X opens the black box of its recommendation algorithm: launches algorithm inspection tool, allowing users to view account "demotion" labels

Odaily Planet Daily News The X platform has officially open-sourced the code related to how posts are ranked for exposure in the "For You" recommendation timeline, and launched a new transparency tool that allows users to see whether their account or posts have been tagged with labels that may limit visibility. The goal is to let users judge for themselves whether the platform is restricting the spread of their content, whether the recommendation system is fair, and why they see specific content.

At the same time, X has also introduced a transparency page called "Under the Hood." Eligible users can view the labels applied to their accounts and posts over the past month through this page, and download related data. X describes the two newly launched features as "two complementary pieces of the puzzle":

1. The open-source code reveals how the "For You" timeline is generated, including the content labels that affect post exposure and how they work;

2. The transparency tool allows users to check whether their own accounts or posts have restriction labels that affect visibility.

X stated that it has invited third-party recommendation system experts to review and test this code release in advance, in order to ensure that the information disclosure is clearer and more practical. The company said it will continue to update these transparency mechanisms regularly as the recommendation algorithm evolves. The features are still in a pilot phase and are only open to randomly selected eligible accounts. Initial test subjects are users who have been registered for more than a year and have published more than 10 posts in the past month. X plans to gradually expand coverage based on user feedback.