The Limited Times

Now you can see non-English news...

Twitter reveals how part of its recommendation algorithm works

2023-04-03T16:30:53.265Z


Elon Musk released the source code for the algorithm that decides what content appears on users' feeds on Friday afternoon.


Elon Musk promised it a year ago.

It is now done.

The boss of Twitter released part of the source code for the blue bird platform on Github on Friday afternoon.

Specifically, these are the algorithms that are used to recommend tweets to users in the

“For You”

section .

“Being transparent about code is going to be very embarrassing at first, but should lead to quick improvements in the quality of recommendations,” added

the whimsical billionaire.

He hopes that some users spot security flaws and suggest modifications to optimize its operation.

Collect, classify, filter

In a blog post, the company reveals gathering

“the best tweets among a hundred million recommendations”.

Around 1,500 posts are collected to build a page that is 50% tweets from followed accounts and 50% off-network profiles.

These tweets are then ranked to keep only those with the best chance of causing positive engagement, by a like, a retweet or a reply.

The selection is then filtered by further criteria and restrictions, such as whether the posts have negative feedback, are from mostly the same account, or blocked or muted users.

Other content is then added to the feed, such as sponsored tweets or recommendations of accounts to follow.

Internet users have already started to crack the code, like Aakash Gupta.

The user notices that a like has much more impact than a retweet or a reply to a tweet.

If images and videos help increase the visibility of a tweet, links do just the opposite.

The tweet then falls into the category of spam, unless the account is extremely followed.

Publications containing disinformation are not highlighted by the platform.

Finally, subscribers to the paid Twitter Blue offer have an advantage.

Another user, Jean Manchun Wong, notes that

"Twitter's algorithm explicitly indicates if the author of the tweet is Elon Musk."

The researcher also specifies that the platform assigns labels to certain accounts indicating whether the author of the tweet is an intensive user or whether he is a Republican or a Democrat.

Elon Musk has since assured that he would withdraw the offending part of the code.

"That shouldn't be part of it

," he said.

“Consider it deleted.”

Reassure advertisers

The transparency measure could appease authorities and lawmakers who want to know how the platform works.

Many elected officials, especially Republicans, believe that it is biased against them, while human rights associations criticize the service, especially under Elon Musk, for not fighting enough against misinformation and harassment.

The Tesla founder is also trying to win back advertisers on his network.

Despite a slight increase in daily users since the start of 2022, Twitter's revenue has fallen 50% since October due to a

"massive drop"

in advertising, Elon Musk said.

The billionaire's attitude scares advertisers.

The brands, which account for 90% of the platform's revenue, believe that Twitter no longer guarantees a safe storefront for their ads.

Read alsoTwitter faced with the flight of advertisers in France

For now, only the codes for the

“for you”

page recommendation algorithm have been released.

Neither the search algorithm nor the code that powers the ad recommendations has been shared.

“We have excluded any code that would compromise the security and privacy of users or the ability to protect our platform

,” the firm justified in its blog post.

The rest of the algorithms will be available

"in the coming weeks

," Elon Musk said in a tweet.

Source: lefigaro

All news articles on 2023-04-03

You may like

News/Politics 2024-03-02T16:24:25.639Z
Life/Entertain 2024-02-22T14:32:30.394Z

Trends 24h

Latest

© Communities 2019 - Privacy

The information on this site is from external sources that are not under our control.
The inclusion of any links does not necessarily imply a recommendation or endorse the views expressed within them.