1/ I was lucky enough to soak in Twitter's data for years. So I could see what features people actually responded to, and which we just *assumed* they'd respond to.
Buckle up, I have some interesting news that you, as a Mastodon user, are going to disagree with.
People don't like reverse-chron timelines. (head explodes)
Let me explain myself.
2/ As a Mastodon user, you're waaaaaay more likely than the average user to love reverse-chron with no algorithm. It's one of the main selling points for Mastodon!
But it's not rooted in what we saw in seemingly hundreds of different #designresearch findings.
The reason Twitter almost died so many times at the start is that normal people aren't looking for an IRC or RSS experience. So people joined, got frustrated, and left.
It took THREE YEARS to turn users into "healthy users." That's bad.
3/ Designers love talking about #empathy , but it can be hard to walk the walk. What do you do when literally 97% of your users are saying "I don't like trying to curate a good timeline, and I don't like having to scroll to keep up with everything, and I feel like I'm missing out on the good stuff, and I'm confused?"
Bad designers say "Deal with it."
Good designers say "Hm. This is a problem to solve."
And that's where algorithmic timelines, guide, moments, editorial content, etc came from.
4/ I remember when we moved from strict reverse-chron (which 97% struggle with) to "You might like" prompts (which 3% struggle with) and hearing from the VERY LOUD minority that we were destroying Twitter.
But we saw as the 97% had a much better time. We saw that every step forward we took (I have a whole presentation on this) was helping people more and more.
The data told us we were making a better product. And that reverse-chron kinda sucks. For most people.
5/ I remember when we added quote tweet, and we have the receipts. It does *not* increase abuse. (Source: I was the lead designer on the abuse team)
You know what increases abuse? Reverse-chron replies without any sorting or algorithm, because someone can say "Kill yourself" as the first response. It becomes a game. It silences people.
6/ My team rolled out mute keywords, which people SCREAMED at us about, saying it was killing the concept of "just show me everything" Twitter.
Victims of abuse, who aren't nearly as loud, thanked us quietly.
We rolled out "hide replies" which almost caused an internal riot, and was shelved for three years, but eventually it went live.
Again, some people said we were killing Twitter. Again, they were wrong.
7/ Mastodon is exciting for a ton of reasons. I love it. I want it to succeed. But ... have you read Animal Farm?
As Mastodon grows up, and tries to reach beyond the early adopters, please *listen to your users.*
They'll guide you to some features that cannot be optional in a system like this. And, sorry, a lot of them will feel like traditional Twitter.
Not because people have baggage, but because humans are going to human and certain things are inevitable.
8/ Good luck with everything. I hope sharing some of these opinions and experiences helps the fediverse in some small way.
I'm rooting for you.
@jon Reverse-chronological timeline is best for when you are Extremely Online, like me (and the other top percentage of all Rattata, er I mean, Twitter users)
@PavelASamsonov @victoriadecapua @jon I learned to hit the search bar with "filter:followers -filter:replies" every. single. time. That brings up JUST the people you follow and in chronological order. No ads. It was magic. And then Musk bought Twitter and now it sucks and I'm no longer there at all.
@victoriadecapua @jon @PavelASamsonov I’m surprised nobody has rolled out a basic #algorithm builder that would let you design your own sorting algorithm, and flip between it and reverse chronology. I’m obviously not a #SoftwareEngineer but it would seem like a plausible feature. We have mute and block filters. Personal, controllable algorithm would be harder but not impossible, right?
@bradpilcher @victoriadecapua @jon @PavelASamsonov Not impossible, and not even hard, but computationally expensive to scale if you want it "fancy". I have code for doing that with the Twitter API using Bayesian network over search results (train it by liking/disliking). Very good results with very simple code.
Doing something which mixes and reorders/ranks a limited set of feeds would still give pretty good results and be easier to scale.
@bradpilcher @victoriadecapua @jon @PavelASamsonov Many options. Could do it standalone, using the API to obtain posts to filter and sort, so you'd get another page to go to, or could implement it in the core of Mastodon, and expose it in the standard UI. Also an option to implement a specialized. instance you can follow accounts on which would boost posts according to different algos. I expect we'll see people try out different options. I sure intend some experiments when I get time.
@vidarh @victoriadecapua @jon @PavelASamsonov Seems unlikely it’ll show up in core Mastodon anytime soon. The culture and leadership here (however you definite it) has obviously placed not one bit of premium on making exposure or content processing efficient, on purpose. Since an algorithm’s primary value would be to supplement what we can process with our own time going chronologically, seems like something they’d have a strong aversion to.
@gerbrand @bradpilcher @victoriadecapua @jon @PavelASamsonov Yeah, I see a few models: 1. you can "just" reorder your own feed. That can be done either client or server side. Many options like priority for people you engage with a lot; spreading posts out so lack of timezone overlap doesn't mean you miss everything from people etc., but highly individual what people will want/need so I think there's space for many different variations there.
2. "topic" feeds. These can be done reasonably ...
@gerbrand @bradpilcher @victoriadecapua @jon @PavelASamsonov .. cheaply server side, but they don't need to be done separately by every instance. E.g. we already have people operating "groups" that needs to be mentioned, but topic feeds can be done similarly but aggregating the content these accounts boost in all kinds of different ways. This ties in fine with option 1, in that you can just follow them.
