Most users of online services have unique behavioral or usage patterns. These behavioral patterns can be exploited to identify and track users by using only the observed patterns in the behavior. We study the task of identifying users from statistics of their behavioral patterns. In particular, we focus on the setting in which we are given histograms of users' data collected during two different experiments. We assume that, in the first data set, the users' identities are anonymized or hidden and that, in the second data set, their identities are known. We study the task of identifying the users by matching the histograms of their data in the first data set with the histograms from the second data set. In recent works, the optimal algorithm for this user identification task is introduced. In this paper, we evaluate the effectiveness of this method on three different types of data sets with up to 50 000 users, and in multiple scenarios. Using data sets such as call data records, web browsing histories, and GPS trajectories, we demonstrate that a large fraction of users can be easily identified given only histograms of their data; hence, these histograms can act as users' fingerprints. We also verify that simultaneous identification of users achieves better performance compared with one-by-one user identification. Furthermore, we show that using the optimal method for identification indeed gives higher identification accuracy than the heuristics-based approaches in the practical scenarios. The accuracy obtained under this optimal method can thus be used to quantify the maximum level of user identification that is possible in such settings. We show that the key factors affecting the accuracy of the optimal identification algorithm are the duration of the data collection, the number of users in the anonymized data set, and the resolution of the data set. We also analyze the effectiveness of k-anonymization in resisting user identification attacks on these data sets.

Consider a set of probes, called “agents”, who sample, based on opportunistic contacts, a population moving between a set of discrete locations. An example of such agents are Bluetooth probes that sample the visible Bluetooth devices in a population. Based on the obtained measurements, we construct a parametric statistical model to jointly estimate the total population size (e.g., the number of visible Bluetooth devices) and their spatial density. We evaluate the performance of our estimators by using Bluetooth traces obtained during an open-air event and Wi-Fi traces obtained on a university campus.

We study the problem of privacy-preserving computation of functions of data that belong to users in a social network under the assumption that users are willing to share their private data with trusted friends in the network. We demonstrate that such trust relationships can be exploited to significantly improve the tradeoff between the privacy of users' data and the accuracy of the computation. Under a one-hop trust model we design an algorithm for partitioning the users into circles of trust and develop a differentially private scheme for computing the global function using results of local computations within each circle. We quantify the improvement in the privacy-accuracy tradeoff of our scheme with respect to other mechanisms that do not exploit inter-user trust. We verify the efficiency of our algorithm by implementing it on social networks with up to one million nodes. Applications of our method include surveys, elections, and recommendation systems.

We conduct an experiment where ten attendees of an open-air music festival are acting as Bluetooth probes. We then construct a parametric statistical model to estimate the total number of visible Bluetooth devices in the festival area. By comparing our estimate with ground truth information provided by probes at the entrances of the festival, we show that the total population can be estimated with a surprisingly low error (1.26% in our experiment), given the small number of agents compared to the area of the festival and the fact that they are regular attendees who move randomly. Also, our statistical model can easily be adapted to obtain more detailed estimates, such as the evolution of the population size over time.