Research On Anonymity Models And Algorithms Of Trajectory Privacy Preservation To Resist Location Linkage Attack

Trajectory data play an increasing important role in intelligent transportation, urban planning, individual behavior pattern analysis, etc. Therefore, many organizations are collecting and publishing trajectory data. However, publishing original trajectory data will threat individuals’privacy. With the increasing of social problems inducing by trajectory privacy disclosure, research on trajectory privacy preservation is becoming a hot topic in mobile computing area.(k,δ)-anonymity model is an effective method to preserve trajectory privacy, but the model is vulnerable to location linkage attack. Therefore, this thesis proposes some improved anonymity models and algorithms to resist spatio-temporal point linkage attack and to resist sub-trajectory linkage attack respectively. The main contributions are as follows:(1) A (k,δ,l)-anonymity model to resist spatio-temporal point linkage attack is proposed. The (k,δ)-anonymity is an effective model for trajectory privacy preservation, but it is vulnerable to spatio-temporal point linkage attack. To address the problem, this paper proposes a (k,δ,l)-anonymity model. The (k,δ,l)-anonymity model makes sure that each published trajectory is indistinguishable with at least other k-1 trajectories within a 8-radius cylinder, and each spatio-temporal point of the center trajectory of the cylinder is passed by at least l trajectories. This thesis also proposes an AGG-NWA algorithm to implement the (k,δ,l)-anonymity model. Experimental results show that the proposed model can reserve similar utility as the (k,δ )-anonymity model, and outperforms the (k,δ)-anonymity with respect to security.(2) A (k,δ,l)m-anonymity model and (k,δ, s)-anonymity model to resist sub-trajectory linkage attack are proposed. The (k,δ,l)m-anonymity model is directed against continuous sub-trajectory linkage attack. The model requires that each published trajectory is indistinguishable with at least other k-1 trajectories within δ-radius cylinder, and to any continuous sub-trajectory whose length is less than m, there exist other l-1 trajectories containing the same continuous sub-trajectory. The (k,δ,a)-anonymity model is directed against sub-trajectory linkage attack. The model requires that each published trajectory is indistinguishable with at least other k-1 trajectories within a δ-radius cylinder, and to any trajectory, there exist at least s-1 other trajectories which are same with the trajectory. This thesis also proposes m-NWA algorithm and Center-NWA algorithm to realize the (k, δ, l)m-anonymity model and the (k, δ, s)-anonymity model respectively. Experimental results show that the (k, δ,l)m-anonymity model and (k, δ,s)-anonymity model can reserve similar utility as the (k, δ)-anonymity model, and outperforms the (k,δ)-anonymity model and the (k, δ,l)-anonymity model with respect to security.