A probabilistic inference attack on suppressed social networks

Official URL: http://dx.doi.org/10.1109/ASONAM.2012.132

Social Networks (SNs) are widely used by internet users to share personal information, which also raises every privacy concern. Hence most service providers offer various preference-based privacy policies, allowing users to suppress any information under their accounts in case they do not wish to share it with public. In this paper, we show that such policies are not sufficient to provide privacy mainly because they do not allow users to control data belonging to other users they are linked with. We show experimentally that one can predict a suppressed boolean label (e.g, being rich or having voted for a specific political party) of a node from other released information in neighboring nodes when there is a known correlation between the links and the label.