> they’re binary: either you link to something or you don’t (if there are different implementations of PageRank that address this, please let me know in the comments!).
This isn’t the case at all!
PageRank is modeling a person going from website to website by doing two things:
(1) clicking a link on the page at random
(2) going to a random page on the web (a poor approximation of visiting a content aggregator / pasting a link into the URL bar / being sent links from friends / visiting a favorite)
Typically, you “simulate” someone doing this a large number of times (say, 1 million), then ask, “What is the probability they ended up on site <some-site>?”
Nowhere in the above model does it require that the probability of every action is equal! More probable links (e.g., those with bigger font, or closer to the top of the web page) can be given more weight in the simulation.
> There’s also another issue: PageRank doesn’t take into account sites linking to themselves, and so we can’t use data about a pokémon type being weak to itself or effective against itself.
This doesn’t have to be true. Those links don’t add anything, but they don’t hurt the model — and they don’t let you “cheat” any more than building your own spammy self-linked web ring.
In the probability interpretation above, there’s some chance of the surfer doing nothing — and staying where they are. (And maybe this is more likely on, say, a game website than a news article)
