Privacy preference mechanisms in Personal Data Storage (PDS).

In this thesis, we study frameworks for managing user's privacy when disclosing personal data with third parties from Personal Data Storage (PDS). PDS is a secure digital space which allows individuals to collect, store, and give access to third parties. So, PDS has inaugurated a substantial change to the way people can store and control their personal data, by moving from a service-centric to a user-centric model. Up to now, most of the research on PDS has focused on how to enforce user privacy preferences and how to secure data stored into the PDS. In contrast, this thesis aims at designing a Privacy-aware Personal Data Storage (P-PDS), that is, a PDS able to automatically take privacy-aware decisions on third parties access requests in accordance with user preferences. This thesis first demonstrates that semi-supervised learning can be successfully exploited to make a PDS able to automatically decide whether an access request has to be authorized or not. Furthermore, we have revised our first contribution by defining strategies able to obtain good accuracy without requiring too much effort from the user in the training phase. At this aim, we exploit active learning with semi-supervised approach so as to improve the quality of the labeled training dataset. This ables to improve the performance of learning models to predict user privacy preferences correctly. Moreover, in the second part of the thesis we study how user's contextual information play a vital role in term of taking decision of whether to share personal data with third parties. As such, consider that a service provider may provide a request for entertainment service to PDS owner during his/her office hours. In such case, PDS owner may deny this service as he/she is in office. That implies individual would like to accept/deny access requests by considering his/her contextual information. Prior studies on PDS have not considered user's contextual information so far. Moreover, prior research has shown that user privacy preferences may vary based on his/her contextual information. To address this issue, this thesis also focuses to implement a contextual privacy-aware framework for PDS (CP-PDS) which exploits contextual information to build a learning classifier that can predict user privacy preferences under various contextual scenarios. We run several experiments on a realistic dataset and exploiting groups of evaluators. The obtained results show the effectiveness of the proposed approaches.