Rapidly identifying network communities

Members of networks are joined to other members by connections of varying types and often cluster into modules, groups, or communities. Identifying and characterizing these communities is a fundamental problem in network analysis. One way to tackle this problem is to optimize the quality function known as ‘‘modularity’’ over possible divisions of a network. Mark Newman reports a mathematical technique for quickly identifying and analyzing communities that form in large networks, and the algorithm may be useful for studying communities in social, computer, metabolic, and regulatory networks. Newman showed that network modularity can be expressed in terms of a ‘‘modularity matrix,’’ which leads to new formulas that reveal the community structure. The author tested the matrix method on classic and new networks, including social networks, the metabolic network of the worm C. elegans, coauthorships between condensed matter physicists, and networks representing the political leanings of blogs and books. The method was found to deliver more rapid, higher-quality results compared with three other published algorithm methods; in the case of the largest network, a collaboration network of 27,000 physicists, Newman’s algorithm required only 20 min on a modern desktop computer to find community linkages. — P.D.