Modeling the Risk & Utility of Information Sharing in Social Networks

With the widespread of social networks, the risk of information sharing has become inevitable. Sharing a user's particular information in social networks is an all-or-none decision. Users receiving friendship invitations from others may decide to accept this request and share their information or reject it in which case none of their information will be shared. Access control in social networks is a challenging topic. Social network users would want to determine the optimum level of details at which they share their personal information with other users based on the risk associated with the process. In this paper, we formulate the problem of data sharing in social networks using two different models: (i) a model based on \emph{diffusion kernels}, and (ii) a model based on access control. We show that it is hard to apply the former in practice and explore the latter. We prove that determining the optimal levels of information sharing is an NP-hard problem and propose an approximation algorithm that determines to what extent social network users share their own information. We propose a trust-based model to assess the risk of sharing sensitive information and use it in the proposed algorithm. Moreover, we prove that the algorithm could be solved in polynomial time. Our results rely heavily on adopting the super modularity property of the risk function, which allows us to employ techniques from convex optimization. To evaluate our model, we conduct a user study to collect demographic information of several social networks users and get their perceptions on risk and trust. In addition, through experimental studies on synthetic data, we compare our proposed algorithm with the optimal algorithm both in terms of risk and time. We show that the proposed algorithm is scalable and that the sacrifice in risk is outweighed by the gain in efficiency.