Twitter showed us its algorithm – what does it tell us?

https://knightcolumbia.org/blog/twitter-showed-us-its-algorithm-what-does-it-tell-us

by randomwalker • 3 years ago

225 163 3 years ago

jawns• 3 years ago

I would assume that if "probability that other users will positively engage with a tweet" is the primary determiner of reach, then the more you can help Twitter accurately predict that probability, the better, because otherwise the default probability is likely no higher than middle-of-the-pack.

If that assumption holds, then I would guess this type of algorithm favors consistency of content. In other words, someone who picks a certain topic and consistently tweets only about that topic is going to be easier to form predictions around versus someone whose tweets show much more variety, in topics, styles, etc.

What that might mean, from a "gaming the system" point of view, is that if you're a person who intends to primarily tweet about two or three disparate things, you might be better off creating a separate account for each, rather than a single account where engagement is harder to predict.

delecti2• 3 years ago

Incidentally, those same conditions seems to apply to other sites. Lots of Youtubers have multiple channels (a main channel, a livestream channel, a shorts channel).

netcraft• 3 years ago

I've thought many times over the years that I would love to be able to subscribe to a particular playlist or "show" from a channel. There is several channels that I want to see their main stuff, but not their side content. Or a particular game from a lets-play-er, but not their other games.

thrashh2• 3 years ago

I don’t think it’s only helpful to the algorithm. It’s also helpful for me as a subscriber.

If I want to watch episodes of Breaking Bad, I don’t want you to randomly throw in episodes or M*A*S*H (even though both are good).

iamflimflam12• 3 years ago

I wrote about this - https://www.atomic14.com/2023/03/06/the-curse-of-youtube.htm...

To be honest it's pretty horrible. Why can't I just post about what interests me right now?

just_boost_it• 3 years ago

That doesn't really make sense because most big accounts tweet about a range of topics. There's pretty well established ways for estimating the probability based on how the range of topics you might tweet about would match with the range of topics a user likes. That means you have to try and figure out what your base likes to see, and be like that. Tweeting about only a single topic means that you're only targeting people who are likely to like tweets from accounts that tweet about that one topic.

teruakohatu• 3 years ago

That doesn't really make sense because most big twitter accounts tweet about a range of topics.

People engage with celebrities on everything. If an A list celeb announces they enjoy a slice of lemon in hot water, twenty news articles will be published around the world.

whimsicalism6• 3 years ago

It's probably less this and more "if you only talk about one topic, then when we show your posts to similar users, they are more likely to like it"

gregbander• 3 years ago

As the article calls out, the code is right there. Post your results of tests, not knee jerk conjecture. Wrong opinions are a dime a dozen.

JenrHywy• 3 years ago

Uh, the article also points out that the algorithm itself can't tell you anything definitive about how a given tweet will be ranked without the training data.

winternett• 3 years ago

In all the infinite wisdom of developers and tech gurus, we get dictatorial systems that rigidly force dumb and obtuse rules and stupid "success scripts" on creative people that only end up rewarding trust fund babies, scammers, and/or hackers... Leaving the most talented and original creators to wither in obscurity.

Social media sites like Twitter (Now Titter) have brought out the worst in us, and now that they're collapsing under their own arrogant leaders, no one is really considering the lessons learned... They can't exist without creators, but creators can exist and function on any platform. Screw Twitter's dictatorial algorithm, mastering it does not help anyone to be creative and successful unless they're trust fund babies, scammers, and/or hackers.

TeMPOraL2• 3 years ago

Let's be honest here: vast majority of "talented and original creators" on social media are the ones seeking out, sharing, and geeking about "success scripts". And, of course, social media isn't unique here. This is a story as old as time: any activity that has a competitive element to it gets dominated by those "playing to win". It's true in every aspect of life. I feel the only way to avoid it is to suppress the entire competitive angle - which means removing the rewards altogether, or making them purely random, as any correlation between activity and reward will attract people willing to game the activity.

mountainofdeath• 3 years ago

Given what has been open sourced so far, it makes sense that content that is likely to be controversial, or content that generates neutral to negative engagement would have a smaller probability of being displayed.

