Feasibility of structural network clustering for group-based privacy control in social networks

Users of social networking sites often want to manage the sharing of information and content with different groups of people based on their differing relationships. However, grouping contacts places a significant configuration burden on the user. Automated approaches to grouping may have the potential to reduce this burden, however, their use remains largely untested. We investigate people's rationales when grouping their contacts for the purpose of controlling their privacy, finding six criteria that they commonly considered. We assess an automated approach to grouping, based on a network clustering algorithm, whose performance may be analogous to the human's use of some of these criteria. We find that the similarity between the groups created by people and those created by the algorithm is correlated with the modularity of their network. We also demonstrate that the particular clustering algorithm, SCAN, which detects hubs and outliers within a network can be beneficial for identifying contacts who are hard to group or for whom privacy preferences are inconsistent with the rest of their group.