Interesting problem with the bipartite network analysis I'm using: I think it breaks if there are even a couple of authors who write for a ton of fandoms.
Basically, you start by finding all of the connection points (author to fandom but also shared authors within a fandom/shared fandoms within an author) and then you make sets, and then you narrow those sets down by first removing any subsets (if set 1 is {a1,a2,a3,f1,f2,f3} and set 2 is {a1,a2,f1,f2} then you get rid of set 2 because it's just a subset of set 1), and then you narrow down further by combining sets with a high degree of similarity. And so on until you none of your sets are similar enough to combine.
But if you have even a couple of authors who write for a ton of fandoms (and/or one or two fandoms e.g., Harry Potter that a huge fraction of authors write for) you can end up in a situation where all or nearly all of the sets are just subsets of or extremely similar to one giant set, which is just all or nearly all of the authors and all or nearly all of the fandoms all in one group together.
Which isn't wrong necessarily (they are all part of one broad community) but it keeps you from getting any actual useful information about the sets within that community.
As far as I can tell the paper that I'm getting my process from doesn't account for this issue, I think because the data they were using didn't have this characteristic. I'll have to see if I can find a process that can account for this, or if there needs to be a step in the beginning to pull certain authors out of the data before starting the main analysis.
