Twitter Pulls Back The Curtain On Its Recommendation Algorithm – SlashGear

At the operation’s core is a three-pronged recipe that begins with what Twitter calls “candidate sourcing.” It categorizes the best 1,500 Tweets into several lanes from both in-network and out-of-network sources, the former being those you follow and the latter being those you don’t.

It’s easy enough to pool together a bunch of Tweets from everyone you follow, but Twitter has to employ a logical formula for those you don’t. It uses grouping sub-algorithms called “social graphs” and “embedding spaces” to determine whether a Tweet might be relevant for you based on several factors, including whether your friends follow them or have engaged with their posts, as well as the activity of people who have similar engagement patterns to yours.

Another element of this approach is embedding spaces, which dives deeper into your specific profile in relation to the community at large. Twitter recognizes more than 145,000 “communities” across its network and uses your profile to determine which of these you primarily belong to. Once ascertained, it delivers the most relevant and influential Tweets from others in that community. This could be anything, from politics and finance to gaming and sports.

After ranking Tweets in order of engagement likelihood and filtering out bad apples and misfits based on content, author diversity, your personal mute preferences, and more, it’ll throw in a few personalized ads and deliver the resulting web to your timeline.