The secret sauce behind TikTok’s recommendation algorithm

Eugene Wei explains why TikTok’s recommendation algorithm is so good. It has to do with an “algorithm-friendly” design.


September 29, 2020

Eugene Wei — a former product exec at Amazon and Oculus — has penned the most concise explanation of what makes TikTok tick (sorry).

His 2nd entry in a 3-part series on the video app is a must-read for anyone trying to understand TikTok’s recommendation engine (AKA the For Your Page [FYP]).

The secret?

An “algorithm-friendly” design.

Here are some key takeaways:

  1. TikTok’s actual machine learning (ML) recommendation algorithm isn’t out of the ordinary
  2. However, the data inputs into TikTok’s algorithm are differentiated and — all things equal — better data inputs create better algorithms
  3. To get the most valuable inputs possible for its algorithm, TikTok’s design is very unique: It is only one video at a time with a number of indicators as to whether or not the user likes it (length of viewing, re-watches, likes, comments, song choice, video subject, shares)
  4. Typically, UX design is meant to be user-friendly. However, to improve its algorithm, TikTok has made its product a bit less user-friendly (users scrolling through multiple pieces of content is a more frictionless experience than just a single video view)
  5. Eugene calls this product decision an “algorithm-friendly” design
  6. Compare this with a traditional social feed (Twitter, Facebook), both of which offer an endless scroll of content. The user inputs are less clear (“liking” something doesn’t transmit a ton of information)
  7. With such clear signals — whether positive or negative — TikTok can quickly understand a user’s preference and serve up more similar content
  8. This creates a tight feedback loop and kicks off the flywheel that continually improves TikTok’s recommendations and data inputs 

P.S. If you prefer audio, listen to his explanation on the a16z podcast.

Daily briefings, straight to your inbox

Business and tech news in 5 minutes or less

Join over 1 million people who read The Hustle

Psst

How'd Bezos build a billion dollar empire?

In 1994, Jeff Bezos discovered a shocking stat: Internet usage grew 2,300% per year.

Data shows where markets are headed.

And that’s why we built Trends — to show you up-and-coming market opportunities about to explode. Interested?

Join us, it's free.

Look, you came to this site because you saw something cool. But here’s the deal. This site is actually a daily email that covers the important news in business, tech, and culture.

So, if you like what you’re reading, give the email a try.