Research On Multiple-Strategies Of Differential Privacy Protection Based On Sparse Tensor Factorization

At present,we have entered the era of the 5G Internet of Things.The 5G network has the characteristics of large capacity and fast transmission speed.Therefore,thousands of devices(including mobile phones,computers,watches,AR/VR glasses,etc.)are connected through the Internet of Things.Fast-selling entertainment forms such as micro-video spread rapidly on the Internet.The huge number of users and the complex IoT service system also generate High-Order,High-Dimension,and Sparse Tensor(HOHDST)data,which contains users’private information.HOHDST usually uses Sparse Tensor Factorization(STF)technology to discover rules,obtain complete data,and perform accurate analysis in real-time.Therefore,in order to solve the problem of privacy protection in the process of network connection,this paper studies the underlying basic traffic data and application business layer data involved in the network connection.This paper proposes a Multiple-strategies Differential Privacy framework on STF(MDPSTF)for network traffic recovery,which jointly considers the underlying data recovery and privacy security issues,and is used for data recovery of underlying basic traffic.and privacy protection.The framework includes three privacy mechanisms:ε-differential privacy,centralized differential privacy,and localized differential privacy,respectively dealing with the application scenarios of trusted,untrusted,and local clients of third-party servers.Finally,the theoretical proof of the privacy bounds of each mechanism is given.This paper conducts experiments on two real network traffic datasets,Abilene and GèANT.The experimental results show that MDPSTF can meet the requirements of different degrees of privacy protection,and has a high recovery accuracy for HOHDST.This paper proposes a Multi-modal Micro-video Recommendation System based on STF and Differential Privacy(MRTFDP).Considering the security issues of application business layer data,it is used for micro-video recommendations and user points of interest privacy protection.Pay attention to the current popular short-video applications,because the micro-video itself has multi-modal characteristics,and also contains information such as personal attributes and points of interest,so micro-video recommendation needs to protect the user’s private information during the recommendation analysis process.MRTFDP consists of three parts:bilinear Tucker fusion,differential privacy,and neural collaborative filtering,and the theoretical proof of differential privacy are given at the same time.Extensive experiments on publicly available datasets Movielens and Tiktok show that the MRTFDP model proposed in this paper has high recommendation accuracy for micro-video datasets and can protect user privacy information.To sum up,the data content involved in the network connection is closely related to the user’s privacy.In order to protect the user’s privacy and security,it is necessary to pay close attention to the privacy information contained in this type of data and protect it.People’s increasingly beautiful life needs to rely on the development of the network in the new era,but at the same time,it must also take into account the comprehensive development requirements of network connection and privacy protection.