The modularity of the new network is quite poor (we suggestedhere that a network is definitely modular only when the modularity is 0.6 or above):
|||partition = community.best_partition(tag_network)|
|||nx.set_node_attributes(tag_network, partition, "part")|
Apparently, counting raw co-occurrences is not the best way to describe similarities—indeed, correlation-based networks are much more flexible, and you will learn about them later (Chapter 14, Similarity-Based Networks). However, even the coarse network that you have allows some meaningful interpretation.
The partition that you extracted ...