Key facts
- X has released the source code for its 'For You' page recommendation algorithm, 'Phoenix'.
- The algorithm assigns positive weights to predicted shares, replies, quotes, follows, and reposts.
- Predicted URL copies are weighted 40 times higher than predicted likes.
- Negative interactions such as 'report,' 'mute,' 'not interested,' and 'block' carry significant negative weights.
- A pilot feature called 'Under the Hood' will provide users with aggregate data on content labels impacting visibility.
X, formerly Twitter, has released the source code for its recommendation algorithm, known as 'Phoenix,' which dictates the content shown on its 'For You' page. The company stated the move is part of a broader effort to enhance transparency and allow users to assess the fairness and reach of content on the platform.
The 'Phoenix' algorithm ranks posts by predicting user interactions. A predicted share via URL copy is weighted approximately 40 times more than a predicted 'like.' Other positive interactions like replies, quotes, direct message shares, follows, and reposts also contribute to a post's ranking score with varying weights.
Conversely, predicted negative interactions significantly limit a post's visibility. A predicted 'report' carries a negative weight 468 times that of a predicted 'like,' while a predicted 'mute' is 118 times as negative, and a 'not interested' interaction is 86 times as negative. A block is approximately 62 times as negative as a like.
In addition to releasing the code, X is piloting a feature called 'Under the Hood,' which will allow participating users to view aggregate information about labels applied to their accounts and posts that may affect their visibility. These labels, which can include categories like spam and adult material, are determined by X's content classification systems.
Elon Musk, who acquired the platform last year, has previously called the algorithm 'dumb' and pledged greater transparency. The release of the source code is the latest step in this direction, with Musk aiming to improve fairness and solicit feedback for platform enhancements.