BTW, you could instead apply some of the analytical approaches I used here:
viewing hearts as Approvals on ballot “candidates”. Proportional representation schemes excel at identifying the consequences of “factions”, but multidimensional scaling is a visual method that’s much better at showing who the factions consist of.
It could all use some tweaking to better fit this different context, though. The thought inspires me to try a conceptually much simpler approach based on counting a version of n-grams (“how many posts were hearted by all of X, Y, and Z?”). If that proves interesting, I’ll write it up in the referenced topic tonight or tomorrow.
The intuition is, I think, clear enough on first sight: factions reveal themselves by which people tend to like mostly the same posts Which leaves those who fear to express a view invisible, but then they already are.
Off topic? Mostly. But The State could use it to automate building Enemies Lists without even bothering to read through content, so it’s not really hostile to power 
.