Commute times in dense graphs

Abstract

In this paper, we introduce the approach of graph densification as a means of preconditioning spectral clustering. After motivating the need of densification, we review the fundamentals of graph densifiers based on cut similarity and then analyze their associated optimization problems. In our experiments we analyze the implications of densification in the estimation of commute times.