Today is the day: In an effort to boost transparency between the public and the social media platform, CEO Elon Musk has made Twitter’s recommendation algorithm open source...for some reason.
It would appear that Musk is true to his word and actually did release Twitter’s recommendation algorithm on GitHub, as well as a write-up from the Twitter Team describing what the code algorithm actually does on Twitter’s blog. Twitter claims that, at its core, the algorithm that recommends tweets—which it called “Home Mixer”—pulls them from sources using a process called “candidate sourcing,” ranks them against a machine learning model, and then filters them based on benchmarks like users you’ve blocked and if the content is NSFW.
Twitter says the Home Mixer candidate sources consist of both people you follow and people you don’t, and the For You tab usually consists of tweets from 50% of each. Home Mixer pulls about 1,500 tweets during this first step. When the algorithm ranks the tweets, it ignores what the source of the tweet is, and “takes into account thousands of features” in order to assign them one of ten labels—each label corresponds to a certain probability of engagement. Home Mixer will then filter out tweets based on its own internal sieves and then throw it onto your For You page.
Musk previously announced that the algorithm Twitter uses to recommend tweets to users would go open source on March 31 in an effort to provide code transparency to users which will, hopefully, lead to trust. Whether the majority of Twitter users are even interested in learning more about the code remains yet to be seen. Regardless, he himself acknowledges that providing code transparency will be “incredibly embarrassing at first” as Twitter continues to find ways to fine tune the algorithm.
“Our ‘algorithm’ is overly complex & not fully understood internally. People will discover many silly things, but we’ll patch issues as soon as they’re found!” Musk tweeted earlier this month. “We’re developing a simplified approach to serve more compelling tweets, but it’s still a work in progress. That’ll also be open source.”
One thing that doesn’t appear to have been made public is the list of VIPs that Twitter pushes to users. Platformer reported that Twitter has a rotating list of noteworthy users that included everyone from Representative Alexandria Ocasio-Cortez to YouTuber Mr. Beast to Daily Wire founder Ben Shapiro, according to internal emails. Twitter allegedly uses this list to monitor changes to the website’s recommendation algorithm by seemingly increasing the visibility of these power users at will. This list stands in stark contrast to Elon Musk’s hare-brained quest to treat everyone equally.
Some portions of Twitter’s code were actually already made open source—just not in the way Musk had planned. Earlier this week, news broke that portions of Twitter’s source code were uploaded to GitHub according to court documents, in which Twitter demanded that the offending code portions be removed from the online repository. Twitter also filed a DMCA takedown notice with GitHub, which was complied with the request to remove the code.
It’s not clear how much of the code was uploaded or for how long it was left online, though The New York Times reported that it could have been months. The chunks of source code was uploaded by a user who goes by “FreeSpeechEnthusiast”—perhaps a nod to Elon Musk’s fallible attempts at being a “free speech absolutist.” Whoever FreeSpeechEnthusiast is may be gleaned soon enough, however, as Twitter successfully obtained a subpoena to unmask the hacker. The subpoena also requires GitHub to provide Twitter with the identity and information on anyone who downloaded the code during the time that it was uploaded.