Social media sites do this because all their stats show that users actually ‘like’ non chronological order even when they say they don’t. Telling people what they want when they ask for something else is a big nono, so they just change it for you on the sly.
Of course, the main way they measure ‘liking’ the site is via engagement, which may actually just measure compulsion to use, not enjoyment. And, of course, the however small, cohort of people who genuinely ‘like’ chronological order are ignored.
> Social media sites do this because all their stats show that users actually ‘like’ non chronological order even when they say they don’t.
I think the issue is a little deeper: it's not that people want chronological vs. non-chronological timelines, it's that they want a way to get the "best", "most relevant" content to be surfaced from those timelines and presented in a sane way. Non-chronological timelines are better at producing the "best", content (for some relative definition of "best", at least), but by presenting it out-of-order, it requires more of the user.
My guess was actually that it's more efficient to show users a bunch of popular, cached content than it is to show them a truly chronological timeline full of new and unpopular content, which is unlikely to hit their cache so readily.
I kind of doubt that it's a caching issue. Twitter feels a lot like email to me -- sending a tweet is a lot like emailing to a group of followers. (Though it's unlikely they implement it that way because of the number of tweets that go unread.) Email providers operate at a scale similar to Twitter and don't replace your email with popular emails to increase cache hits.
Maybe to improve profitability you'd want to improve how much you can serve out of memory, but I think Twitter has more than enough compute to just generate your page for you when you visit. (I haven't seen a fail whale for over a decade!)
> sending a tweet is a lot like emailing to a group of followers. (Though it's unlikely they implement it that way because of the number of tweets that go unread.)
My understanding (from a watching a Twitter Tech talk, maybe about redis?) is that that is actually how it is (was) implemented. They were so focussed on time to first render, they took the efficiency hit.
I have a deep hunch that it gets far stupider even than measuring compulsion.
Suppose that your feed refreshes itself while you're trying to read a particular tweet. You go back looking for it--now badly ordered, irrelevant content is positively correlated with your amount of scrolling and time spent in-app.
So the app isn't just optimizing against your lazy attention, it's probably in some cases also rewarding itself for actively hindering you.
These software patterns are anti-human. Imagine using a hammer that is trying to maximize your engagement with the hammer itself. Well, I know I'll be more engaged with the hammer if the head keeps falling off, but that isn't what a hammer is for.
I certainly engage with content more when 'the algorithm' shows it to me repeatedly, which is what happens on Twitter. Of course I don't necessarily like that.
Even beyond "compulsion", it might just take longer to see what I want to see because there is other junk mixed in. If they are just measuring how long I scrolled and how many ads I accidentally clicked in the processed, they might be measuring inefficiency and mistaking it for engagement.
Making the UX worse so it takes a longer time and more clicks to get what you want out of a site could look like improved engagement. What if I just want to quickly check-in, see what's new, then bounce?
Most people wanting chronological quickly find out they don't really like it anyways, is my guess. Most twitter users I see now follow over a thousand people, scrolling through all that every day is impossible, it needs to be curated somehow.
That seems far from impossible actually. Especially if you consider the likelihood of all of their followers posting even once per day. You can scroll pretty quickly if it never stops.
Of course, the main way they measure ‘liking’ the site is via engagement, which may actually just measure compulsion to use, not enjoyment. And, of course, the however small, cohort of people who genuinely ‘like’ chronological order are ignored.