In a new blogpost, Twitter has revealed how its algorithm
works to rank and filter tweets on users’ timelines. Explaining the process in
a simple manner, Twitter says that it occurs in three main stages.
According to Twitter, its algorithmic system first gathers
the “best” tweets from various recommendation sources. On the For You timeline,
for instance, Twitter aims to display around 50% tweets from accounts that a
user follows and 50% tweets from accounts that a user doesn’t follow.
In the ranking stage, the system uses a machine learning
model to rank tweets based on engagement analytics including Likes, Retweets,
and Replies.
In the final step, tweets are filtered out to be displayed on users’ timelines. The filtering happens in categories like tweets from people one has blocked, tweets one has already seen, or tweets that are not safe or appropriate. The aim of this process is also to ensure that a user isn’t seeing too many tweets from the same account.
In addition to providing this information, CEO Elon
Musk announced that will be open source starting today. The code, which Musk believes to be “overly
complex,” is meant to provide users and researchers with transparency into the mechanisms
that decide which tweets show up on people’s timelines.
Most of the recommendation algorithm will be made open source today. The rest will follow.
— Elon Musk (@elonmusk) March 31, 2023
Acid test is that independent third parties should be able to determine, with reasonable accuracy, what will probably be shown to users.
No doubt, many embarrassing issues will be… https://t.co/41U4oexIev
Twitter has shared the code on GitHub, and while it doesn’t contain
every little detail, the insights it provides are pretty significant as it contains
core models and features that extract latent information from tweets, users,
and engagement data on Twitter.