Research On Trajectory Privacy Preserving Methods Based On Anonymization

With the popularity and spread of location-based services,a large number of trajectories of mobile users have been widely collected by service providers.For commercial and academic purposes,these trajectory data sets are often released by service providers to third parties for data analysis and research.If the raw trajectory data is released directly without any privacy preservation,the results of data mining and data analysis may reveal the privacy of users.With the help of big data technology,attackers can obtain a large amount of background knowledge,and make inferences about users’ sensitive information,including life habits,health status,social relationships and so on.Therefore,privacy protection in the trajectory data publishing is of great significance to safeguard users’ personal information security,and has high research value.In this thesis,we propose two anonymization-based trajectory protection methods in data publishing scenarios to address the shortcomings of existing approaches from different perspectives.Specifically,the main research works are list as follows:(1)This thesis proposes a trajectory anonymization method to resist social relationship attack.It is shown that the linkage of spatio-temporal features of trajectories can disclose the social relationships of anonymous users.Traditional kanonymity method only prevents re-identification attack.Existing protection methods only consider weakening the connection between trajectories and social relationships in the spatial dimension.Our method uses a Markov chain to capture the spatiotemporal features of users’ mobility behavior,and defines the social sensitivity among different trajectories by combining Bayes theorem.The social sensitivity is combined with the similarity of trajectories in temporal and spatial attributes to obtain a socialaware trajectory distance.This social-aware distance is used to improve the traditional trajectory k-anonymization algorithm based on clustering,so that the anonymized trajectory dataset is not only resistant to re-identification attack,but also prevents the privacy leakage of users socia l relationships.Two social relationship attack models are used in the experiments.The traditional k-anonymization method and existing protection method are used as comparisons.The results show that the anonymized trajectory data generated by our method can effectively reduce the accuracy of the attack models to infer social relationships.And the privacy-preserving effect is better than the existing methods while showing stable protection performance.(2)This thesis designs a dummy trajectory synthesis method based on location semantic information.Existing trajectory synthesis methods only consider the spatiotemporal attributes of trajectories,ignoring the rationality of locations’ semantic information.The semantic information of location can reveal the purpose of users visit,and attackers can combine it with the background knowledge to identify whether the trajectory is fake or not.Our method uses the users historical trajectory data to synthesize dummy trajectories,first clusters the historical locations based on semantic information,and then converts the real trajectory to the semantic dimension according to the similarity between the real locations and semantic clusters.Then we combine movement direction and location reachability in the real trajectory,sample from the candidate location sets and synthesize dummy trajectories.Finally,a dummy trajectory dataset satisfying k-anonymity requirement is generated to hide the real trajectories.A real trajectory dataset is used in the experiment to analyze the dummy trajectories generated by our method.And our method is compared with two existing dummy trajectory synthesis methods.The experiment results show that compared with the existing methods,the dummy trajectories generated by our method are more reasonable and can retain more features of the real trajectories,which can effectively confuse the attackers and protect user ’s real trajectory.