Computing heat kernel pagerank and a local clustering algorithm

The quantitative ranking of vertices obtained with heat kernel pagerank can be used for local clustering algorithms. We present an efficient local clustering algorithm that finds cuts by performing a sweep over a heat kernel pagerank vector, using the heat kernel pagerank approximation algorithm as a subroutine. Specifically, we show that for a subset S of Cheeger ratio ϕ, many vertices in S may serve as seeds for a heat kernel pagerank vector which will find a cut of conductance O(ϕ).