Title:Multiscale Community Mining in Networks Using Spectral Graph Wavelets

Abstract:For data represented by networks, the community structure of the underlying graph is of great interest. Instead of uncovering the overall "best" partition of nodes in communities, a more elaborate description is proposed in which community structures are identified at different scales. To this end, we take advantage of the local and scale-dependent information encoded in graph wavelets. After some developments for graph wavelets for which we introduce proper scale boundaries, physical scale and scaling functions, we propose a method to mine for communities in a scale-dependent manner. It relies on clustering nodes according to their wavelets or scaling functions, using a scale-dependent modularity. An example on a graph benchmark having hierarchical communities shows that we uncover successfully its multiscale structure.