This thesis revolves around the topic of quality and privacy in user-generated Big Data, focusing on the problems regarding privacy-preserving continuous data publishing that we summarize below.
We reviewed the existing literature regarding methods on privacy-preserving continuous data publishing, spanning the past two decades, while elaborating on data correlation. Our contributions are:
\item We categorized the works that we reviewed based on their input data in either \emph{microdata} or \emph{statistical data} and further separated each data category based on its observation span in \emph{finite} and \emph{infinite}.
\item We identified the privacy protection algorithms and techniques that each work is using, focusing on feature like the privacy method, operation, attack, and protection level.
\item We organized the reviewed literature in a tabular form to allow for a more efficient indexation of the related works, using a number of relevant features.
We presented ($\varepsilon$, $L$)-\emph{{\thething} privacy}, a novel privacy notion that is based on differential privacy allowing for better data utility in the presence of significant events. Our contributions are:
\item We introduced the notion of \emph{{\thething} events} in privacy-preserving data publishing and differentiated events between regular and events that a user might consider more privacy-sensitive (\emph{\thethings}).
% \item We proposed and formally defined a novel privacy notion, ($\varepsilon$, $L$)-\emph{{\thething} privacy}.
\item We studied {\thething} privacy under temporal correlation, which is inherent in time series, and observed the effect of {\thethings} on the temporal privacy loss propagation.
\item We designed an additional differential privacy mechanism, based on the exponential mechanism, for providing
% additional
protection to the temporal position of the {\thethings}
% \kat{what is the name of the mechanism? how do you quantify 'additional' ?}
by generating dummy {\thething} set options.
\item We experimentally evaluated our proposal on real and synthetic data sets, and compared {\thething} privacy schemes with event- and user-level privacy protection, for different {\thething} percentages.
% \kat{what are the conclusions that show the quality/benefits of the proposed solution?}
We showed that our methodology can provide adequate differential privacy guarantees while achieving better data utility than the user-level scheme.
