Hi, I am writing here since I’ve been searching for something like this for over a month now, but I cannot seem to find anything that is satisfying, possibly because there isn’t something out there that fits what I’m looking for.

Basically, I am interested in decentralized recommender systems that are:

  • currently being used

  • customizable, on the client side, at run time, to a certain extent (e.g. regulate exploratory vs safe recommendations, tag based)

  • if not being used, at least “usable out of the box”, via code released on github

  • if nothing like this, at least a promising projects that are currently very active

Best case scenario would be a decentralized recommender system already applied to the fediverse. I am trying to get into this eco-system, but the lack of filters/tools of discovery for the content I’m being shown is a major hindrance for usability, at least for me. Being able to direct my own reccomender system, in order for it to be alligned with my goals and not with my addictive tendencies would be a bit of a dream, but I feel that a decentralized system would be the only way to achieve such a thing.

  • paequ2English
    1·
    6 hours ago

    Ooooh, related: https://www.piki.nyc/

    Using Piki data, we find that while the like rate increases monotonically with artist popularity on Spotify, this does not hold true for superlike rates

    This company’s product seems to be music recommendations. We need something like this, but open source and federated.

  • paequ2English
    8·
    12 hours ago

    Being able to direct my own reccomender system, in order for it to be alligned with my goals and not with my addictive tendencies

    AGREE! There are options for controlling the data side of things, Lemmy, Mastodon, Jellyfin, torrents, but I’ve definitely noticed the recommendation side of things is basically non-existent. What I miss the most from Spotify or Netflix isn’t the music or movies, it’s the recommendations. There’s a ton of content outside the megacorps, but we don’t have a good way to find it.

    It would be awesome if we had an algorithm that we could control. We could tune it to whatever we want, instead of letting these giant megacorps shove their shit in front of us.

    • angrytako@lemmy.mlOPEnglish
      4·
      12 hours ago

      The dream!

      I’ve seen some papers on some experimental systems that can work in a peer-to-peer environment, that try to balance our privacy (since somehow we need to collaborate without exposing all the content we like or are interested in) and good recommendations. It is however quite an endeavor implementing such systems.

  • Scrubbles@poptalk.scrubbles.techEnglish
    3·
    13 hours ago

    I’ve thought about this, an open recommendation/review system that could be based on your fediverse profile. Anyone could use their existing account to review things, but the reviews could be ranked based on how trustworthy you are on the fediverse, so bots/spam accounts wouldn’t be taken seriously. Just haven’t had any time to make it…

    • angrytako@lemmy.mlOPEnglish
      2·
      13 hours ago

      To add to this, most platforms, at least in Europe, but I guess its the same everywhere, are required by law to give you all the data they have on you. By building “adapters” for the major platforms, in order to process their formats, you wouldn’t even need to start from 0 when “migrating” to the open source socials. If rendered automatic, even the average person could use it.

      Btw I as well I am starved for time, otherwise I would love to try my hand at it, but the best I could do right now it’s just maybe help. I am a Machine Learning dev, so I have some practical knowledge in that domain at least, although not on reccomender systems :/.