TikTok's Recommendations

TikTok. It’s taken the world by storm. One of the only Chinese apps to have become insanely popular globally. Matthew Brennan’s book, Attention Factory, says TikTok’s recommendation system is the differentiator.

 

But first, let’s see what TikTok is not. It is not a social media app, unlike Facebook or Twitter. In a social media app, you create an account (to identify yourself), then pick your “friends” or people you want to “follow”, and then the algorithm decides how to prioritize the feeds from those folks.

 

In TikTok, on the other hand, you don’t pick anyone as a friend or someone to follow. The algorithm decides what to show you next: it could be from literally anyone on the planet (Since you don’t tell the app whom to follow, the app isn’t restricted to showing only the videos from those people).

 

In fact, if you’re a new user, TikTok does not insist you create an account:

“(This approach) allowed people the freedom to experience TikTok without committing.”

This is possible because its recommendations are based on what you did with the videos so far (Watched them again? Skipped midway?), i.e., your actions. Not your identity or your friends. But if they like it, people will create an account. Why? Well, if they switch phones, they’d want the app to continue working as before, and to do that, they’d need to identify themselves via an ID. And once they create an account, other options are available (to share, to like), which act as additional feedback to the app’s algorithm to learn their preferences.

 

Here’s how TikTok’s founder, Yiming Zhang, explains the difference between search and recommendations:

“(It is the) shift from people looking for information to information looking for people.”

Going further, he says that search is a great way to find information, except for one problem. Here’s how the book describes the problem with search:

“You must know the term you wish to search for… This make search less ideal for… entertainment, which hold a strong discovery element.”

Recommendation systems rely on two approaches. First, there’s content-based filtering (same/similar content as topics you already liked). And then there’s collaborative filtering (what else others who liked the same topic as you enjoyed).

 

Recommendation algorithms aren’t easy to get right. Obviously. But TikTok got it done so well that users don’t need to pick what to watch next, they trust the algorithm to show them something they’d like. In the age of the Internet, users have ceded control to the system to decide. Imagine that!

“TikTok could be viewed as the true smartphone successor to television.”

It’s like TV in the sense that you don’t control what’ll come next. But unlike the TV, TikTok is able to show you an endless stream of content you like love.