Scalable Seed Expansion for Identifying Web Communities

Abstract

We study the problem of identifying Web communities around some seed vertex. In this work, we propose a fast graph algorithm to expand Web communities in a scalable style. Given a seed vertex, our algorithm computes approximate personalized PageRank vectors with better and better approximations, and finds the smallest conductance sets on these vectors as candidate communities in nearly-linear time. At the end, it returns the candidate community with the smallest conductance as the result community. We also define local community profile (LCP) to investigate structural and statistical properties of Web communities in a local range. Theoretical analysis and primary experiments both show the efficiency of the proposed algorithm and the quality of the results.