ABSTRACT
In part II of my talk , I will discuss how to solve stage 2 of the clustering problem, namely, going back from a continuous solution to a discrete solution. I will review Yu and Shi’s approach, and present a mild generalization. Kunegis et al proposed a way to deal with signed graphs for two clusters (allowing negative weights). I will show that it is actually possible to extend the method of normalized cuts for signed graphs to any number of clusters, and present an application to word clustering developed by Joao Sedoc.
Slides may be found here…