• tal
    link
    fedilink
    English
    arrow-up
    3
    ·
    edit-2
    3 months ago

    I’d actually like to have an effective recommendations algorithm. Originally, Reddit was intended to use voting as a way to predict what you wanted to see and show a personalized recommendations set. That…basically didn’t work. What happened instead was that subreddits were introduced, and I used those to filter instead, just subscribed to some and only look at content from those. I do the same thing here.

    I do think that there’s a problem where recommendations algorithms either need to store a lot of data about you – which I don’t want to hand to one entity along with everyone else doing so, too much potential for data-mining – or need to recommend to the preferences of some “aggregate user” that reflects what the typical user wants, which usually isn’t what I want. Google News did the latter. I really liked their recommendations early-on. But over time, what they recommended shifted, got a lot more sensationalist and lowbrow. I suspect that this reflected their changing userbase. Maybe they could do better recommendations if I created an account and let them profile me, but I’m not willing to do that. Google has too much data, in my view, on everyone already.

    In theory, my home instance of lemmy.today – a small Lemmy instance – could profile me, but because the Threadiverse is decentralized, it’s harder for any one party to get a large dataset to do that for everyone. Any instance operator can see what I comment on (as with Reddit) and vote on (unike Reddit), but not what I view. They can’t directly link that to my IP address either, which helps in profile-building, though as I’ve pointed out in the past, as things stand, the Threadiverse doesn’t do a great job of isolating one’s IP address, and a dedicated person could probably do a reasonable job of harvesting IP addresses.