Aaron McDaid

The use of modularity for community finding is based on what I would call a minimum likelihood approach. The idea behind modularity involves creating a model that is as wrong as possible, called a null model, then find the parameters for which the fit between the data and the model is a bad as possible.

This is in stark contrast to more common modelling, where you try to fit a good model as well as possible to the data. Hence I refer to a modularity-based method as a minimum likelihood estimator.

Two wrongs don't make a right, and therefore it no surprise that modularity is not suitable for community finding.