It was a longtime promise of Elon Musk. As confusion reigns between accounts that still allow their certifications to be displayed, between historical users and those who paid for Twitter Blue’s subscription, Twitter’s source code was partially revealed on Friday, March 31.
“For this release, we sought the highest degree of transparency, excluding any code that could compromise security and privacy or our ability to protect our platform from harmful actors, including code that would undermine our efforts to combat child exploitation and manipulation.” . specified the social network in a publication loading
Last month, Twitter announced that snippets of its code had been partially leaked. Some of the social network’s code is now freely accessible on the GitHub platform, including the algorithm for recommending tweets visible in the “For you” section.
50% tweets between followed accounts
In a blog postTwitter reveals that 1,500 tweets out of hundreds of millions are chosen to populate this section, with 50% tweets coming from people followed by users, and the other half from unfollowed accounts, “although this may vary depending on the user.”
“The classification [des tweets] It is achieved through a ‘neural network’ of 48 million parameters, continuously trained on tweet interactions to seek engagement (such as ‘likes’, retweets and replies)”, specifies the social network.
Users do not see the full selection of 1,500 tweets, as they are filtered by other criteria and restrictions, such as whether the posts have negative comments, are mostly from the same account, blocked or muted users.
Political triage and VIP users
Twitter would also use the accounts of personalities categorized as important (from the athlete LeBron James, to the United States Democratic congresswoman Alexandria Ocasio-Cortez to the conservative radio host Ben Shapiro) to verify the changes in its recommendation algorithm, putting them before more than the others, as revealed last week by the American platform Platformer. This list of “VIP” accounts was not revealed by the code revealed by Twitter, says the specialized site Gizmodo.
The platform’s recommendation algorithm, on the other hand, shows that it assigns labels to users, whether they’re Elon Musk, a major user, or… a Republican or a Democrat, as noted by the researcher Jane Manchun Wong.
During an official audio sharing session on Twitter Space, one of the platform’s developers claimed that it was a simple means of collecting statistics, notes Matt Binder on the Mashable site. The journalist adds, however, that a note in the code states that these metrics are collected to ensure that changes to Twitter’s algorithm do not negatively influence any of these four groups.
Elon Musk has since assured that he would remove the offending part of the code, although it has not been made public in its entirety, according to a report. platform reporter. “That shouldn’t be a part of this,” the CEO said. “Consider it gone.”
Source: BFM TV
