algorithmic choices I would like to experiment with in my mastodon feed:
1. show me boosts, but hide any particular boost that I've seen more than 2 times
2. show me responses in context of the thread... i.e. no click needed to figure out the context
3. not strictly part of the feed logic, but I'd love quote toots. Because I want to boost but with some context around why I find something interesting
I mention these because I think Twitter has anchored us in a false dichotomy: reverse chronological or high engagement rage posts. Those aren't the only two choices. There's an infinite number of smaller choices that could possibly (need experimentation) make my feed better.
so my largest request would be that there be an experimental platform for experimenting with feed sorting algos. And have that logic able to be exported by a user and shared. Could drive interesting innovation
The closest thing to an algorithmic timeline for Mastodon is @[email protected] 's Mastodon Digest - an open source python tool that delivers posts from your timeline. With customizable algorithm options! https://github.com/hodgesmr/mastodon_digest
Do you have a sense of what it would take to build critical mass around this? I’d really like a similar thing and have been exploring options on how to get started.
@dhadfieldmenell I don’t know. Two things I suspect:
1. most folks will only ever use the default settings. Internet explorer taught me that.
2. The first implementations will likely be client side hacks because that’s the easiest entry point, but that will be bumpy in many ways. Needs to be server side to scale.
Both of those are good points (the default one especially). I’m planning to experiment in the space I have some funds to find a couple smart students and see what they put together. There’s a limited amount that we can do, but I would love to hear thoughts and suggestions for concrete projects that seem like easy wins here.
@Cmastication Wonder if this customization is possible at the app level (I don’t see why not) vs. at the instance/server level.
But I agree, having options for timeline customization would be nice down the road.
@kdpsingh @Cmastication Usually this kind of algorithmic stuff is done on the server because it has access to all the data. Clients, especially mobile, are optimized to download as little data as possible so they’re fast.
Consider a really simple algorithm that always loaded @Cmastication toots at the top. The client would need to download your whole timeline to find all those posts. Basically, you’d need to mirror everything on your client and keep it in sync. So slow, big and error prone.
Nate Rich
JD Long ✅
Andreas
Joe Wasserman
Dylan Hadfield-Menell
Alex
Karandeep Singh
Jay Lyerly
Joe Harris
Dunc