Power and limitations of TikTok’s recommendation algorithm

It seems as though TikTok’s sophisticated recommendation algorithm is reading your mind when it comes to suggesting videos for you to watch. Its hyper-personalized “For You” feed gives off an almost psychic vibe, as though it knows people very well.

Does it, however, truly pick up on your innermost desires and thoughts? A detailed analysis indicates that the real picture is more nuanced. The TikTok algorithm does not quickly determine your true desires; instead, it develops your interests over time to enhance involvement.

In contrast to other platforms, TikTok’s algorithm can quickly determine a user’s preferences based on just one crucial signal. Every moment that passes when you pause or replay a video gives the algorithm crucial information. Afterward, it makes use of that information to present interesting, customized material that leads viewers down “rabbit holes” that are unique to their tastes.

The phrase “down the rabbit hole” effectively conveys the idea of the TikTok algorithm rapidly leading users into increasingly specific and sometimes problematic content, in an almost uncontrollable manner, like falling down the “rabbit hole” into an alternative world, as referenced in Alice in Wonderland. This idiomatic expression captures the sense of being drawn into a deep, immersive, and perhaps unwanted experience, much like Alice’s journey from the real world into the fantastical realm she discovers at the bottom of the rabbit hole.

As explained here, this degree of customization has advantages and disadvantages for marketers. The algorithm on TikTok can make advertisements and branding initiatives seem eerily current. However, without human supervision, content can potentially stray into more extreme niches. Comprehending the system’s operation is crucial to establishing a connection with consumers while avoiding dangerous detours.

TikTok algorithm

Some broad information regarding TikTok’s recommendation system’s work has been made public. To recommend new videos, it takes into account elements like likes, comments, captions, sounds, and hashtags. Experts from outside the field have also attempted to decipher the algorithm.

According to a Wall Street Journal analysis, TikTok heavily influences watch time in order to entice users to scroll endlessly. This engagement-driven strategy may occasionally lead younger viewers to objectionable content, such as material that encourages self-harm. According to TikTok, it actively removes any videos that violate its guidelines.

TikTok’s popularity is partly due to how simple it is to create videos with integrated memes and music. Its ability to identify users’ interests and direct them toward specific “sides” is startlingly accurate for a large number of users.

Several headlines stating the algorithm’s nearly supernatural ability to understand someone better than they know themselves serve as evidence of the app’s deep insights into people’s inner lives.

The article “The TikTok Algorithm Knew My Sexuality Better Than I Did” is a notable example of how the platform’s suggestions may provide users with incredibly intimate self-reflections, bringing feelings and thoughts from subconscious levels to the surface.

These anecdotal reports demonstrate how precisely the algorithm has mapped the human psyche.

TikTok’s recommendations

As I’ve already indicated, the way that TikTok’s For You stream presents videos that correspond with users’ unspoken feelings and thoughts seems almost uncanny. However, this is not an accident. It is the outcome of a highly developed recommendation system that the company has spent nearly ten years perfecting.

Knowing the history of TikTok’s algorithm is helpful in order to fully understand it. ByteDance, a Chinese startup, owns TikTok. Douyin is an app that used a similar suggestion mechanism in the past. ByteDance relaunched Douyin as TikTok after entering new markets. However, the powerful algorithm stayed the same.

The New York Times was able to receive a leaked internal document that stated TikTok’s main goals are to increase “user value,” “long-term user value,” “creator value,” and “platform value.”

It specifically aims to optimize for “time spent” (the amount of time a user spends on the app) and “retention,” two metrics that are closely related. The goal of the algorithm is to maximize the amount of time you spend watching videos.

The recommendation formula

The document reveals that TikTok calculates a score for each video based on a formula factoring in:

  • Predicted likes: the number of likes a video is expected to get, based on machine learning predictions;
  • Predicted comments: the expected number of comments;
  • Predicted playtime: the predicted total playtime if shown to a user;
  • Played: whether the video was played or not.

The basic formula is:

Plike x Vlike + Pcomment x Vcomment + Eplaytime x Vplaytime + Pplay x Vplay

In this case, the “V” variables stand for weights that modify the importance of each prediction, and the “P” variables stand for the predictions. This is “highly simplified,” according to the document, and the real formula is far more intricate.

tiktok algorithm goals
TikTok algorithm goals