[aperson]

Originally Posted by xXOpkillerXx

I don't think Google's SE's implementation has any public documentation (?), but you can probably get some of their policies. Other than that, it's pretty much hypothesis; the engine must be quite complex tbh

Ya that's the joy of opaque nonlinear correlation algorithms. Even if they show you what all the categorization features are they are also probably similarly opaque and unclear. Gives you lots of wiggle room to do questionable stuff.

Same deal with insurance markets using correlation algorithms that take stuff like zip code as features. Oops, accidentally made the network racist.