Chris TrottierMay 29, 2023
#Bluesky now allows you to “choose your own algorithm”.

Which sounds “incredible” and “sci-fi”—but it really isn’t.

What it essentially does is give a Twitter-like service Reddit-like features.

As an aside, now I’m wondering why Reddit doesn’t offer an alternative web front-end to make it more Twitter-like!

But how does this apply to the #ActivityPub flavour of the #Fediverse? This feature now makes me realize how big a deal Fediverse groups are going to be, and if I were @Gargron, I’d be even more excited about rolling out Mastodon’s group functionality.

Because while groups aren’t exactly relevancy algorithms, once you add a “New”, “Hot”, “Best”, etc. feed to groups, now you’re in business.

I don’t know if choosing your own algorithm is the killer feature that Bluesky thinks it is. My experience is that most people hate choice. Nevertheless, I still thinks it’s important.

@[email protected]
Show threadChris TrottierMay 29, 2023
If this is "choosing your own algorithm", #Calckey has actually had this feature long, long ago.

We call it "Antennas" -- and you can easily build an Antenna yourself.

Here's a screenshot for Antenna creation settings.
Show threadChris TrottierMay 29, 2023
So I'm thinking about this in more detail, and I think the "choose your own algorithm" feature with #Bluesky is really not hard to implement, and something very doable on the #Fediverse.

What do we exactly want from an algorithm? Topics.

And we want topics sorted according to the following:

1. Hot
2. New
3. Top
4. Rising

Some people would like a "controversial" feed, but we don't have to give it to them 😉

As for "Top", we can sort it according to time parameters.
Show threadSteve BateMay 30, 2023
@atomicpoet How would you determine "Hot", "Top", "Rising"? Mastodon, and probably other implementations, don't federate likes. That may be true for reply counts too.
Show threadeshepMay 30, 2023
@atomicpoet @steve, could you expand on "don't federate likes" please?
Show threadGustav Lindqvist
@eshep @steve @atomicpoet IIRC they are sent one way from the one doing the like to the one receiving it. The one receiving it does not broadcast it out to everyone else. So you can see a pretty good count for likes on your own posts and you can see your own likes on others posts but everything else is miscounted.
May 30, 2023 at 7:14amWeb
Show threadSteve BateMay 30, 2023
@gustav @eshep @atomicpoet That makes sense. Thanks for the clarification. My point is that this signal can't be reliably used for timeline ranking, at least in smaller instances. I'd personally like the ability to use custom ranking based on post data like content, author, instance, and so on (could include local likes too). I'm thinking of something like a Bayesian or ANN boost/filter trained with my own historical likes and/or boosts.