The recommendation pipeline has four steps:

1. Candidate sourcing - Fetch the best tweets from different recommendation sources
2. Ranking - Rank tweets such that the most relevant ones appear first in the feed.
3. Filtering - Apply heuristics and filters to create a balanced and diverse feed.
4. Mixing non-tweet content like Ads, Follow Recommendations, and Onboarding prompts.

The recommendation pipeline runs 5 billion times per day and completes in under 1.5 seconds on average.

Gets the best 1,500 recent tweets:

• 50% from people you follow (In-Network)
• 50% from people you don't follow (Out-of-Network)

In-Network

• Uses a logistic regression model to rank the recent tweets of people you follow. The higher the likelihood of engagement between you and another user, the more of their tweets will be included.

Out-of-network

• Uses a social graph of engagements to find tweets that people you follow have recently engaged with, and tweets liked by people who are similar to you.
• Uses an embedding space of users and tweets to determine what tweets and users are similar to your interests.