Repel Communities and Multipartite Networks

Abstract

In graph theory, there are two complementary approaches to grouping nodes using the links for guidance. The first is finding the components; the second is finding multipartite partitions (vertex colorings). In network mining, the technique of community detection can be thought of as a relaxation of component finding. The results are groups that are all tightly connected but with some few connections between the groups. One can also envision a relaxation of multipartite partitions such that the groups have very few connections within the group but most of the links connect nodes in different groups. These groups, called repel communities, are discussed and three new algorithms are introduced to detect them.