Research On The Privacy-preserving Recommendation System In Social Networks

With the rapid development of Internet,social networks has become an important way for online communication.Lots of interactions among the users bring the problem of social overhead and information overhead.The social overhead and information overhead make the users unable to search the information correctly and efficiently.The recommendation system is considered to be an effective tool to solve these problems.By predicting items the users may like,recommendation system can effectively help the users to filter useless information,and provide personalized recommendation services.Thus,it can improve recommendation accuracy and user experience.However,the recommendation service provider may be untrusted,and he may modify users' data or leak data to a third party.Therefore,the users face the risk of privacy disclosure when they enjoy the personalized recommendation services.Thus,it is important to design a privacy-preserving recommendation system.Currently,it is one of the hot topics to design a privacy-preserving recommendation system based on cryptographic techniques in the academic community.Although all existing cryptography-based privacy-preserving recommendation systems can compute predicted ratings for users under ciphertexts,there are other problems such as complicated scheme,heavy computation cost and communication cost,which leads to impractical schemes.Based on the research of existing schemes,this thesis proposes privacy-preserving recommendation system schemes in social networks.The main result has been accepted by the international journal International Journal of Embedded Systems(EI source journal).The main contributions of this thesis are as follows:(1)Based on the proxy re-encryption technique,we propose a novel privacy-preserving recommendation system in social networks.Compared with the existing schemes,our scheme not only supports friends offline in the process of computing predicted ratings,but also addresses the problem that the user encrypts a rating with different public keys when sharing it with different users.Therefore,the proposed scheme can reduce the computational cost for the user.The security analysis shows that the proposed scheme can guarantee the confidentiality of user's rating under the honest-but-curious model.(2)The proxy re-encryption proposed by Blaze et al.will disclose the message “0”.Thus,we propose a privacy-preserving recommendation scheme by using the discrete logarithmic enhanced technique.The proposed scheme can protect the message “0”,namely,it can protect the information about the item whether rated by the user or not.Meanwhile,it achieves better security at the cost of spending a little extra computational cost.(3)We not only analyze the computational cost and communication cost of our scheme in theory,but also simulate the proposed scheme based on the Movie Lens dataset.The results show that our scheme can effectively reduce the computational cost and communication cost compared with existing schemes in theory and practice.Thus,our scheme can be better applied into practice.