Among the changes that the social network Twitter has undergone since its purchase by Elon Musk is its algorithm. In recent days there has been much discussion about the For You tab which, for many, suggests irrelevant content to users. For this reason, Twitter released the source code of its algorithm on GitHub to learn what is behind these recommendations and how it works.
Recent changes and announcements on Twitter have caused many users to migrate to other platforms. The publication about the functioning of its algorithm seeks to retain such users, in addition to dispel doubts about the reliability and transparency of Musk’s social network.
The latest Twitter news has generated a new controversy. The removal of the verified check mark is causing many companies to block comments on their profiles to avoid scams. Some companies even inform their customers that the service is provided by other networks or Whatsapp. Users are not sure that they are interacting with official profiles, so they are easy prey to virtual scams.
How does the Twitter algorithm work?
It should be noted that, prior to Twitter’s release of its algorithm, part of its source code was published on GitHub without authorization. The social network had to send a court notice for the content to be taken down. In addition to this, Twitter also requested information about the user who disseminated the information and those who downloaded it. A few days later, at the end of March, the source code of the social network’s algorithm was finally made public.
To choose the content that appears in For You there are many interconnected services and jobs. Basically, the algorithm works in 3 stages:
Popular tweets are analyzed based on likes, retweets and replies. Then another team is in charge of analyzing the possibilities for the user to interact with that account. The trending topics and communities in which it usually participates are also taken into account. This first stage is known as candidate selection.
A second stage is in which each of the tweets is classified through a machine learning model. Those with higher rankings will have a better chance of appearing in the recommended section. Finally, the third and last stage filters the publications based on the user’s preferences. Tweets from blocked users, already seen in the past or else classified as NSFW, not appropriate for work, are left out here. The number of tweets from the same account is also reduced, in addition to other processes.
Is the source code explaining how Twitter’s algorithm works real?
So far there is nothing to indicate that the released source code is not the one that was released. However, there are doubts about whether it is the source code that is actually used in Twitter. The company has reported that there is more information on its blog about how its algorithm works. On the other hand, it was also mentioned that no information was included on how to classify advertisements for inclusion in the recommendations.