Abstract
We present a SR-2 method to extract overlapping community structures from a human relationship network. It is a method of spectral graph analysis and is a variation of a SR method, which can extract a set of densely connected nodes from a network, changed for the more detailed extraction of co-occurrence networks. To investigate the characteristics of a SR method, a SR-2 method, and a k-community method, we visualized the results of extraction by their methods and evaluated the performance of extraction on a small human relationship network extracted from Web space. And we analyzed the distribution of node set sizes on a large co-authorship network. These results show that a SR-2 method can extract adequate community structures corresponding to real human communities.
