Photo Credit: Alexander Shatov
X is expanding its open-source codebase to include its ranking algorithm, providing transparency on how “going viral” works on the former Twitter.
Elon Musk’s X, formerly Twitter, released the source code that determines how posts are recommended to users in a bid to increase transparency. The move arrives amid Musk’s broader shift to “improve fairness” on his platform. He explained that the decision to open-source the code aimed to enhance transparency and gather feedback for improvement.
The algorithm, dubbed “Phoenix,” examines a user’s recent viewing history to predict how that person will interact with a post from an account they don’t follow. Posts on the user’s For You page are ranked based in part on the way Phoenix predicts that the user will interact with them.
If Phoenix predicts that a user will share the post by copying its URL, the post is given a weight approximately 40 times larger than if Phoenix predicts that a post will receive a “like.” Replies, quotes, and shares by direct message each carry a weight of five, which is ten times the 0.5 weight assigned to a predicted “like.” The weight of follows is equivalent to eight likes, while the weight for reposts is two likes.
Combining those predicted probabilities with their respective weights, Phoenix produces a ranking score, which X then uses to decide the order of posts in that particular user’s For You feed. Phoenix assigns negative weights to posts that it predicts will elicit a “displeasure” response from the user, severely limiting their visibility.
Phoenix views a predicted “report” as 468 times a predicted “like” in the opposite direction. A predicted mute carries a negative weight of around 118 times that of a predicted like. “Not interested” interactions are treated as 86 times larger (in the negative), while a block is about 62 times larger.
Interestingly, the system indicates that posting deliberately inflammatory comments meant to evoke an angry response is less likely to grow a following. This might come as a surprise, considering the platform’s reputation since Musk’s takeover.
X also announced a pilot feature called “Under the Hood,” which lets participating users see aggregate information about any labels applied to their account and content that could affect visibility. Based on its source code, the platform classifies content and accounts across categories such as spam and adult material, information that can then be used by a separate system to determine whether a post is shown.
So far, X says the feature has only been activated for a “randomized test group of eligible accounts.” A wider rollout depends on the feedback the company receives from the initial group.
Back in January, Musk derided the existing algorithm, calling it “dumb” and vowing to make the system more transparent for users.
“At least you can see us struggle to make it better in real-time and with transparency,” Musk wrote. “No other social media companies do this.”
Leave a Reply