Re: re. It's time to reveal all recommendation algorithms
This, and what the poster just above said. I make some effort to obfuscate my online presence, but not nearly as much as other posters here; I'm fairly sure that, in principle, Google knows more about me than most of my relatives. And I still get tons of utterly nonsensical recommendations.
If I try to use Google News, it still frequently gives me soccer and horoscopes, in spite of the fact that I've never searched for either in my entire life anywhere, and I have in fact explicitly told it dozens of times that I do not want those topics (before giving up on it altogether and switching to another aggregator).
Spotify? I have over a thousand songs in my favorites, and yet if I tell it to play from my favorites at random, there's a half dozen songs that pop up like 10% of the time.
Amazon's recommendations sort-of make sense. Most of the time. Unfortunately, any time I buy something like a vacuum cleaner, it then spends the next few months attempting to sell me dozens of vacuum cleaners. Is it that hard to understand that there are items nobody gets more than one of?
Ads served by Google are - I dunno, sometimes they have something to do with my recent web searches, but most of the time they are... not quite random, but apparently fixated on things that come completely out of the blue. These days it's big in apartments in places I've never heard of, for example, but up until a couple weeks ago it used to show me some kind of manga.
I believe there are two kinds of recommendation algorithms right now: ones that are based on neural networks, and those that are based on a ginormous mess that nobody has understood in years.
Unfortunately for the article's author, there's something those two categories have in common. You cannot explain the output just by knowing the input and the algorithm. And you cannot apply effective constraints on their output just by tweaking the input.