Identifying User’s ID From Anonymous Mobility Trace Set Via Asynchronous Side Information

With the increase of mobile devices capability and development of mobile social networks, location based service has made a great progress in recent years. To meet the demand of mobile system designing and scientific research, plenty of location trace in-formation has been collected and published. Most public traces apply the anonymous ID and noise to protect the privacy of users. However, there is still hidden location privacy leaking possibility, and the location information is still easily attacked by ad-versary based on a little side information. For example, the Bayesian approach would implement the identifying problem optimally via synchronous side information. This paper analyzes the Bayesian approach further to identify its effect on asynchronous condition. Experiments show that, although the Bayesian approach can obtain the good performance of the synchronization condition, but under the condition of asynchronous side information the accuracy is less than 20%. Therefore, we discusses the scenarios for asynchronous side information, and propose two approaches, Hot Matrix and Mo-bility Spectrogram, to realize the asynchronous condition. We analysis and pointed out that even in asynchronous anonymous information processing conditions are also vulnerable.Our work and contributions are listed as follows:1. Propose asynchronous condition of identify user’s status from anonymous mo-bility trace set, and discussed the possibility of existence of the problem, as well as asynchronous attack. As collectors of publicly available data sets will take protective measures in data collection duration, get synchronization side information will obvi-ously be more difficult. The asynchronous side information is not restricted by the constraints above, thus the access method is more flexible and the side information is more easily obtained by attacker.2. Propose Hot Matrix approach to realize the problem of identifying user’s sta-tus from anonymous mobility traces via asynchronous side information. Hot matrix method makes use of the users’mobility feature of relative concentration in geograph-ical space, and make geographical hot-spots as the mobility pattern of mobility trace. Propose two methods, cosine vector and frequency distribution vector, to calculate the similarity of hot matrix. Experiments show that cosine vector outperforms frequency distribution vector a lot. Accordingly, we propose the scheme of identifying based on hot matrix:Firstly quantity the hot matrix of both side information and public traces, Then measure the similarity of the side information and each public trace based on hot matrix, Finally choose the public trace with highest similarity as the result. Due to our experiments on human, taxi and bus trace sets, this approach achieve accuracy of 65%, 55% and 95% respectively.3. Propose Mobility Spectrogram approach to realize the problem of identifying user’s status from anonymous mobility traces via asynchronous side information, mo-bility spectrogram is analogous to the spectrogram of light and wireless radio. It is quantified by a two-dimensional matrix, which describes the temporal-spatial charac-teristics of node movement. Propose three methods, overlap region, Jaccard similarity coefficient and Hausdorff distance, to calculate the similarity of hot matrix. Exper-iments show that overlap region and Jaccard similarity coefficient outperforms fre-quency distribution vector a lot. Accordingly, we propose the scheme of identifying based on mobility spectrogram:Firstly quantity the hot matrix of both side informa-tion and public traces, Then measure the similarity of the side information and each public trace based on mobility spectrogram, Finally choose the public trace with high-est similarity as the result. Due to our experiments on human, taxi and bus trace sets, this approach achieve accuracy of 70% ,60% and 95% respectively.Above all, this paper proposed the asynchronous condition of identifying from anonymous mobility traces. And to realize the asynchronous condition, two different approaches are proposed, whose performance is much better then existing methods on asynchronous side information.