Research On The Privacy-Preserving Techniques For The Recommender System

In the digital era with the rapid development of information technology,contextaware recommender systems are of increasing popularity to recommend personalized items to users and help them quickly and accurately find their interests from massive amounts of Internet data.However,the recommendation accuracy of a recommender system usually depends on the amount of information provided by users,which involves the protection of user data privacy.In the field of recommender systems,how to ensure user data privacy while remaining high recommendation accuracy is widely considered a challenge.This challenge is embodied in the following two aspects: First,the recommender system is necessary to perform calculations on negative numbers when processing data,but the existing secure operation protocols still have shortcomings in the calculation of signed integers.Second,due to the resource difference between users and cloud servers,users often hope to participate in the computation of cloud servers as little as possible and require to obtain accurate recommendation results under the premise of ensuring that user information is not leaked.Focusing on the above issues,the work of this paper mainly includes the following two aspects:1)Secure operation protocols based on additive secret sharing.This paper improves the existing additive secret sharing scheme,designs new data splitting and recovery methods to support the sharing of signed integers,further proposes secure comparison protocol(SCmp),secure division protocols(SDiv P,SDiv S)and secure square-root extraction protocol(SSqrt),which are support signed integer operations,and utilizes strict security analyses proves the security of the proposed protocols under the semi-honest model.Compared with the state-of-the-art,the proposed SCmp and SDiv P have achieved at least33% performance improvement,and the designed SDiv S has higher security.2)Privacy-preserving protocol for the recommender system.Based on the proposed secure operation protocols,this paper constructs a privacy-preserving protocol for the context-aware recommender system in the cloud environment.This protocol not only supports off-line users but achieves stronger data privacy preservation,compared with the latest work,by further protecting the intermediate data calculated during the system training.This paper analyzes the security of the proposed scheme under the semi-honest model and shows that the proposed scheme can protect user data confidentiality from active adversaries.In addition,the accuracy of the scheme is verified through experiments on real-world datasets.At the same time,the proposed scheme can be highly parallelized and can be effectively deployed in a real-world cloud environment.