Krylov Subspace Approximation for Local Community Detection

Community detection is an important information mining task in many fields including computer science, social sciences, biology and physics. For increasingly common large network data sets, global community detection is prohibitively expensive, and attention has shifted to methods that mine local communities, i.e. methods that identify all latent members of a particular community from a few labeled seed members. To address this semi-supervised mining task, we propose a local spectral subspace-based community detection method, called LOSP. A sampling technique is first applied around the seeds to significantly reduce the scale of the search. Then we define a family of local spectral subspaces based on Krylov subspaces, and seek a sparse indicator for the target community via an one norm minimization over the Krylov subspace. Variants of LOSP depend on types of random walks with different diffusion speeds, regular random walk or inverse random walk, dimension of the local spectral subspace and steps of diffusions. The effectiveness of the proposed LOSP approach is theoretically analyzed based on Rayleigh quotients, and is experimentally verified on a wide variety of real-world networks across social, production and biological domains, as well as on an extensive set of synthetic LFR benchmark datasets.