The Research On Trajectory Synthesis-based Privacy Protection Technology

With the rapid development of mobile computing technology,location-based services(LBS)have become an indispensable basic service in contemporary social life.However,while enjoying the LBS,users face the risk of trajectory privacy leakage.Currently,trajectory synthesis,as the main technology for trajectory privacy protection,only considers temporal and spatial attributes of trajectory data but ignores semantic and social attributes.However,the semantic and social attributes of trajectory data are inherently related to the trajectory of human activity.This connection brings new challenges to trajectory privacy protection,making the existing trajectory privacy protection technology unable to resist a variety of privacy attacks and fail to ensure the data utility.For this reason,this thesis considers the issue of trajectory privacy protection from the perspective of privacy attacks.This thesis first studies the social relationship inference attack in trajectory data,aiming at revealing the influence of the spatial,temporal,and semantic features inherent in trajectory data on social relationship privacy leakage,which can provide theoretical support for trajectory privacy protection.On this basis,this thesis integrates social attributes into the technical framework of trajectory privacy protection,and combines spatial,temporal,and semantic attributes to develop a privacy-preserving trajectory synthesis model combining multiple attributes.The research of this thesis has important theoretical significance and reference value for the analysis and design of the LBS system.Specifically,the research work of this thesis is as follows:(1)This thesis proposes a social relationship inference model based on fusing spatial and temporal features.The trajectory data implies the user's social-related information.Existing social relationship inference models only analyze the influence of spatial attributes of trajectories on the social relationship inference.However,it is observed that the time context of the co-location has a significant influence on inferring social relationships.Therefore,this thesis proposes a social relationship inference attack based on fusing spatial and temporal features.This thesis analyzed the influence of spatial and temporal attributes in trajectory data on privacy leakage of social relationships through feature extraction,metric formalization,and feature fusion.This research provides a new dimension,i.e.,the social attribute,for trajectory privacy protection.(2)This thesis develops a trajectory k-anonymity algorithm based on social attributes.Based on the previous research,it is found that existing methods only consider the spatial and temporal attributes,and ignore the impact of social relationships on users'mobility behaviors,thus failing to resist de-anonymization attacks based on social attributes.Therefore,this thesis proposes a trajectory k-anonymity algorithm based on social attributes.The algorithm first models the target user's mobility from temporal,spatial,and social dimensions.Then,it samples the locations from trajectories that show similar three-dimensional mobility.Finally,the model uses the sampling locations to synthesize a set of fake trajectories,which can provide the privacy guarantee of k-anonymity for the target user.It is the first algorithm that introduces the social attributes of trajectory data into trajectory privacy protection,which ensures the data utility of synthetic trajectories while achieving privacy protection.(3)This thesis designs a semantic-aware social relationship inference attack.Existing related works mainly consider the spatial and temporal attributes of trajectory data.They estimate the strength of social connections between users by calculating explicit statistical information such as the co-location frequency and the stay time of user pairs.It will not only cause a high computation and storage overhead but also bring obvious estimation errors due to ignoring the semantic features of user mobility behaviors.To this end,this thesis proposes a semantic-aware social relationship inference attack model.The model learns a mobility feature vector for each user by deriving the probability distribution of the visit purpose of the user's stopover locations and combines the spatial and temporal attributes of the trajectories.Finally,to calculate user similarity,binary coding technology is used to compress the feature vector.The attack model has high accuracy,strong scalability,and is robust to the data size of users.(4)This thesis proposes a trajectory synthesis algorithm with utility awareness and dual privacy guarantees.Existing trajectory synthesis-based privacy protection technologies are failing to resist multiple attack methods at the same time,and unable to balance data utility and privacy protection.For this reason,this thesis proposes a trajectory privacy release model with utility awareness and dual privacy guarantees.First,to accurately and efficiently model human mobility behavior,this thesis designs an adaptive space-time discrete grid structure and constructs a time-dependent Markov model based on this grid structure.Secondly,to evaluate data utility and privacy protection,the space-time distance,semantic distance,and social distance are proposed to measure the correlation between location points in the same trajectory,the correlation between the trajectories of the same user,and the correlation between the trajectories of different users.On the basis of the metrics,this thesis models the trade-off between privacy preservation and data utility as a bi-level program,and proposes a method to solve the problem,thereby obtains synthetic trajectories that meet the requirements.The synthetic trajectories generated by this model can resist a variety of trajectory de-anonymization attacks and social relationship inference attacks,while also ensuring the data utility.