SOUPS: Feasibility of Structural Network Clustering for Group-Based Privacy Control in Social Networks

Simon Jones presented Feasibility of Structural Network Clustering for Group-Based Privacy Control in Social Networks this week at SOUPS

The researchers accessed the participants friend connections (list of friends), they also looked at connections between their friends. Used list of friends to do a card sorting exercise. One contact per card (“cards” were digital and shown on the computer).

They found six common grouping criteria

• Social circles & cliques
• The strength of their relationship – commonly used to divide other groups into people with strong ties or weak times
• Geographical locations
• Organizational boundaries
• Temporal episodes – For example childhood or undergrad
• Functional roles – People they had met at events

Used the groups created by participants and compared them with groups created by a clustering algorithm. You can read the details of the algorithm in the paper.  Their algorithm was 45% similar with the user created groups.

Had users find a privacy sensitive item and asked them to rank their willingness to share with different contacts in their network. People who were outliers in the social network were more often not shared with. The authors hypothesize that outliers could be used to automatically identify people who users may not want to share sensitive information with.