I suspected this and told my far-right/left wing acquaintances, that no, Twitter (and Facebook too) isn't suppressing you, your content is just a net-negative from the platform's perspective. The platform is in the business of keeping the bulk of its users and advertisers happy.

akira2501• 3 years ago

When people talk about suppression do they mean that their own tweets are suppressed, or the tweets of people they follow are? Or tweets from news organizations are, depending on their content?

whimsicalism6• 3 years ago

The evidence is abundant for active suppression of certain political views beyond just predicted engagement. No clue about Twitter's algo, but Facebook certainly does this.

dekhn2• 3 years ago

Can you, uh, share that abundant evidence?

simplotek6• 3 years ago

I was disappointed by this article and how it omitted the fact that Twitter hardcoded how references to Russia's invasion of Ukraine should be downranked.

https://news.ycombinator.com/item?id=35410841

GenerocUsername• 3 years ago

I don't know when this dataset was published from, but there was def a time when Ukraine news was beyond saturation.

phailhaus• 3 years ago

Because if he's going to claim to be the internet's town square, he can't pick and choose topics that he's personally 'tired of'.

madeofpalk• 3 years ago

It's the sort of thing I would expect on a highly opinionated Hacker News (iirc like how posts about Apple have a penalty applied to them to counteract massive usual interest in them), but less so on something more general audience like Twitter.

I'm not really looking to twitter to say "Actually, we've all heard a bit too much about Tennis". I don't want Twitter's timeline to have an editorial voice.

paulpauper2• 3 years ago

If I had to guess it's because it gets poor engagement and would crowd out other topics due to the popularity of the topic.

simplotek6• 3 years ago

If that was remotely true then why wouldn't the topic be handled at the training level? Instead, Twitter is censoring references to Russia's invasion of Ukraine through the same mechanism used to kill DMCA violations. This is not an approach motivated by "engagement".

theteapot2• 3 years ago

That HN artcile is "[flagged]". Reading the comments I think it's because there is no analysis of what "UkraineCrisisTopic" actually means or does? Author seems to just grep code base for "Ukraine" then draw what conclusion fits narrative.

ROTMetro2• 3 years ago

Versus the push that we should draw the conclusion that it's a nothing burger? It definitely highlights that Twitter is happy to categorize and treat Ukrainian content differently which is insightful to know.

joshxyz2• 3 years ago

This is quite nuanced.

Damned if you do: people will question your political stance.

Damned if you don't: you'll lose a lot of users who uses twitter for various reasons other than politics and warfare. People that just want to relax and unwind, be entertained, and plug into the matrix.

Seriously though I think a better UI for muting topics would be a better user interface than hardcoding these things. In twitter you can mute topics but you gotta dig through the settings first.

takeda2• 3 years ago

I think if you don't (boost or demote) is generally perceived better than if you actively interfere.

They could simply provide option to block tweets with specific hashtags (or maybe they already do, I don't use twitter), and I think everyone would like that.

btilly• 3 years ago

Anyone who is interested should also read https://twitter.com/aakashg0/status/1641976869460275201 for a rather different take.

It particularly interested me that Twitter under Musk is trying NOT to discuss Ukraine, and PENALIZES people who attempt to interact with those outside of their general political circle. I can give arguments for why they should do both, but I think both are ultimately bad ideas.

simplotek6• 3 years ago

It particularly interested me that Twitter under Musk is trying NOT to discuss Ukraine, and PENALIZES people who attempt to interact with those outside of their general political circle.

While Musk's Twitter explicitly censors references to Russia's genocide of Ukraine, Musk himself feigns ignorance and false indignation accusing the "western press" of insisting "on pushing such a lopsided view of the conflict".

https://twitter.com/VsimPohuy/status/1645699649003569152?t=v...

lumb63• 3 years ago

The two ideas aren’t necessarily in conflict with one another. In fact, they make sense to go hand-in-hand. Take for instance affirmative action or reparations. They’re instances of trying to correct for something that’s the opposite of how the group desires it, by doing the opposite of what supporters claim is being done to the marginalized group.

hathawsh2• 3 years ago

From your link (thanks!):

9. Making up words or misspelling hurts
Words that are identified as “unknown language” are given 0.01, which is a huge penalty.

Does that mean if I tweet about coding and use identifiers like "setUserName", which is not an English word, the tweet gets a huge penalty? If so, that's disappointing.

strken• 3 years ago

That jumped out at me as a possible misreading of the code. Is it detecting the language of the whole tweet, or just a word as the author claims?

Demoting a tweet that's entirely unidentifiable as any human language seems fair enough.

giraffe_lady2• 3 years ago

Yeah that struck me too. I can see the reasons why you'd want it but the collateral damage on that must be huge.

For example do they check for the common but nonstandard transliteration systems arabic speakers use? There have to be similar systems in other languages that don't use the roman or cyrillic alphabets too right?

Or for that matter what about languages twitter simply isn't aware of? There are thousands with native speakers after all, does this make it basically impossible for them to organically use twitter together?

stingraycharles• 3 years ago

The actual code comment doesn’t mention “words” but rather if the “tweet language” isn’t in one of the user’s “understandable” languages. As such, I assume your example is perfectly fine (would be extremely surprising if it wasn’t).

