Pri-View: Privacy-Preserving Views for Data Analysis and Publication

• Autor(a) principal: Costa, João Miguel Pereira
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/138793
• topic: Relational Databases; Statistical Data; Differential Privacy; Views; Proxy; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Data is being generated and processed at an unprecedented scale. Statistical data analysis is in high demand, with many organizations using it for a broad range of interests, from researching to guiding business decisions. However, this massive generation of data raises privacy concerns, as most of this data contains sensitive information about individuals. In turn, several regulations have emerged to give people more control over their data, such as the European General Data Protection Regulation. For organizations, the challenge is how to analyze and publish data without compro- mising an individual’s privacy. In the context of Relational Databases, they still lack features for this, with solutions involving manually removing identifying information from the data or only allowing certain aggregate queries. However, these solutions can be susceptible to attacks and do not provide strong privacy guarantees. In this thesis, we propose to explore a solution to address the challenge of privately analyzing and publishing data on Relational Databases. To this end, we present a new type of Views - privacy-preserving views - which allow for computing statistical aggre- gations on data while preserving privacy. We focus our studies on Differential Privacy, a recent mathematical definition of privacy, and explore how to turn common aggregation functions into their private counterparts. We present our solution in two parts. In the first part, we present a solution to create privacy-preserving views for a specific database, namely PostgreSQL. In the second part, we present the design and implementation of a database proxy, which supports any SQL database and produces private statistical results. The experimental results show that our proposed solutions can achieve balanced performance - views containing count functions perform better than views containing other functions. They also show that both solutions are capable of providing accurate privacy-preserving data for large databases and sample sizes.