Overlapping community search in very large graphs

Abstract

"In this master thesis we present a novel approach to finding communities in large graphs. Our method finds the overlapped and hierarchical structure of communities efficiently, outperforming previous proposals. We propose a new objective function that allows to evaluate the quality of a community naturally including nodes shared by other communities. This is achieved by implicitly mapping the nodes of the graph in a vectorial space, using as a basis a construction presented by Lóvasz in 1979. We present and analyse several algorithms to decompose a given graph into a set of not necessarily disjoint neighborhoods. This has applications for analysing and summarizing the large-scale structure of complex networks."