UC Berkeley

Online social networking services (e.g., Facebook, Twitter, and Blogger) bring new benefits to almost all aspects of our lives. They have completely transformed how we communicate with each other, how we process information, and how we diffuse social influence. However, these social networking services are also plagued by both conventional and emerging threats to security and privacy. For instance, two fundamental security risks are 1) users’ accounts are compromised by attackers or get lost and 2) attackers create massive fake (or Sybil) accounts to launch various malicious activities. In this thesis, we first design secure and usable account recovery methods based on users’ trusted friends to recover compromised or lost user accounts. Second, we construct a scalable semi-supervised learning framework, which is based on probabilistic graphical model techniques, to detect Sybil accounts. Third, we demonstrate that diverse private information (e.g., private user demographics and hidden social connections) can be inferred with high accuracies from data that is publicly available on social networking sites, which has implications for the design of privacy-preserving online social networking services.