Show me how you move and I will tell you who you are

Reviewer: Jesus Villadangos

Sanitization algorithms and methods for preserving geoprivacy are evaluated in this paper. It is motivated by actual situations where a large amount of geolocation information is available, and the analysis of it could compromise personal information by identifying important places, predicting the movement patterns of an individual, or discovering social relations. Sanitization algorithms or techniques such as randomization, aggregation, sampling, and perturbation introduce some uncertainty and remove some information from geolocated datasets, in order to increase the privacy of individuals whose movements are contained in the dataset. An interesting contribution is the summary of existing sanitization algorithms. More importantly, as sanitization leads to a loss of information, the authors point out the requirement to define a utility function to evaluate the impact of the sanitization mechanism, in order to extract valid conclusions from the modified (sanitized) geolocated information dataset. The authors present some results by using their geoprivacy-enhancing toolkit (GEPETO) software, which can be used to visualize, sanitize, attack, and measure the utility of a particular geolocated dataset. Often, there is a tradeoff between the utility of the sanitization process and the privacy protection of individuals. Gambs et al. provide the results of the evaluation of sanitization methods and inference attacks on geolocated data. In particular, they describe some clustering algorithms and heuristics that can be used as inference attacks and evaluate their efficiency for the identification of points of interest (POI), even after the application of sanitization mechanisms such as sampling and perturbation. Of particular interest is the proposal to define utility functions to evaluate sanitization methods, taking into account their effect on practical applications. The paper analyzes the effect on POI identification. The paper also opens an interesting line of research: sanitization methods could have a different impact on different location-based services. For example, geomarketing applications require a different precision level than promotional offers, and their sanitization will have different effects. The design of algorithms meant to extract conclusions from geolocated datasets has to take into account the application scope of algorithms, and utility functions must be defined as a way to compare such algorithms. Such functions have to be designed for each practical application or group of similar applications. Some of them focus on exact locations, while others are valid in a larger region around the exact position. Modifying personal information might be enough for the former, while perturbation might be enough for the latter. Online Computing Reviews Service