A User-centric and Agent-based Approach to Multi-user Privacy in Online Social Networks

Student thesis: Doctoral ThesisDoctor of Philosophy


Among the issues that arise from services on online platforms such as social networks (OSNs), there is an increasing concern about privacy and data protection. This is exacerbated when looking at multi-user privacy (MP), i.e. when the privacy decisions of an individual impact multiple stakeholders, which is an issue that has gathered little attention so far.

Following an iterative Value Sensitive Design approach and informed by the literature in Privacy and Autonomous Systems, in this thesis I investigate the design of autonomous systems that can effectively support OSNs users manage MP. This investigation culminates in ELVIRA, a user-centric multi-agent architecture that, by engaging in practical reasoning, recommends optimal collaborative solutions to MP conflicts. The optimality of the solutions is measured by considering not only the contextual privacy preferences of all the users involved, but also their moral values. Furthermore, ELVIRA justifies such privacy recommendations by producing tailored explanations, whose format has been investigated and validated with users.

Through software simulations and a user study, I demonstrate how the agent ELVIRA presents a combination of features that enables it to provide a more satisfactory support for users than alternative state-of-the-art approaches for managing MP in OSNs. In particular, ELVIRA’s privacy recommendations are more acceptable across demographics, and ELVIRA’s explanations nudge users to be more respectful of others’ preferences and more appreciative of fair solutions to MP conflicts. Additionally, drawing from evolutionary game theory and simulating a wordof-mouth marketing strategy, I show how ELVIRA could be widely and stably adopted by OSNs users.

Finally, I outline possible extensions of the ELVIRA model, such as the definition of interactive explanations and the management of noncollaborative behaviour.

Date of Award1 Mar 2023
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorJose Such (Supervisor) & Peter McBurney (Supervisor)

Cite this