Discovery and the Fediverse (algos, curation, interfaces)


Convener: Gabe Kangas

Participants (add yourself here if you want to be listed, no obligation): Name and Fediverse handle

  • Hannah Aubry (@haubles@fosstodon.org)
  • Jesse Baer (@misc@mastodon.social)
  • David Slifka (@davidslifka@mastodon.social)
  • Davide Eynard (@mala@fosstodon.org)
  • Matt Terenzio (@librenews@mastodon.social)
  • Christel van der Boom (@xtel@flipboard.social)
  • Marlene Breitenstein (@BreitensteinArt.mastodon.art)
  • Mike McCue (@mike@flipboard.social)

Key points:

  • fediview, fedifeed, lots of things coming out right now

  • Chris (@UncensoredNews) says: 9 recommended articles about problems with the MSM algorithms. https://uncensorednews.us/algorithms/

  • Davide says: https://fosstodon.org/@mala/109914623812016969 - a thread collecting different approaches to search and timeline algorithms on the Fediverse

  • The fediverse is currently a “human-based” algorithm

  • David Slifka says: Also: There is a group @algorithms@venera.social that you can follow/mention from Mastodon, for this topic.

  • Any algorithm should be open from the beginning

  • “You either trust all of the system or you trust none of it” - Gabe

  • Weighted hashtags: e.g. if a post has two hashtags, each one has only 1/2 weight, three tags -> 1/3, etc.

  • Algorithms must be opt-in

  • User-side algorithms - open & transparent, explanations of why recommendations happen

  • Views & interfaces rather than a chronological view

  • Also weighting the people you follow (more important, less important), and…

  • Limiting the number of posts you see from each per day or time period.

  • Davide says: Re: engagements, I think this is an interesting read: https://medium.com/understanding-recommenders/whats-right-and-what-s-wrong-with-optimizing-for-engagement-5abaac021851

  • List makes are “curators of the curators” - Twitter never leaned into lists, but they offer all sorts of possibilities, e.g.: Filter them for topicality central to the list; normalize frequency among list members; much more

  • Gabe Kangas says: Here’s some extended thoughts from me if anybody is interested: https://gabekangas.com/blog/2022/08/decentralized-discovery/

  • Reputation trust systems

  • Davide says:On limiting content/users: https://github.com/mitotic/pinafore-mah

  • Jon Udell says:Marlene, might’ve been me: https://blog.jonudell.net/2023/01/16/mastodon-relationship-graphs/

  • Clients can define different ways to introduce algorithms, and you choose the method you like by app.

  • Marketpace for algorithms: clients as a view, opt-in to servers with alo-based feeds

  • A sort of directory: Self-selecting into various categories (like selecting hashtags in bios) to be filterable by users who want to follow those types of content.

  • Xtel says:Here is the account that Mike is talking about: https://flipboard.social/@NewsDesk

  • Newsie.social took the approach to manage the links (news) in our instance to ensure that ONLY real and legitimate news orgs percolate. https://newsie.social/explore/links

  • Readily is adding Brave Goggles to it’s daily digest to allow users to choose what gets boosted https://readily.news