3. Alternatively you can have your own personal "topic"/discovery feed. This is what gets ..
@gerbrand @bradpilcher @victoriadecapua @jon @PavelASamsonov ... computationally hairy, in that it'd potentially take processing a lot of posts for every user running it. But I think if you do option 1 and 2, you don't really need option 3 - you can just follow a wide variety of such topic feeds and then apply ranking of them client side from your timeline.
@vidarh @gerbrand @victoriadecapua @jon @PavelASamsonov Option 3 would definitely feel both unnecessary and too far out of line with the prevailing norms and culture here. It is also, arguably, *the* thing that most people hated about algorithms on other social media sites. But I would love option 1 and 2. Algorithms for discoverability where you’re not purposefully directing it seems like a path to hell in terms of UX, but as a way to help parse more from your feeds… <chefs kiss>
1) if a user doesn't want to go through the trouble of curating their own feed, then they almost certainly don't want to go through the trouble of learning how to program the platform.
2) No one wrote Twitter's content curation algorithm. Instead they wrote an algorithm that would intake a bunch of user activity data and output a content curation algorithm. The output would then be used to determine what to show to users, but it's important to note that it wasn't taking an approach that was human readable.
@bradpilcher @victoriadecapua @jon @PavelASamsonov instead, it would identify that the current user responds well to content that was "glorpy", but not very well to content that was "fexish"-- we don't necessarily know what those categories even are. They're just the result of a computer crunching a bunch of data and finding some patterns in it.
@PavelASamsonov @jon It's also great if you follow a lot of people. No one has time to read everything. Start with the newest and work downwards as you have time.
That said, I hate the Twitter app or webpage with a passion and never would've stuck with Twitter if a co-worker had not introduced me to Twitterrific. I hate being bombarded with info I never asked for, so I guess I'm one of the 3%. And now I'm suspended for removing a phone # that I was forced to give, but the timing seems good.
I do feel sometimes the recency obsession of rev-chron has unhealthy aspects too: addictive "what's happening right now this very second", and a perspective on the news that encourages instant rather than considered responses, emotive gut reactions, and hot takes. .
I would like the option of a feed (here or RSS/Atom or wherever) which composed reading materials with a little clustering to help readers compose a higher level view of what happened than the "he said she said they said" that we get from the endless stream.
Those who only follow a handful of feeds may be kind of doing this for themselves, in a way.
Users should be able to pick their timeline. To sort it by who you interact with most and/or to boost up those who post infrequently over those who are constantly posting.
What *shouldn't* happen is *other* people's likes altering *your* feed. And the fights they're involved in getting into your feed. *That's* what makes Twitter toxic.
The OP mixes up satisfaction with engagement :)
I hope I'll live to see a day when engagement engineers will be brought to justice for knowingly creating destructive behaviour loops!
When I say "justice", I mean literal prison sentences for Instagram, Twitter and Facebook engineers and managers.
@jon One of the big problems with "the algorithm" for me and many others is that we simply STOPPED seeing content from people we want to see, and started seeing MORE content from people being reactionary, fandom drama, call-outs, and the like.
The algorithm made the experience a lot worse for me and others, and turned our timelines into a shit show.
I'm okay with some subtle algorithmic sorting, but rather then bringing conflict forward, to bring accounts forward we frequently interact with.
@jon Also the algorithm shouldn't completely take over. Bring SOME content forward, but make sure ALL the content is displayed. Don't suppress content with certain keywords like "commission" and make sure we see stuff reliably from EVERYONE we follow.
Yes, even people we don't interact with often. Because suppressing their content is a self-reinforcing problem!
@jon When I switched to "Latest Tweets" after being on "Home" unknowingly for a long time I started seeing tweets from people I hadn't seen anything from in months, and that's when it really struck me just how BAD the algorithm is.
So yeah, it might be a good idea but it was a horrible execution. Enjoy your social media vacation. :)
@zorinlynx @jon Interestingly, I had the opposite experience.
I changed to 'latest tweets' at some point, and after a while I noticed I hadn't heard from people. If I checked their profiles, they were still posting, but I had just missed them by not being online at the time.
When I turned the algorithm back on, I suddenly got their tweets showing up again, raised to my attention even if they were made 4-5 hours before.
Now I use the algo for the first morning visit and change to timeline after
@_intothevoid @jon @zorinlynx this is because Twitter never allowed you to save your reading position and why using 3rd party clients was a superior reading experience on Twitter. I could never use the official client because of these reasons and it’s one of the reasons why we developed Twitterrific in the first place.
@gedeonm @_intothevoid @jon @zorinlynx I rarely encountered most of the problems people complain about with Twitter. No ads. No crazy tweets elevated by the algorithm. No missed posts from infrequent posters of great content.
How was that possible? Third-party browsers (Talon, Plume, etc), chronological mode, extensive use of lists, and TwitterListManager. I really hope a rich ecosystem of 3rd-party software develops around Mastodon.
@pgcommunication @gedeonm @_intothevoid @jon @zorinlynx I stopped using 3rd party apps when I realised they were stuck using polling to find new tweets, unlike the official app which supports streaming. At least at the time, which was years ago. Probably would go back though if I still used Twitter.
Hometown is adapted from Mastodon, a decentralized social network with no ads, no corporate surveillance, and ethical design.