A Scheme for Activity Trajectory Dataset Publishing with Privacy Preserved

Facilitated by ubiquitous computing techniques and the wide development of location-based service (LBS) and smart card systems, users´ trajectory data, which is usually attached with user activities, are recorded more and more easily. Inappropriate publishing these trajectory data may jeopardize users´ privacy. However, existing works on privacy-preserving trajectory data publishing fail to consider the activity information. In this paper, we propose a scheme under an existing privacy framework for solving the problem that how to preserve users´ privacy when publishing an activity trajectory dataset. In this scheme, it should be ensured that the posterior probability about users´ privacy inferred by adversaries is not too much larger than the prior one. We assume that user behaviors could be described by a Markov model and that the adversary has a knowledge of both the user model and our privacy preserving mechanism. The adversary could infer users´ predefined sensitive information by Bayesian inference. Then we propose a dataset publishing algorithm, PPPAT, which preserves user privacy. It iterates on all the events of a given user to decide whether they could be published or not. Finally, we evaluate our algorithm by conducting experiments on two datasets, compared with several baselines. The results verify the effectiveness of our algorithm and the utility of outputted activity trajectory dataset.