Many domains of human activity are characterized by directed graphs: interactions in social networks, citation networks, hyperlinked domains like the web, trade relations between companies, patterns of air travel. Clustering is a general technique to discover global structure in such networks of local interactions. Presently, networks with asymmetric relations are first symmetrized then clustered by methods applicable to undirected graphs. This proposal shows that symmetrization often loses the information about structure and propose to develop a theoretical framework and algorithms for clustering of directed networks. I will approach the task using the random walks view that I developed and which has proved successful at clustering in undirected graphs.
