X is significantly expanding the amount of its recommendation system that is available as open-source code.
The company is publishing more details about the algorithm behind its “For You” timeline and its core ranking engine, while also introducing a new tool that will show users whether its ranking systems have affected their accounts or posts.
The source code for the “For You” timeline is being published on GitHub under the Apache 2.0 license. X says the expanded release includes additional details such as model configuration, filtering systems and core ranking components, including parameters used to determine how different signals are weighted when deciding which posts appear in users’ feeds.
According to X, the expanded codebase is now roughly 10 to 15 times larger than the previous release. Developers can see the core ranking code used to select and rank posts for individual users, as well as systems designed to filter potentially problematic or rule-breaking content. Some of the ranking components can also be run outside X, allowing researchers and developers to examine how they work.
X is also introducing a new transparency tool through its “Under the Hood” settings page. Users who have posted at least 10 times during the previous month will eventually be able to download aggregate information about their account as a JSON file. The data will indicate whether labels were applied to their account or posts during the previous calendar month.
The feature will initially be tested with a limited group of accounts that are at least one year old before being made available more broadly. X says users without technical knowledge can also use an AI tool to help interpret the information by providing the JSON data along with the company’s open-source code.
Before the release, X gave external researchers access to its expanded codebase. The researchers were able to train and run the company’s Phoenix scoring system using the publicly available code, which X describes as an important step in its transparency efforts.
The GitHub repository will also allow developers to submit changes through pull requests. X engineers will review those contributions and decide whether they should be incorporated into the platform’s algorithm.
Not every part of X’s recommendation infrastructure is being released. Systems that use Grok to predict whether posts could violate platform rules are being kept private. X says withholding those components is intended to prevent bad actors from using the information to bypass its moderation systems and spread spam.
The changes come amid long-running concerns over how social media recommendation systems can influence political discussions, elections, misinformation and the visibility of individual posts. X says its goal is to allow the public to audit and critique how posts are distributed and determine whether the system is operating fairly.
If this article helped you, please consider supporting our work. Every small contribution keeps Abijita.com independent and running.
While X is making more of its recommendation technology publicly available, the company has faced criticism over transparency in other areas since Elon Musk took the platform private. User metrics, revenue information and government takedown requests have historically been less publicly available under private ownership. X has since merged with SpaceX, making some user metrics public again.
