Back to search

Hellenic Open University

A service - oriented framework for the specification, deployment, execution, benchmarking, and prediction of performance of scalable privacy - preserving record linkage techniques

Abstract

dc:description

At the dawn of a new era of computing and the growth of big data, information integration is more important than ever before. Large organizations, such as corporations, health providers, public sector agencies, or research institutes, integrate their data in order to generate insightful data analytics. This data integration and analysis enables these organizations to make certain decisions toward deriving better business outcomes. Record linkage, also known as entity resolution or data matching, is the process of resolving whether two records that belong to disparate data sets, refer to the same real-world entity. The lack of common identifiers and the presence of typos and inconsistencies in the data, render the process of record linkage very challenging and mandatory for organizations which need to integrate their records. When data is deemed as private, then specialized techniques are employed that perform Privacy-Preserving Record Linkage (PPRL) in a secure manner, by respecting the privacy of the individuals who are represented by those records. For instance, in the public sector, there are databases which contain records that refer to the same citizen holding outdated information. Although, there is an urgent need for integration, the lack of common identifiers poses significant impediments in the linkage process.Due to the large volumes of records contained in the data sets, core component of PPRL is the blocking phase, in which records are inserted into overlapping blocks and, then, are compared with one another. The purpose of the blocking phase is to formulate as many as possible matching record pairs. The blocking methods proposed thus far in the literature apply empirical techniques, which, given the particularities and technical characteristics of the data sets at hand, produce arbitrary results. This dissertation is the first to provide theoretical guarantees of completeness in the generated result set of the PPRL process, by introducing a randomized framework that allows for easy tuning of its configuration. Its flexibility lies in the fact that one can specify the level of its performance, with respect to the completeness of the results, taking into account multiple factors, such as: the urgency of the problem being solved, the desired response time, or the criticalness of the results completeness. Additionally, we enhance its main functionality, by providing certain extensions, and illustrate its applicability to both offline and online settings. The framework has been materialized by a prototype that is freely available so that it can be used by practitioners and researchers in their tasks.This dissertation is divided into several chapters; we first introduce the core of our framework and its capabilities, and, then, we present its several extensions, such as the integration with the map/reduce paradigm for scaling up large volumes of records, or the add-on for performing PPRL using numeric values. In each of these chapters, we report on an extensive evaluation of the application of the constituent methods with real data sets, which illustrates that they outperform existing approaches.

Degree

thesis:*
Grantor dc:publisher
Hellenic Open University
Year dc:date
2016

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Karapiperis, Dimitrios
  • Καραπιπέρης, Δημήτριος

Subjects

dc:subject × 8

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/39789

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Karapiperis, Dimitrios; Καραπιπέρης, Δημήτριος. A service - oriented framework for the specification, deployment, execution, benchmarking, and prediction of performance of scalable privacy - preserving record linkage techniques. Hellenic Open University, 2016. http://hdl.handle.net/10442/hedi/39789