Από τα στοιχειοσύνολα και τις τροχιές κινούμενων αντικειμένων στις υπηρεσίες θέσης: μια προσέγγιση προστασίας της ιδιωτικότητας μέσω απόκρυψης γνώσης

Abstract

Since its inception in 2000, privacy preserving data mining has gained increasing popularity in the data mining community. This line of research can be primarily attributed to the growing concern of individuals and organizations regarding the violation of privacy in the mining of their data by the existing data mining technology. As a result, a whole new body of research emerged, providing novel approaches for the mining of data, while prohibiting the leakage of private and sensitive information. In this dissertation, we investigate a set of methodologies for the preservation of privacy in different contexts, data domains, and application scenarios. The dissertation is divided into three parts: In the first part, we investigate methodologies for the hiding of sensitive knowledge in the form of association rules, extracted from large transactional databases. Our research in this area led to the proposal of a new direction of approaches that guarantee optimality in the hiding solution by introducing the least side–effects and by causing minimal distortion to the original data. In the second part of the dissertation, we extend the applicability area of association rule hiding (and, generally, semantic knowledge hiding) by applying similar techniques for the hiding of various types of temporally and spatially annotated data. Trajectory data leads to a more powerful kind of knowledge than the one that is produced by association rules and transactional data. First, there exists an ordering of the elements in the trajectories, thus we are faced with sequential data. Second, since mobility data is of spatiotemporal nature, both the spatial and the temporal dimension of the data require special handling. Our contribution in this part of the thesis is a privacy aware trajectory tracking query engine, which enables a set of untrusted end users to query trajectory data that is stored in a database. The proposed engine guarantees that the information that is returned to the end users as part of a query, does not violate the privacy of the users whose movement in recorded in the database. Moreover, special care is taken to shield the database against different types of attacks. In the last part of the dissertation, we go even further by investigating trajectory hiding in a dynamic, real–time environment. In particular, we consider real–time (rather than historical) traffic data produced by a number of users who are on the move. Each user is equipped with a mobile device that regularly transmits his/her location to a given station. Users subscribe to a set of services that depend on user location and are free to request any service at any time. The research problem we investigate deals with the offering of the requested services in a privacy aware manner, so that the identity of the requester is adequately protected. We have contributed a set of methodologies that protect the trajectory of the requester from the time of request, until the provision of the requested location based service. To shield the identity of the requester, we use his/her historical movement traces, as collected by the system, to build mobility patterns that capture the locations and times where his/her privacy is in danger. Then, we derive a set of privacy methodologies both for unconstrained and network–constrained user movement, which generalize (anonymize) the actual location of request as well as the subsequent locations of the user until the completion of the service. Moreover, we present PLOT, the first contributed open–ended toolbox for the offering of privacy in location based services.