Real-Time & Personalized Notifications @Twitter

Twitter developed a personalized fanout algorithm to send you notifications only about the best content from the people you follow.

The algorithm relies on engagement data:

• Your current interests = recent hashtags, topics and people you have engaged with.
• Your top followings = the people you really want to hear from. These are periodically calculated offline because they don't change that often (it takes time for relationships to develop between users).

Using this data, Twitter notifies you about a tweet only if it matches your current interests and if it's from one of your top followings.

Twitter chooses which users to send notifications to based on two main factors:

1. The volume of notifications a user usually receives:

• Users with lots of notifications don't get any more notifications.
• Users with few notifications receive recommendations to help them get the most value of Twitter.

2. The likelihood of a user engaging with notifications:

• Twitter keeps track of your historical open-rate for notifications i.e. whether you liked them and engaged with them.

Twitter sends notifications with content recommendations only to users that have a high enough chance of being interested in them.

Not sending them to all users is a win-win:

• Users who don't like notifications don't get notified.
• Active users are not overwhelmed with more notifications.
• Twitter doesn't waste resources calculating recommendations for users who aren't going to see them anyway.