How TikTok algorithms work

 

Before selling the American branch, TikTok revealed the essence of the recommendation algorithms in the application feed. All in order to prove to the public and potential buyers that the algorithms, in fact, are not “evil, manipulative, Chinese and spy”, but simply super-effective and engaging.

Point

When users first open TikTok, they are shown eight popular videos in different styles, with different music and on different topics.

Each new iteration of 8 videos will be formed taking into account how the user interacted with the previous eight-the image on the video with which the user interacted the most, text stickers in the video, captions and hashtags to it, music and sounds are taken into account. Plus — the device, account settings, language, and geography of the user are taken into account.

And all this is only the first stage.

Once TikTok collects enough data about a user, the app can match their preferences with those of other users, and assign the new user to an existing group with similar interests.

Videos are grouped in exactly the same way.

As a result, using machine learning, the algorithm shows the video to the user depending on its proximity to other groups of users with similar content preferences. Collaborative filtering, in general.

Additionally, the algorithm works to avoid duplication of content in the feed.

Captivity of recommendations

TikTok recognizes that the ability of algorithms to take into account preferences so effectively means that users are essentially creating “recommendation bubbles”, beyond which it is difficult to go — existing user preferences are strengthened, recommendations become more narrow and focused, content becomes less diverse,” monotonous”, fundamentally different, fresh content does not get into the feed.

TikTok itself studies these “bubbles”: average life time, methods of formation and destruction. Developers consider this phenomenon undesirable: sometimes it can spoil the user experience and contribute to the avalanche of fake content among the audience. In other respects, this is also the sin of YouTube.

How do I use this information?

Be sure to look in the account analytics section “what is popular with my subscribers”. It’s basically a list of videos that you’re most likely in the same feed with right now, not just a selection of popular videos.

This section is probably much more useful for audience profiling than the standard geo, gender, and age statistics. In fact, you can quickly understand what kind of “bubble” you are in. The audience may not be the one you were originally aiming for.

Consider these aspects and increase your conversion rate. If TikTok is new to you, it’s time to write to your manager at 1win partners and get detailed recommendations on how to earn money on this platform.