A personalized and semantic-aware approach for trajectory protection

State-of-the-art privacy protection research often aims to reduce the computational costs required of entire trajectories by typically omitting less significant location information. Considering locations where users frequently stay for a longer duration or frequently visit as stay points, techniques such as location generalization, location deception, location perturbation, $k$-anonymity, cryptography, and the involvement of a trusted third party (TTP for short) are employed to achieve privacy protection at these stay points. Semantic-aware trajectory privacy methods typically either categorize semantic values or use user role differences in locations to establish LBS queries with similar or different semantic types of point of interest (POI for short) to protect users’ semantic privacy. However, techniques such as generalization, deception, and perturbation often yield less accurate results. The $k$-anonymity technique requires handling numerous service requests, cryptography entails significant computational costs, and TTP might become a target for attacks leading to severe privacy breaches. Identifying stay points or user role differences can only be done after the trajectory has been completely established. Classifying semantic values cannot effectively achieve the semantic privacy users require. To address these shortcomings and establish spatial–temporal correlations between trajectories and semantic values, we propose a novel personalized semantic-aware obfuscation scheme (PSAS for short) combined with differential privacy. PSAS utilizes Markov chains to establish spatial–temporal correlations and to predict user movement points to reduce query frequency. This study introduces a novel graph structure to represent semantic relationships, and calculates semantic importance using term frequency-inverse document frequency (TF-IDF for short). By adopting differential privacy, trajectories are added with noise based on different location privacy budgets to protect users’ privacy of locations, POIs, and trajectories. Experimental results demonstrate that PSAS effectively and comprehensively protects trajectory data and semantic privacy without sacrificing quality of service (QoS for short).