Whether the user implies the reader or author, I don’t know, I assume the reader as that would make most sense.

https://twitter.com/aakashg0/status/1641976943141699584?s=61...

Doddler2• 3 years ago

I think the correct reading of the code is that if English is your only language, non English tweets would be weighted 0.01x on your timeline.

randomwalker• 3 years ago

OP here. Unfortunately this thread is mostly misinformation. There were a bunch of viral threads from the growth hacker / influencer crowd, including this one, within hours of the code release with a very superficial understanding of the code (and how recsys work in general). That's partly what motivated me to write this article.

See here for a rebuttal of the main tweet in that thread (near the bottom of the article). https://solomonmg.github.io/post/twitter-the-algorithm/

ROTMetro2• 3 years ago

If this is for their Crisis Misinformation Policy why only one specific callout and specifically directed to Ukraine? Seems like a generous assumptions to make on your part that it's a nothing burger. The takeaway we should go with is that we now know that internally they are willing to programatically segment out Ukraine related topics. The question to me that this new knowledge should lead to is why a policy to segmenting this? (not to call immediately jump to 'nothing burger' or as you put it in the above post 'misinformation').

simonw• 3 years ago

I was not at all impressed with the analysis in that thread. It makes a bunch of assumptions that don't feel very thorough to me, but announces them as if they are unimpeachable facts.

Biggest example is this one:

"9. Making up words or misspelling hurts - Words that are identified as “unknown language” are given 0.01, which is a huge penalty."

The code in the screenshot for that looks like this:

    // Boost (demotion) if the tweet language is not one of user's
    // understandable languages, nor interface language.
    optional double unknownLanguageBoost = 0.01

That doesn't match the description of "Making up words or misspelling hurts" at all!

LegitShady• 3 years ago

input youtube thumbnail of cat in the hat enraged "DR SUESS CANCELLED?! TWITTER WON'T COMMENT!" ragebait youtuber.

treis• 3 years ago

The systemic racism people are going to go wild about this

eli• 3 years ago

I agree that seems presumptive. But of course twitter conveniently didn’t share the source code for how that factor is calculated. Turns out it’s hard to evaluate a big complex distributed system by looking at just one slice and no data.

Doddler2• 3 years ago

Yeah on the surface this looks more like "This user doesn't know German we should make German tweets much less likely to appear on their timeline" kind of thing. Just a complete misinterpretation, presented as fact without any supporting evidence.

muzani• 3 years ago

Ironically, sounds like a flaw of using Twitter as a medium. I'm sure he has more to say, or at least a more polite way to say it, but there's a character limit.

wunderland5• 3 years ago

It’s unclear if it penalizes discussion of Ukraine equally though.

There have been many stories that have come to light in the last few months. Merkel and Macron admitting the Minsk agreements were used to buy time for CIA and British to arm rebels since 2014 was big story. Large amounts of money the US has supplied Ukraine and lack of oversight to where this is going (the total US aid now surpasses Russia’s entire military budget per year). But this same poster (aakashg0) claims these stories have been suppressed, even though they would be counter to dominant narrative in western media.

I think algorithmic moderation on a particular topic is hard; you still need someone in there boosting the stories you want people to read and downplaying the stories you don’t.

stefan_3• 3 years ago

Tell us more, who are the "rebels" in this story and what arms did Merkel send?

(Is this what news in the PRC feel like?)

matthewdgreen4• 3 years ago

I mean, the fact that the Ukraine re-armed itself after Russia invaded their territory isn’t news, is it? I think it was reported on pretty substantially. And a good thing too since they were invaded a second time, this time with a strike towards their capital. I sort of assumed that was obvious public knowledge and don’t understand why people are making it into a “story.”

YeBanKo• 3 years ago

Merkel and Macron admitting the Minsk agreements were used to buy time for CIA and British to arm rebels since 2014 was big story.

Wtf are you talking about? What rebels in Ukraine did CIA and British arm? Or is one of those “it’s complicated” comments?

kspacewalk2• 3 years ago

Merkel and Macron admitting the Minsk agreements were used to buy time for CIA and British to arm rebels since 2014 was big story.

You don't mean "rebels" (who were/are in fact simply Russia proxies), you mean the other side, Ukraine.

Not sure what percentage of all arms delivered to Ukraine has come after February 2022, but it must be well north of 90%. So apparently all these plans to secretly and slowly arm Ukraine amounted to nothing, and it was the invasion itself that triggered the flow of arms. Some dastardly plan.

