How the Twitter algorithm works, which Elon Musk made open source

Elon Musk first said it about a year ago: “Twitter’s algorithm should be open source,” explained the then aspiring number one of the social network at TED 2022. In recent months, the management of the platform founded by Jack Dorsey by the owner of Tesla has been talked about a lot, among others controversy every time different (the latest is with the New York Times) and the introduction of the Blue paid service.

A promisehowever, has been respected: the algorithm code that governs the Twitter homepage has actually been made available online. The code has been posted on GitHub And the company released a post on the official blog to explain, in greater detail than usual, how the social network’s recommendation system works.

A clarificationfirst of all: when we talk about Twitter’s algorithm and recommendation system, we are referring to the formula that manages the homepage of the social network, which today, in its default For You section, it’s a mix of tweets from profiles that the user follows and on topics they might find interesting. At the base are user data, which the artificial intelligence behind the system analyzes to select and classify the most interesting content for each profile. This filtering operation takes place starting from personal information, but also from a series of objective factorswhich Twitter has made public.

How tweet selection works

Let’s start from the beginning. Every time we open Twitter, the system is called a select a series of contents for us (1500 at a time, let the social network know) to compose our homepage. It does so by drawing from two main groups of tweets, the first of which is that of the profiles we follow: in this case, it takes into consideration the user’s relationship with whoever posted the tweet. In other words, if it happens often interact or read content posted by a particular profile, it is very likely that this will continue to appear constantly in our timeline.

The second group it is much broader, because it includes everything else, i.e. tweets that come from unfollowed profiles. The approach is twofold: on the one hand, contents with which the user’s network has interacted in some way are taken into consideration; on the other, tweets on topics to which the user himself has shown a some kind of interest. In short, if we read and interact a lot with content that the system identifies as related to technology, it is very likely that the system will continue to show us tweets of that kind, even from profiles we don’t follow.

The factors that increase the visibility of a tweet

At this point, the question arises: how does the system organize these 1500 contents? The underlying principle has to do with interaction. The algorithm chooses tweets based on probability that the user opens them or leaves a like or a share.

However, not all interactions are created equal. According to what is learned from the code, each like increases the visibility of a content by 30 times, a retweets of 20, a comment only by 1. Other positive factors, which favor the probability that a tweet will be seen by more users, are the presence of a multimedia content, video or image that is; in the latter case, visibility doubles. It warrants an increase, as Musk also announced, too signing up for Twitter Blue, which is considered by the algorithm a spread multiplier. And that, starting from the next few weeks, will also be decisive for the possibility of appearing in the For You section.

There are, however, too negative factors: basically, putting a link is not well received by the system, as well as a too high number of hashtags or having been blocked or muted by many users. Also post content on topics outside your niche can decrease the reach of a certain content: if the system identifies a user in the Technology category (for example), a tweet about football could be penalized and distributed to a smaller number of users.

Social networks

Why does Twitter show us tweets from profiles we don’t follow?

by Francesco Marino

March 16, 2023

Doubts about Elon Musk and Ukraine

An algorithm is a complex system. And apart from the attempt to simplify, within lines of code published by Twitter hide a series of even curious questions, which give a sense of the very high number of parameters hidden behind the scenes of social networks.

A curiosity it concerns a particular attention that the algorithm has for Musk himself. In the code, there is a section in which the system has the instruction to divide users into specific categories: there are power users, or strong users, then the two political groups of Democrats and Republicans and there is Elon Musk . Well yes: Twitter exclusively monitors the performance of the content published by its number 1, which more than once is in fact intervened to orient the algorithm in its favor. In this case, they let the company know, it is only a tracking for analysis purposes.

Another interesting topic is the disinformation, a very hot topic for Twitter and social networks in general. In this sense, the algorithm identifies, through an automatic system, violent, toxic or disinformation tweets, and penalizes them. What has caught the eye of many is that, within these categories, there is also the war in Ukraine. According to what can be read in the code, the distribution of content that talks about the conflict is decreased by the system. On this issue, Musk himself had an ambivalent attitude: first, with the provision of the Starlink service to the Kiev government, then with the disputes with Zelensky over the peace plan that Twitter’s number one himself had proposed.