Filter entities associated with docs in corpora (e.g. low frequency)
Generate a cooccurrence network of entities (and normalise edge weights?)
Apply community detection algorithm at different levels of resolution.
One advantage of community detection is that there are some inherently overlapping community detection algorithms. There is also SBM in graph-tool which generates nested communities.
Proposed method:
One advantage of community detection is that there are some inherently overlapping community detection algorithms. There is also SBM in graph-tool which generates nested communities.