query data processed by the privacy protection,the scheme can mine the users' interest in the location category and the users' hot activity area,and make personalized location recommendation results for the users.The main contents are as follows:1.Design the location privacy protection algorithm.The scheme adopts the system architecture of both the client and the LBS location service provider.There is no trusted third party,and the location privacy protection algorithm is done independently on the client side,which enhances the users' privacy protection.Among them,we use the location ambiguity technology to translate the users' real position randomly to generate an alternative fake position,in order to achieve the purpose of protecting the location privacy;we use k-anonymity technology to confuse the user's real interest with other(k-1)dummy ones to achieve the purpose of protecting the query privacy.2.Design the location recommendation algorithm.Based on the history query data processed by the privacy protection sent by the client,we construct the user-interest matrix,and use the regularized matrix factorization collaborative filtering model to avoid the occurrence of the over-fitting of the recommended results,and based on the consideration that the fake location sent by users is around the real location and the visit to location usually presents the phenomenon of geographical aggregation,we use K-means clustering algorithm to mine the users' hot activity area to make personalized recommendation.At last,in the client based on the distance dimension by secondary filtering to provide users with more accurate location recommendation results.3.Analysis of experimental results.A series of simulation experiments are carried out on the real data set collected from the Foursquare website.The experimental results show that the accuracy of the results of the scheme has been declining with the increasing value of anonymous k value.However,when the value of anonymous k is less than 8,the precision of the proposed scheme is higher than that of the conventional collaborative filtering recommendation scheme,and the safety analysis and usability analysis of the privacy protection are proved.The above analysis shows that the proposed scheme can guarantee the effectiveness of personalized recommendation and protect the privacy of users' location information.