Run community detection now that we have tractably-sized networks

[jwzimmer-zz]

Use methods from https://github.com/jwzimmer/tv-tropes/issues/21

• Run community detection on the network of tropes based on the tropes they link to
• run community detection on the network of indices based on the tropes they link to
• maaaybe run community detection on the combination but that may again be too large? (as in https://github.com/jwzimmer/tv-tropes/issues/22)

compare the categories we get to:

• the indices (their categories)
• the stanford neighborhoods

[jwzimmer-zz]

Index network clusters

Girvan-Newman 1) Girvan-Newman 6 clusters (b/c of the number of neighborhoods found by Stanford): https://github.com/jwzimmer/tv-tropes/commit/96e70cc68aed8db7cef072b5060aa98959c3889d on this network: https://github.com/jwzimmer/tv-tropes/issues/22#issuecomment-727626405.

• I think because the tropes are so much more tightly connected to each other than the indices, the clusters basically ended up putting all the tropes together and then each index in a cluster? 2) Girvan-Newman 12 clusters on the same graph as (1): https://github.com/jwzimmer/tv-tropes/commit/5cf06dd93e9a69584456fb0e9b91085707b8ee43 3) Girvan-Newman 6 clusters using graph in BigFour_tropes_all4_top10000_top50_top20.gml: https://github.com/jwzimmer/tv-tropes/commit/8f1d92a164ba4a459ee870fb18f5852d8fd7a18f
• Omitting the links between the tropes the tropes link to (difference between this graph and the graph in 1)
• Viz: https://github.com/jwzimmer/tv-tropes/commit/eb5a1b4b626df2be70c23d37e6e597abe0129e9e

4) Girvan-Newman 12 clusters using graph in BigFour_tropes_all4_top10000_top50_top20.gml: https://github.com/jwzimmer/tv-tropes/commit/2e378170891c2a19c5c3726527fe1252f7f69ac6

[jwzimmer-zz]

for now i'm only doing girvan-newman clustering for the indices: we're not sure if seeing how they cluster will be interesting compared to how the tropes cluster; if it looks like it is we can come back and do different kinds of clustering with the indices as well

[nguyenhphilip]

woo! finally got something to work. here's a group of 5 clusters with 100 nodes and 313 edges. each node was limited to 5 edges in adding them to the graph. more could have been added to a particular trope as edges were created for the tropes that were later added in the for-loop. but the overall size looks manageable!

https://github.com/jwzimmer/tv-tropes/commit/0f400773a32eb0aee3e57710ef130d3a5df7f84b