3 / 8 Social & Feed Systems

Sources Rankers And Filters

The open sourced X recommendation stack splits timeline construction into candidate sourcing, ranking, and filtering. Roughly half of the posts in a For You timeline come from the in-network source; the rest arrive from out-of-network sources before anything is scored.

More than one option is correct. Which of these components produce candidate posts, as opposed to scoring or filtering posts that some other component already produced?

A

tweepcred, which computes a PageRank style reputation score per user

B

navi, the Rust model serving runtime

C

follow-recommendations-service (FRS), which surfaces accounts to follow and posts from them

D

search-index (Earlybird), which finds and ranks in-network posts and supplies about half of the timeline

E

user-tweet-entity-graph (UTEG), which walks the recent engagement graph on top of GraphJet

F

visibility-filters, which applies legal, quality and trust rules