Thanks for your quick reply!
You were right - the algorithm just didn't converge to anything meaningful:
This is a hard problem to solve in general. It is usually the case that
you should have some insight on the total number of communities, and
AFAIK there is no known reliable method which will give you this
information easily. If your networks are not too large, you can try
increasing n_spins values, and stop if the modularity no longer changes
Thanks for your advice! My networks are large, so I can't really
afford multiple iterations...
I have so far also tried the MOSES tool that is quite fast, but the
downside (for me) is that it produces overlapping communities and that
the granularity is too small for my liking...