X is expanding access to the code behind its recommendation system while giving users tools to check how ranking systems affect their accounts and posts.
X is expanding its open-source codebase to include the code behind its “For You” timeline and core ranking engine. The code is available on GitHub under the Apache 2.0 licence, giving developers and researchers access to more details about how posts are ranked and displayed.
The new codebase includes the model architecture, filters, and core ranking system. It also includes the weights applied to various signals that influence which posts are shown in users’ feeds. X has also allowed external researchers studying recommendation systems to train the Phoenix scoring system using the open-source code.
Along with the code release, X is offering users features that allow them to analyse how its ranking system has affected their accounts or posts. The “Under the Hood” transparency feature will be available through the app’s settings. Users who have posted at least 10 times during the previous month can download aggregated statistics in JSON format showing whether their account or post received any labels during the previous calendar month.
The transparency initiative also includes access to the source code used for recommendations. However, some parts of the system are not being made public, including systems that use the Grok machine learning model to predict whether a post may violate a rule. X says these details are being withheld to prevent malicious users from exploiting the system.
The changes expand X’s efforts to improve transparency around its recommendation system while retaining restrictions on information that could be used to manipulate the platform. Concerns remain about how X’s algorithms can influence politics, elections and the spread of disinformation.
