X has made a notable change to the open-source algorithm that shapes the "For You" feed. The new filter, activated on October 22 in the code published on GitHub, aims to reduce candidate posts for the feed in a more balanced manner based on their sources.
Previously, every time a user opened their feed, the system created a large pool of candidates consisting of followed accounts, popular content, posts liked by similar users, and new posts. However, because the number of posts the algorithm could score was limited to 2,800, some content was filtered out without being evaluated when the pool became crowded.
According to Peter Mick, who tracks these changes, new posts were often among the first to be filtered out in the old system because they remained at the end of the ranking. This made it difficult for posts from accounts with low follower counts to gain visibility.
The newly added filter, called FairShareRankingFilter, groups candidate content by source and, when necessary, makes cuts from each group at similar rates. This is intended to prevent a single content source from being systematically pushed back.
As reported by Mick, the "new posts" category generally covers posts that are less than 2 hours old, have fewer than 8 likes, and fewer than 500 views. Therefore, the change could be significant in terms of allowing posts from smaller accounts to at least enter the evaluation stage by the algorithm.
On the other hand, it is unclear how significant the impact of the update will be. The code does not specify how often the 2,800-post limit is reached. It is stated that no new changes have been made to the interaction weights used in post scoring, such as likes, replies, quotes, and link copies, with this update.
News Source: 12punto
