A Constrained Power Method for Community Detection in Complex Networks

For an undirected complex network made up with vertices and edges, we developed a fast computing algorithm that divides vertices into different groups by maximizing the standard “modularity” measure of the resulting partitions. The algorithm is based on a simple constrained power method which maximizes a quadratic objective function while satisfying given linear constraints. We evaluated the performance of the algorithm and compared it with a number of state-of-the-art solutions. The new algorithm reported both high optimization quality and fast running speed, and thus it provided a practical tool for community detection and network structure analysis.