Multiscale community mining in graphs using graph wavelets
Classical community detection tools output only one partition in communities, usually by optimizing a function like modularity on all the possible partitions of the network. Two issues naturally arise. In a network with multiple scales of description, these methods only output one of the possible solutions, and do not look at different scales. Moreover,
in a network with no community structure at any scale (like Erdös-Renyi graphs), these methods always output a solution even though it is not relevant.
My research tackles both these problems by developing a tool that looks for partitions in community at different scales, detects the relevant scales of description of the network, and outputs only the relevant partitions of the network - if there are any.