BTW always love the not-at-all-loaded use of "admitted".

Large amounts of money the US has supplied Ukraine and lack of oversight to where this is going

These are the kind of fact-free notions that seem to start their life in the TuckerCarlsonVerse and spread outward from there. Just saying there's "lack of oversight" does not make it so, particularly when there's plenty of oversight by the US gov't. In fact, oversight is one reason why the buildup of weapons supplies (in terms of quantity and weapons systems types) has been so gradual. The US wanted to make sure Ukraine's army knows what to do with the stuff and won't leave on the field of battle so that it ends up arming the enemy, like the Afghans did.

(the total US aid now surpasses Russia’s entire military budget per year).

That's completely meaningless. Russia's military budget pays for many times more personnel and weapons systems than the equivalent number as part of the US military budget, because the purchasing power of, say, $1 million is vastly different when it's spent by the Pentagon in the US, paying US prices and employing Americans, or by the Russian MOD, paying Russian prices and employing Russians.

XorNot• 3 years ago

This is an example of "a lie can get half way around the world before the truth gets its pants on".

I'm no fan of Musk these days, but there is plainly no evidence in the repo that Ukraine is being suppressed - the linked code is very obviously a model dispatch from an initial classification system, and it makes perfect sense Ukraine would need a call out there since a major war most likely would normally run afoul of profanity, violence, and calls to violence filters without actually violating them.

paganel• 3 years ago

It particularly interested me that Twitter under Musk is trying NOT to discuss Ukraine

Because of stuff like the NAFO trolls I can understand why that discussion is not brought to the top anymore.

Anyway, whoever is interested in the war going down there can most definitely have access to both sides' views, I personally find Twitter one of the few media/online outlets that still makes that possible (and props to them for that).

boringuser2• 3 years ago

That's a great way to eliminate misinformation, I don't really see the problem.

nonethewiser• 3 years ago

It particularly interested me that Twitter under Musk is trying NOT to discuss Ukraine

What makes you say this? I’m aware tweets about Ukraine are down ranked. This seems analogous to subreddits (such as world news, geopolitics, etc) stickying a live thread for the war. This is to prevent it from taking over the subreddit and it happens to slow down discussion.

Seems like the Ukraine war topic is similarly tuned on Twitter. Ukraine can still trend it just takes more activity than other topics.

cultureswitch3• 3 years ago

This is exactly what "trying not to discuss Ukraine" is. Without outright banning it.

When subreddits create a mandatory megathread, it is always with the purpose of kneecapping that specific topic. Putting a large topic into a single thread makes sharing information about the topic extremely difficult.

I'd like to add that there's no reason to believe that such topics that get buried into megathreads would have taken over a subreddit otherwise. The news cycle turns. For example, worldnews today would look no different if the Ukraine war topic had been allowed to breathe. At most it would have caused the minor inconvenience of going to page two for the fraction of people not interested at all in the Ukraine war at that time.

paulpauper2• 3 years ago

Tweets with hashtags and links do worse.

It would seem anti-woke tweets do very well. I see such tweets a lot when logged out.

Replying to an account with an unverified account is automatically collapsed. Only twitter blue accounts get to post replies in comments and not have those comments be collapsed. They can also post replies to their own tweets without the replies being collapsed.

SV_BubbleTime2• 3 years ago

It would seem anti-woke tweets do very well. I see such tweets a lot when logged out.

I have no idea if this is true. However... I have a coworker that was just banned from Reddit, her daughter is a high school swimmer, and there is a trans issue going on. I saw the comment she was banned for, while I don't necessarily agree with all of her points, it was absolutely not ban-worthy. She is waiting on an appeal that won't come. I've now seen with my own eyes that moderators and at least some admins at Reddit are shutting down anything that takes "the wrong position" on trans topics as "hate", when it's definitely not. Let's say over-woke to promote a consensus.

So, my question to you is this... How could we tell if anti-woke tweets do well on Twitter, or if they are overly censored and penalized on other sites?

How can you be sure it's not your reference and the systems you are comparing to are fair or not?

kenjackson• 3 years ago

There’s a lot of anti-trans stuff on Reddit. Here’s one with 800+ comments.

https://www.reddit.com/r/Conservative/comments/1241rxj/a_tra...

I’d have to see your coworker’s post to really know if it was hate speech.

liotier• 3 years ago

Tweets with hashtags and links do worse

I don't understand the rationale... My feeds are mostly comments with a link to the article, and following a hashtag is the way to form a community. What does Twitter want to be ? Some glorified public chatroom ?

appleaday1• 3 years ago

I think it shows us that I got the most thumbs-down for suggesting we rewrite everything in Python lol.

