That’s it: Twitter’s recommendation algorithm has been made public! Objective ? Guarantee total transparency and help community managers and social media managers implement an effective social media strategy as demonstrated in this training in community management…
A promise from the firm’s CEO
Over the past few months, a “For You” thread has been implemented on the Twitter interface. Its algorithms are suspected of classifying users according to their political affiliation or importance. Elon Musk, the current owner of the social network, then announced his intention to make public and open source the lines of code at the core of this timeline of recommendations.
And it is now done on the blue bird side! A portion of its code was exposed on Github on March 31, 2023.
“For this release, we aim for the highest level of transparency, excluding any code that may compromise security and privacy or our ability to protect our platform from harmful actors,” said the platform through a blog post.
Remember, Twitter’s recent actions have created various controversies, including the issue of certificates that caused confusion among users as well as the previously mentioned recommendation algorithm.
Additionally, Twitter said last month that snippets of its code had been leaked …
How do “for you” threads work on Twitter?
The posting of the source code made it possible to assimilate the workings of the For You thread. home mixer, As it is called, it is based on three steps to provide recommendations to the users.
first thing is to collect “Best Tweets from various recommended sources”, Therefore, About 1500 tweets are collected To appear in your timeline made up of 50% of people’s Tweets You’ve followed and the other half are tweets from people outside your network.
Of course, Twitter users don’t see this entire selection of tweets, These are filtered keeping in mind several criteria and restrictionsThis is part of the second phase.
In fact, the algorithm goes on to classify tweets made through machine learning models. The goal is to optimize the way tweets are sorted to select only those most likely to drive positive engagement through likes, retweets or replies.
The goal of the third phase is Filter selection of Tweets so that they don’t offer you publications that you’ve already seen from accounts you’ve blocked and/or that contain content that isn’t compatible with your workplace.
Home Mixer also helps avoid displaying too many tweets from one person. Plus, other content can be added to your feed: sponsored tweets or even recommendations from accounts to follow.
According to the social network with 330 million monthly users, this process is repeated about 5 billion times a day and takes about 1.5 seconds. Twitter also adds that a “48 million parameter neural network trained to continuously seek engagement (likes, retweets, replies)was necessary to obtain the classification of tweets.
Sorting by political affiliation and VIP users
Twitter will also use the accounts of users classified as important within the platform (from athlete LeBron James, to Democratic Congresswoman Alexandria Ocasio-Cortez and conservative radio host Ben Shapiro) to verify changes brought to its recommendation algorithm.
According to the American media platformer, the social network will thus favor these accounts more than others. Featured site Gizmodo specifies that this list of “VIP” accounts does not appear in the code published on Github.
In contrast, the platform’s recommendation algorithm suggests users are given a label, be it CEO Elon Musk or another prominent user, as research by blogger Jan Manchun indicates. Wong.
During an official audio session on Twitter Spaces, one of the platform’s developers insisted that it was merely a means to collect statistics on users.
However, Mashable Media reporter Matt Binder pointed out that a note in the code claims that these collections are used to ensure that changes to the algorithms do not harm “VIP” users.
Since then, Elon Musk has assured that he will remove the offending part of the code, according to a Platformer journalist, although the latter has not been made public.