Enhanced link clustering with observations on ground truth to discover social circles

Abstract

Along with the development of internet, our personal online social networks become bigger and more jumbled than before, and it is necessary to provide a good way to organize them. Social networking sites allow users to manually categorize their friends into social circles (e.g., ‘list’ on Facebook and Twitter), but it is laborious. The problem of social circles identifying is thus posed on a user’s ego network, while there are currently few efficient as well as effective methods to identify user’s social circles. In this paper, we propose a new method, named enhanced link clustering, for social circles identifying on ego networks. The proposed method integrates node profile and network structure by constructing an edge profile for each edge. Utilization of both node profile and network structure information makes the proposed method more effective. Taking edge similarity instead of node similarity to discriminate nodes into different circles allows us to detect overlapping circles. Moreover, we observe that nodes in one circle appear transitive similarity and some nodes are only densely connected, or share common properties. These observations make the process of edge clustering efficient. Experiments on several real datasets demonstrate that our method is much faster, and also more effective compared with maximization likelihood-like method, which has been proved to dominate most methods.