While it is true that Twitter has been under the microscope since Elon Musk took the reins, for the past few weeks the conversation has been focused on the “For You” tab. The recommendations in this section convince few. Either because it shows posts that are not of interest to the user, or because at some point they considered only showing tweets from people subscribed to Twitter Blue.
To try to calm the waters of a community that is increasingly vocal with the sudden changes in the social network, Elon Musk has released the source code of the algorithm which, precisely, is in charge of recommending publications. All files are available through GitHub.
The idea, clearly, is that the experts inspect the code to make sure that, in effect, there is no anomalous or manipulated operation as is speculated. But, knowing how Elon Musk has moved since he took over Twitter, Can we really trust that this same code is implemented in the social network? Everyone will have an opinion about it…
The steps that Twitter follows before recommending a tweet
Those directed by Elon Musk explain how their recommendation system works, in basic terms. First, it must be said that different services are involved with specific purposes. One of them, for example, analyzes the popularity of tweets based on the user who posted it and the interactions it received (likesretweets, replies, etc.).
«Twitter’s recommendation algorithm is a set of services that are responsible for building and offering the Home Timeline.»
However, before showing it to you on the Timeline, another system evaluate the possibility of you interacting with that account directly in the future.
Also, Twitter analyzes which are the topics and communities with which you most often participate. If you are a person who constantly searches and enters the trending topics When it comes to movies, the algorithm analyzes the trends on that topic and, based on them, shows you posts that might be relevant to you.
The entire process above is known as “Selection of candidates”. Once having this information, the Twitter algorithm proceeds to classify each tweet using a machine learning model. Those who are at the top of the ranking will have a better chance of being recommended.
But before we get to the “For You” tab, one last step needs to be applied. It’s about a filter subject to your preferences. If you’ve blocked an account in the past whose posts might be of interest to you, there’s no point in recommending them to you. Tweets containing content classified as NSFW (not appropriate for work) or those you have seen in the past may also be hidden.
Of course, it’s a pretty straightforward explanation of how Twitter’s recommendation algorithm works in the big picture. Although behind it there is a whole network of information and quite complex analysis.
You can be sure that, over the next few days, there will be people inspecting the code for some strange feature—or taking inspiration from it to implement it in their own projects. It remains for us to trust that the code is the same as the one that runs on Twitter’s servers.