Research On Private Data Leakage And Preservation In "Internet+" Era

The rapid development of the Internet has led to the progress of society,the proposal of "Internet +" development direction has brought out many new industries.On the other hand,it has accelerated the research of computer science.New research fields such as 5G networks,big data,artificial intelligence,and blockchain have sprung up.New industries and new technologies have brought more convenience and high-quality services to people's lives,but with the emergence and application of new technologies,many problems arise.One of the most critical problems is the leakage of user private data.In this highly informative social environment,personal data is rapidly circulating in the form of electronic information.At the same time,the rich perception and computing capabilities of personal smart devices make the data about user privacy constantly generated,even spread to the Internet.On the other hand,big data and artificial intelligence are developing rapidly,some data that implies the behavior of individual users can be used to mine fairly accurate private data.These factors combined to form a great hidden danger of user private data leakage.In addition,there is data leakage caused by imperfect network communication technology.These various threats make it urgent to study the leakage of private data.This paper takes the application scenarios of real life as the research back-ground and explores the risk of private data leakage in the practical application of new network technology.In the process of the research,the method of combining hierarchical networks is adopted,from the perspective of physical layer and appli-cation layer that are obviously different and closely related,this paper separately investigates the security risks in physical layer network technology and applica-tion layer network technology,and pertinently puts forward the protection scheme against the threat of privacy leakage.The innovation of this paper is to study the leakage of user private data in four different technical application scenarios.These security risks involve many fusion scenarios such as smart devices,mobile wireless networks,cloud computing,network crowdsourcing,big data,data mining and so on.This paper finds out the basic reason of privacy leakage through theoretical in-depth analysis and then designs a dedicated privacy protection scheme for these privacy leakage problems.The scheme not only considers different scenarios of application technology(such as network communication quality,channel state,etc.)but also considers attackers with different levels of malice(such as semi-honest,completely malicious,etc.),and all the privacy protection schemes proposed in this paper are at the leading level in the current research field,regardless of the execution efficiency or privacy protection capability.Specifically,this paper starts from the following two perspectives.In the study of private data leakage for the physical layer,this paper combines the latest 5G technology development trend and makes a deep research in the two technical fields of multi-input multi-output system and cognitive radio network.Firstly,this paper proposes an attack method which uses malicious tampering train-ing sequence to mislead the base station and steal the private data of other users in multi-input multi-output systems.This kind of attack seriously violates the private data of honest users.Then,this paper proposes an effective defense scheme by using the fuzzy commitment mechanism and the time-effect of channel state information,which can achieve secure channel state information prediction and downlink data transmission.On the other hand,in cognitive radio networks,the dynamic alloca-tion of spectrum benefits from the correct spectrum sensing of the user,but users'spectrum sensing data contain important personal private data,that is,geographic location information.Malicious users or base stations can easily steal other users'private data.This paper proposes a secure data fusion scheme for multiple partic-ipants and splits the private key to defend against untrusted base stations.Finally,zero-knowledge proof and threshold cryptosystem are introduced to defend against malicious attackers and unreliable network environment.This paper illustrates the correctness and security of the above schemes by theoretical proof and experiments in real scenes.In the study of private data leakage for the application layer,this paper selects two more widely used network services and application scenarios.One is the net-work traffic offloading technology which provides support for content services such as network video or network broadcast,the other is to combine rapidly develop-ing deep learning technology to provide network services of deep data mining for clients on the cloud computing platform.This paper proposes a user geographic location attack based on the user's preference data for the offloading station in traffic offloading and verifies that the attack has high accuracy.In order to defend against this kind of private data leakage,this paper designs a streaming data counter which is used to traffic offloading preference data and satisfies the property of differential privacy,based on this,a matching scheme between privacy protected users and offloading stations is proposed.Theoretical analysis proves that the scheme satisfies a very strong differential privacy property.On the other hand,service providers can provide users with deep learning technology services through cloud computing platform,but there is a serious risk of private data leakage in the cloud computing deep learning mode in which users provide training data.The research work in this paper finds the segmentation property of the network structure in the current mainstream deep neural networks.Based on the principle of segmentation network,a new deep learning scheme of the client-server mode is proposed.The advantage of the scheme is that users can achieve deep learning only with limited computing resources.At the same time,to protect the user's training data from being stolen by untrusted servers or inferred from the intermediate results of the training,this paper proposes a data protection scheme that satisfies differential privacy in this new training model,which guarantees the privacy property of users' training data and the accuracy of learning in practical applications.