X is expanding its open-source initiative by releasing significantly more of the technology behind its recommendation and ranking systems, including code used to determine what users see in their “For You” feeds. The social media platform said the newly released code will provide greater insight into how posts are selected, filtered and ranked.

The material is being published on GitHub under the Apache 2.0 license and includes additional details about the platform’s ranking systems, model configurations and filtering processes.

X Vice President of Product Keith Coleman said the release will allow developers and researchers to examine much of the technology responsible for assembling personalized feeds.

Some of the systems can also be run outside the company, giving researchers an opportunity to study how the ranking process works.
The company is also introducing a transparency feature that will allow some users to examine how its ranking systems have affected their accounts and posts.

Initially, the tool will be available to a limited group of accounts that are at least one year old. Eligible users who have published at least 10 posts during the previous month will be able to download aggregated account information in JSON format.

The information may reveal whether certain labels have been applied to an account or its posts during the previous month.
X says users without technical expertise can also use artificial intelligence tools to help interpret the information and understand what the data means.

External researchers were given access to the system ahead of the public release. According to X, researchers were able to train and operate the company’s Phoenix scoring system using the newly available code.

Developers will also be able to propose changes to the open-source code through GitHub pull requests. X engineers will review those submissions and may incorporate useful improvements into the platform’s ranking technology. However, not every component of X’s recommendation infrastructure has been released.

Systems that use Grok to assess whether posts could violate platform rules remain private, with the company arguing that revealing those systems could allow bad actors to circumvent its safeguards and spread spam or other prohibited content.

The move comes amid years of debate over the influence of social media algorithms on political discourse, elections and misinformation. Critics have also questioned whether platforms treat users and political viewpoints equally.

Before Elon Musk acquired Twitter, some political figures accused the platform of suppressing conservative accounts through so-called “shadowbanning,” a claim the company repeatedly denied.

X now says its goal is to make its recommendation system sufficiently transparent for the public to examine, criticize and assess whether content is being distributed fairly.

Despite the increased transparency surrounding parts of its recommendation technology, the platform has faced criticism in other areas since becoming privately owned.

Information about user numbers, revenue, growth and government requests for content removal has generally been less accessible than it was when the company was publicly traded.

X has since merged with SpaceX, with some user metrics once again being made public. The company says its broader objective is to create a recommendation system that can be independently examined and improved through public scrutiny.

Source: TechCrunch