Jiocus2• 3 years ago

The article has a side-note saying:

“two models trained on the same data but with different algorithms will behave much more similarly than two models trained on different data with the same algorithm. That’s because the training data contains the patterns in user behavior that algorithms mine and use when recommending new content.”

Do we know how ‘true’ this statement is? Or how much % an algorithm must deviate to cause a % change in results? What about difference in data? If someone can point to somewhere to read up on this, I'd like to know!

burnished• 3 years ago

Yes, it is 'true'. The topic you want is fundamentals of machine learning.

Jiocus2• 3 years ago

Thanks, I'll check it out

hnaouesteuho• 3 years ago

I’d love to hear what former Twitter Timeline engineers can say about this.

13years• 3 years ago

Why are there no social media platforms that give the users more power over what they want the system to show them?

For example, this suggestion I posted on the Twitter algo

https://github.com/twitter/the-algorithm/issues/1363

nicce• 3 years ago

Because they generate less money. People don’t get hooked to content they did not know about.

ZunarJ5• 3 years ago

Mastodon??

SCdF• 3 years ago

Because, and I know it's a tired saying, but you aren't the user you're the product.

Advertisers, the actual users, get pretty precise tooling on who to show their ads to.

smeagull• 3 years ago

I really dislike that replying or commenting on things is seen as a positive indicator. This comment would be seen as positive engagement, I think the "algorithm" shown is fairly useless.

Especially since the weights are tied entirely to how well calibrated the individual predictive models are.

smcl4• 3 years ago

Man, all this effort when all most of us want is to see tweets from the people we follow in chronological order. Absolutely wild.

pharmakom• 3 years ago

Wouldn’t this be unusable? Just a huge flood of tweets that you have no hope of keeping on top of?

Alifatisk• 3 years ago

Check out Mastodons federated feed, it’s a mess but fun place.

t_von_doom• 3 years ago

This is what Twitter used to be - I would log in, scroll down until I had seen all updates since I last checked and then left the app. This was, however, before you got tweets in your feed from people you don’t follow (outside of retweets)

RugnirViking• 3 years ago

is that so different from the current system? Besides, its highly dependant on how many people you follow, and whether you care about seeing every tweet from every single one of them

mschuster91• 3 years ago

Only if you follow way too many accounts.

renewiltord• 3 years ago

You could/can get this with Twitter lists, and it's actually way better than anything else.

flohofwoe• 3 years ago

That's why you only follow specific people, and there's also a couple of tools left to tweak the timeline further (e.g. very useful: disable retweets from specific 'noisy' peeps).

smcl4• 3 years ago

If you follow thousands of people then maybe. But in that case you're going to miss tweets anyway, so if you've got way too many tweets buffered and you want to see more recent stuff you can opt to skip them.

Angostura• 3 years ago

Works just fine on Mastodon for me - I've got the people who I follow and I see all their posts in chronolgical order. There's also the federated which flows past swiftly and is quite fun to dip a net into when something interesting drifts past.

ChrisRR2• 3 years ago

Wow, you must be young. This is how the internet used to be, and people liked it.

cultureswitch3• 3 years ago

Is that the point?

Then again if that is indeed the point, just follow fewer accounts.

tbolt• 3 years ago

RIP third party clients.

teekert• 3 years ago

This actually made me stop Facebook. I so often saw something interesting, and was never able to find it back later. It's so incredibly frustrating. You leave the site in one state, you come back and there is not relation between what you left and what you find.

Thanks for breaking my habit I guess.

slazaro• 3 years ago

I use lists for that. I also get no promoted tweets or anything like that. Just chronological order tweets and retweets by the people that are in it.

smcl4• 3 years ago

That's a good idea actually. If I get bored on Mastodon and feel like I wanna give Twitter another whirl I'll try that. I went opened it up a few days back and saw a promoted post for every ~8 real tweets or so. Unreal.

ChrisRR2• 3 years ago

Aka the reason I stopped using facebook. I don't want you select who you think are my best friends, I don't want you to randomly sprinkle in group posts, I don't want to be reminded of random events from years in the past, I don't want notifications that a random friend I've not spoken to in years has posted a picture, I don't want you to exclude some friends from the list based on gut feeling.

I just want a chronological list of all of my friends updates.

v0idzer0• 3 years ago

You can already do this right now

KenArrari• 3 years ago

This is what people say but revealed preferences say otherwise.

noonething• 3 years ago

it tells us, musk bought twitter for reasons other than money...

Twitter showed us its algorithm – what does it tell us?

Related Stories