Twitter has fulfilled Elon Musk's promise by releasing the code that selects which tweets appear on a user's timeline to GitHub. The company also published a blog post, which details how the algorithm chooses and ranks tweets for the For You timeline in three main stages.
The first stage involves collecting the best tweets from different recommendation sources, followed by ranking those tweets with a machine learning model. The final stage filters out blocked users, previously viewed tweets, and NSFW content before displaying them on the user's timeline.
The post elaborates on each step, revealing that the goal of the first stage is to feature an equal mix of tweets from people the user follows (In-Network) and accounts they don't follow (out-of-network). The ranking is optimized for positive engagement, such as likes, retweets, and replies, and the final stage ensures that the user isn't bombarded with tweets from the same account.
Although researchers are already analyzing the code, the most comprehensive information regarding Twitter's algorithm will likely be obtained by scrutinizing the code.
Elon Musk, the CEO of Twitter, had been promising to open-source the algorithm for some time. Back on March 24th, 2022, before he owned the company, he conducted a poll on Twitter asking his followers if they thought Twitter's algorithm should be open-sourced. Roughly 83 percent of respondents voted in favor of the idea. In February, Musk pledged to make the code public within a week but later postponed the deadline to March 31st.
Elon Musk announced on Twitter that the released code contains most of Twitter's recommendation algorithm, with the remaining portion to be disclosed in the future. Musk hopes that third-party analysts will be able to determine, with reasonable accuracy, which tweets will be displayed to users. In a Space discussing the algorithm's release, Musk emphasized his desire to create the least gameable system on the internet, akin to the widely-used open-source project Linux. He also added that the ultimate goal is to maximize users' unregretted time on the platform.
Musk has been preparing his followers for potential disappointment with the algorithm, acknowledging that it is complex and not entirely understood even within Twitter. He suggested that people may discover silly things in the code but promised to address any issues as they arise. Musk also noted that providing code transparency would be initially embarrassing but could lead to significant improvement in recommendation quality.
It is worth noting that code transparency and open-sourcing are two distinct concepts. While code transparency allows users to view the processes responsible for selecting tweets for their timelines, open-sourcing entails allowing the community to submit its own code for consideration and use the algorithm in other projects. While Elon Musk has stated that Twitter's algorithm will be open source, the company must take further steps to earn that label. This involves implementing governance systems that determine which pull requests to approve, which user-reported issues merit attention, and how to prevent bad actors from manipulating the code for their own gain.
While Twitter is working on building tools to manage community suggestions and sync changes to its internal repository, it remains to be seen whether the company will actually accept any community code. Elon Musk's promises to poll users before making major decisions have not always been fulfilled, so it remains to be seen whether the company will follow through on its commitment to open-sourcing the algorithm.
Elon Musk has been openly critical of Twitter's previous management and how they handled moderation and recommendation. He has orchestrated a barrage of stories to expose the platform's "free speech suppression," although it mostly showed how content moderation works. Now that he's in charge, he has faced backlash from users annoyed about their For You pages featuring his tweets prominently, and from his conservative supporters, who are increasingly concerned about their lack of engagement on the platform. Musk has claimed that negative and hateful content is being "max deboosted" in the new recommendation algorithms, but outside analysts without access to the code have disputed this.
In addition, Twitter is potentially facing competition from the open-source community. Decentralized social network Mastodon has been gaining traction, and Twitter co-founder Jack Dorsey is backing another similar project called Bluesky, which is built on top of an open-source protocol.