Overlapping community search in very large graphs

Abstract

The main objective of the thesis is the creation of an algorithm to detect the community structure of large graphs, allowing for nestings and overlappings. Although it has been shown that communities are usually overlapping and hierarchical, we must stress that most of the literature related to community search has focused on nding partitions of the graph. In addition, given the size of modern data sets, most of them typically rely on prohibitively expensive computations. We will propose the algorithm OCA, an algorithm for community detection with nestings and overlaps. It has been able to run in the larger datasets of which we are aware. Our algorithm neither requires the user to set non-intuitive parameters in order to get good results, nor to preassume a certain size or number for the communities, since they are found naturally from the graph structure. The core of the algorithm relies on the de nition of a new tness function that allows to evaluate the quality of a community naturally including those nodes shared by other communities.