Protecting Location Privacy With Differential Privacy Under Temporal Correlations

With the development of positioning technologies and popularity of mobile Internet and smartphones,location-based applications have permeated into our daily life,such as location-based points of interest searching,location-based games,location-based commerce and location-based social networks,etc.Location data generated by mobile users will bring great commercial value for society.However,the user's location belongs to personal privacy,and abusing location data will bring great privacy risks to users.Therefore,the problems of location privacy preservation have been widely concerned by academia and industry.Traditional location preservation methods only consider static scenarios or perturb the location at single timestamps without considering the temporal correlations of a moving user,so that these methods are vulnerable to various temporal correlation inference attacks.In order to solve the temporal correlation inference attack problems,a differential privacy protection method based on -location set is proposed,which adopts the planar isotropic mechanism(PIM).However,the PIM-based approach has two shortcomings as follows :(1)It cannot satisfy the real-time location protection requirements because of its low efficiency;(2)The construction of state transition matrix does not consider the moving mode of users,which will disturb location data largely,so as to decrease users' location service experience.Therefore,we investigate new location privacy protection methods under temporal correlations.The main work includes:(1)We propose a differential privacy location protection method based on Staircase mechanism(SM)and -location set.The implementation of planar isotropic mechanism(PIM)requires two convex hull transformations and one heterogeneous transformation,which leads to low efficiency of the method.At the same time,the implementation of heterogeneous transformation relies on a strict invertible matrix T,so it is difficult to apply to scenes with real-time requirements.To this end,we use the Staircase mechanism to replace PIM.Experiments comparing the size of the -location set,drift ratio,distance,and time efficiency all show that the Staircase mechanism is more efficient than existing approaches with the same privacy budget and generate less noise,increasing the usefulness of location data.(2)We propose a dynamic state transition matrix construction method.The existing construction methods of state matrix do not consider the moving modes of users.Therefore,we first segment the location data according to the user's moving modes,and construct the state transition matrix under each data set respectively.The difference between the transfer matrices is then measured using two indices,namely the F-norm and the matrix heat map.Finally,the experiment shows that considering the influence of user moving modes on the construction of state transition matrix is necessary.