Social Community Detection from Photo Collections Using Bayesian Overlapping Subspace Clustering

We investigate the discovery of social clusters from consumer photo collections. People's participation in various social activities is the base on which social clusters are formed. The photos that record those social activities can reflect the social structure of people to a certain degree, depending on the extent of coverage of the photos on the social activities. In this paper, we propose to use Bayesian Overlapping Subspace Clustering (BOSC) technique to detect such social structure. We first define a social closeness measurement that takes people's co-appearance in photos, frequency of co-appearances, etc. into account, from which a social distance matrix can be derived. Then the BOSC is applied to this distance matrix for community detection. BOSC possesses two merits fitting well with social community context: One is that it allows overlapping clusters, i.e., one data item can be assigned with multiple memberships. The other is that it can distinguish insignificant individuals and exclude those from the cluster formation. The experiment results demonstrate that compared with partition-based clustering approach, this technique can reveal more sensible community structure.