Research On Privacy-preserving Graph Learning-based Recommendation Scheme

In recent years,with the rapid development of computer equipment,mobile terminal equipment and Internet technology,people have entered the era of big data.The massive amount of information resources not only brings convenience to people,but also causes the problem of information overload.Recommendation systems play a crucial role as effective technologies to address information overload issues.However,most existing recommendation systems require the collection and analysis of a large amount of user privacy data in order to provide comprehensive recommendation services,which poses significant risks to user data security.In order to protect user data privacy,significant progress has been made in the research of recommendation algorithms that support privacy protection.However,existing privacy protection recommendation algorithms cannot meet the increasing demand for privacy recommendation accuracy,and cannot cope with the huge computational overhead caused by the explosive increase in user nodes in recommendation systems.In addition,in practical scenarios,data is mainly generated by a large number of mobile users and distributed on multiple data sources.Existing research focuses more on the security issues in recommendation model calculation when storing data in a centralized manner.For distributed stored data,recommendation services cannot be efficiently performed while ensuring data privacy and security.Therefore,this article discusses the data privacy issues in two scenarios:centralized storage and distributed storage of user data in recommendation systems,based on practical application scenarios,and proposes two privacy protection schemes.The details are as follows:(1)For the cloud computing scenario where data is stored centrally,that is,data is completely outsourced to the cloud server,and the cloud server conducts model training and prediction,this paper proposes SecSimpleRec,a privacy map learning recommendation scheme based on Secure multi-party computation.To address the accuracy issue of existing privacy recommendation schemes,this paper proposes a graph learning based privacy recommendation algorithm,which enables the model to effectively utilize graph topology information and enhance privacy recommendation capabilities.In order to ensure the security of the scheme,the article analyzes the privacy issues in different stages of the recommendation system,and designs a secure computing protocol based on additive secret sharing technology to protect the privacy of the entire process for different stages of the graph learning recommendation model.And it has been proven that the proposed protocol is secure in a semi honest model.(2)In response to the scenario where data is distributed among a large number of mobile users who are unwilling to disclose privacy and wish to receive recommendation services,this paper proposes a privacy protected federated graph learning recommendation scheme PPFedRec to achieve privacy protection and effective recommendation for distributed data.This scheme overcomes the problem that graph structure data cannot be effectively used in the Federated learning scenario,and proposes a global graph perspective construction method.In order to protect the user data privacy in the recommendation process,a security map sampling algorithm is designed,and the statistical distribution of the uploaded model Parametric statistics is analyzed.Using Differential privacy technology,appropriate noise is added to the model parameters in this paper,so that it does not lose too much recommendation accuracy while protecting data privacy.In addition,this solution solves the problem of communication overhead caused by the large number of parameters in existing federated recommendation models.