Abstract
dc:descriptionNowadays, the vast adoption of technology, has enabled users to share many aspects of their personal lives. Social networks, GPS-enabled smartphones and portable health sensors are only some examples of data acquisition technologies. Although, the identity of users is removed prior to data publishing, data can still lead to privacy breaches. For instance, a person's identity, home address, financial data or even the medical history or lifestyle preferences may be identified. On the other hand, such data are extremely useful in several studies, such as demographical studies, human behaviour analysis and medical studies. Thus, transforming these data to protect users, in a way that the data remain meaningful, is crucial.Several anonymity methods have been proposed for protecting data, such as k-anonymity, l-diversity, k^m-anonymity and differential privacy. These methods can be applied either in relational datasets (e.g., datasets holding the demographics of a user) or in set-valued datasets (e.g., datasets holding the items bought by a person). As anonymity in relational data is widely studied, we focus our research on anonymity of set-valued datasets such as medical and trajectory datasets. We also formalize the notion of RT-datasets, which are datasets holding both relational and set-valued data (e.g., the demographics and diagnosis of patients).For trajectories datasets, we propose several methods to achieve anonymity and minimize the loss in data utility after anonymization. Initially, we present an approach adapting k^m-anonymity to trajectory data using distance-based generalization. We also develop an effective and efficient anonymization algorithm, which is based on the apriori principle. Then, we propose two more algorithms, preserving different data characteristics, including location distance and semantic similarity, as well as user-specified utility requirements, which must be satisfied to ensure that the released data can be meaningfully analyzed. We also propose a novel framework for anonymizing trajectory data that prevents the disclosure of both identity and sensitive location information, while retaining data utility. Our framework involves: (i) selecting similar trajectories, by employing Z-ordering or data projections on frequent subtrajectories, (ii) organizing the selected trajectories into carefully constructed clusters, and (iii) anonymizing each cluster separately. Finally, we propose four methods, offering protection from attackers controlling a set of disjoint locations. Our methods, employ suppression and splitting, transform the trajectories, and prevent such attackers from inferring locations unknown to them with high certainty. Our methods for the anonymization of trajectory datasets cover a wide range of attacks and a offer broad type of anonymity protection. As we outline in our experimental analysis, our methods protect the privacy of users and are efficient. At the same time they minimally distort the original dataset, thus, they allow useful and meaningful analysis and mining. Next, we focus on RT-datasets. Preserving the privacy and utility on RT-datasets is challenging, as it requires (i) guarding against malicious users, whose knowledge spans both attribute types, (ii) minimizing the overall information loss, and (iii) preserving data truthfulness. Existing anonymization techniques are not applicable to such datasets, and the problem cannot be tackled based on popular, multi-objective optimization strategies. Thus, we propose the first approach to address this problem. Based on this approach, we develop two frameworks to offer privacy, with bounded information loss in one attribute type and minimal information loss in the other. To realize each framework, we propose privacy algorithms that effectively preserve data utility, as verified by extensive experiments.We finally present a prototype system, called SECRETA, in order to assist users with limited background knowledge, to use and evaluate anonymity methods. In more detail, SECRETA allows users, to evaluate a specific algorithm, compare multiple algorithms, and combine algorithms for anonymizing datasets. The analysis of the algorithms is performed, in an interactive and progressive way, and results, attribute statistics and various data utility indicators, can be presented graphically.
Degree
thesis:*- Grantor dc:publisher
- University of Peloponesse
- Year dc:date
- 2014
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Poulis, Georgios
- Πουλής, Γεώργιος
Subjects
dc:subject × 8Rights
dc:rights- Statement dc:rights
-
- BY_NC
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/37084