Learning Social Circles in Ego Networks based on Multi-View Social Graphs

Automatic social circle detection in ego-networks is becoming a fundamentally important task for social network analysis, which can be used for privacy protection or interest group recommendation. So far, most studies focused on how to detect overlapping circles or how to perform detection using both network structure and its node profiles. This paper asks an orthogonal research question: how to detect social circles by leveraging the multiple views of the network structure? As a first step, we crawl ego networks from Twitter to construct multi-view social graphs with six views, including two relationship views, three interaction views and one content view. Then, we apply both standard and our modified multi-view spectral clustering methods on the graph to detect circles. Based on extensive automatic and manual evaluations, we deliver two major findings: first, multi-view clustering methods detect better circles than single-view clustering methods; second, by presuming the sparse social graphs are incomplete, our modified method detects better circles than the standard method that ignores such potential incompleteness. In particular, we believe a direct application of the standard method on a possibly incomplete graph may yield a biased result. By integrating theories from spectral clustering and matrix perturbation, we prove an information-theoretic upper bound for such bias and discuss how it may be affected by the characteristics of the social network.