Research On Privacy-preserving Of Check-in Location Data Based On Local Differential Privacy

With the popularization of mobile Internet and artificial intelligence,a lot of emerging applications,such as Internet of things(Io T),vehicular networks and 5G technology,have rapidly proliferated.All kinds of intelligent terminals are connected to the network,resulting in a large amount of data.The potential value of deep mining data has become a research hotpot.In the meanwhile,the advancement of location technology and wide application of location based services enables the collection of location data to be more accurate and frequent.The analysis and utilization of location data would greatly benefit commercial applications,such as,market analysis,directional advertising and urban planning.Although most published datasets will provide anonymity mechanism in the exiting location data privacy preserving mechanism,the collection of large amounts of data can pose significant privacy risks to both users and data collectors.Based on background knowledge,the attackers can easily infer their sensitive information,such as,occupation and home address,etc.Notice that the users are willing to share their own location information only in a reliable local environment.For example,users' real data can be protected from leakage by properly disturbance before leaving their own devices.Therefore,the privacy preserving in local environment has become vitally critical.Motivated by the above,this thesis investigates an effective method to solve the privacy preserving problem of location data in social networks,referred to as local differential privacy method.Different from the centralized differential privacy method that needs a trusted data collector,the local differential privacy puts the credit boundary at the users' place.Each user disturbs individual data locally,which can gain the user's recognition for the local environment.From the data perspective,this thesis focuses on location data composing of time information and location coordinates.The privacy preserving methods with different emphasis for these two types of data are studied,respectively.In particular,for time information,they are disturbed by the mean value estimation algorithm based on Laplace mechanism.Personalized parameters are also introduced to meet the different degrees of privacy preserving of different users.The goal of mean value estimation algorithm based on Laplace mechanism lies in knowing the user's time preference by adding a large amount of user data,which can eliminate the noise caused by Laplace mechanism.On the other hand,for location coordinates,they are disturbed by the frequency estimation algorithm based on the random response mechanism,aiming to obtain a more accurate user distribution by superimposing a large amount of user,which can eliminate the interference caused by the random response mechanism.Two local differential privacy-based algorithms proposed in this thesis can prevent malicious attacks by attackers with strong background knowledge.In the simulation section,for two types of data,namely,time information and location coordinates,a large number of experiments are carried out on both real data sets and synthetic data sets.The results show that the designed algorithms can strengthen the privacy preserving of users,and ensure the availability of the user data after disturbance,guaranteeing the data validity.To sum up,the designed privacy preserving mechanism for location data in this thesis is feasible,which has a certain theoretical significance for the privacy preserving of location data.