Twitter posts the code it claims determines which tweets people see, and why

Twitter is One of CEO Elon Musk’s many promisesposted on GitHub on Friday afternoon what it claims to be the code for a tweet recommendation algorithm.

Code posted under the GNU Affero General Public License v3.0 contains a lot of insight into what makes a tweet more or less likely to appear on a user’s timeline.

of Blog post accompanying the code release, Twitter’s engineering team (without a specific signature) said the system that determines “the top tweets that end up on a device’s For You timeline” is “composed of many interconnected services and jobs. It is said that Every time Twitter’s home screen refreshes, Twitter pulls “the best 1,500 tweets of him from a pool of hundreds of millions.”

The largest source of these tweets is “in-network sources,” or users someone is following. Tweets at the top of the pile are ranked based on the likelihood that users will engage with the tweet’s creator. The more likely they are to see their tweets on For You. For “out-of-network sources” that the user doesn’t follow, Twitter says it considers tweets that garnered engagement from people the user follows and tweets liked by people who like tweets similar to the user. .

Those who have already examined the code find considerations that raise even more questions. Not surprisingly, many post them on his Twitter itself.

Ólafur Waage, senior software developer at TurtleSec, a Norwegian software consulting service, describes some of the considerations that make a tweet a candidate for the “For You” section in “HomeTweetTypePredicates.scala”: pointing out.

• author_is_elon
• author_is_power_user
• author_is_democrat
• author_is_republican

Elsewhere in the code, a code comment, presumably left by a Twitter engineer, clarifies that these discriminating values ​​are “used for metrics collection only.” The comment said:

These creator ID lists are used for metrics collection only. We track how often we provide Tweets from these authors and how often those Tweets impressed our users. This is useful for A/B experimentation platforms to validate that you don’t ship changes to one group that adversely affect other groups.

The names of the objects in question, such as “DDGStatsDemocratsFeature” and “DDGStatsElonFeature”, appear to support this interpretation, although the available code may not confirm this. However, it’s interesting that Twitter checks and matches these variables. During a Twitter Spaces audio session, a Twitter engineer noticed that the metrics were using Democrat and Republican labels. Musk claimed he was unaware of the label until today, suggesting it shouldn’t be there.

Other considerations for Tweets include whether they are less than 30 minutes long, whether they have photos, and whether they are from “power users.” some people believe means “legacy” authenticated accounts.

Mr. Musk tweeted Alongside the company’s blog post about its recommendation algorithm, it was an “acid test” to see if an “independent third party” could “determine with reasonable accuracy what a user is likely to see.” claims to be

According to The New York Times, Twitter’s posting of the algorithm code comes just days after the social network’s extensive source code was discovered on GitHub, and may have been around for months. . Twitter then obtained a subpoena to force GitHub to disclose his GitHub contributor information.

According to the platformer’s report earlier this week, Twitter was utilizing a secret list of 35 top Twitter users, including President Biden, LeBron James, Ben Shapiro and Musk. Evidence of that list’s implementation, reportedly spurred in part by dissatisfaction with Musk’s own involvement, has thus far not been found in Twitter’s posted codebase. I did.

Most notably, codes arrive hours before “Legacy Verified” users (those with a blue checkmark indicating authenticity or attention before Musk purchased their service). Some users connected to governments and large organizations may Check marks in other colorsonly $8/month Twitter Blue subscribers will receive “Priority Ranking of Conversations.” among other features.

All of these changes happen to happen on April 1st or April Fool’s Day.