How TikTok Reads Your Mind

“This system means that playback time is key. The algorithm tries to get people hooked instead of giving them what they really want, ”said Guillaume Chaslot, founder of Algo Transparency, a Paris-based group that has studied YouTube’s recommendation system and the impact of the Product on dismal opinions considered children, in particular. Mr. Chaslot checked the TikTok document at my request.

“I think letting TikTok’s algorithm run our children’s lives is a crazy idea,” he said. “With every video that a child watches, TikTok gains information about them. In a few hours, the algorithm can identify his musical tastes, physical attraction, depression, drug addiction and a lot of other sensitive information. There is a high risk that some of this information could be used against him. It could possibly be used to target him or make him more dependent on the platform. “

The document says that watch time isn’t the only factor TikTok takes into account. The document provides a rough equation for scoring videos that summarizes a machine learning-driven prediction and actual user behavior for each of the three bits of data: likes, comments, and playing time, plus an indication that the video was played:

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

“The recommendation system rates all videos based on this equation and returns videos with the highest ratings to users,” the document says. “For the sake of brevity, the equation shown in this document is greatly simplified. The equation used is much more complicated, but the logic behind it is the same. “

The document details how the company tweaks its system to identify and suppress “like bait” – videos designed to play the algorithm by explicitly asking people to like it – and how the company thinks through more nuanced questions.

“Some authors may have cultural references in their videos, and the only way for users to understand those references better is to watch more of the author’s videos. Therefore, the total value that a user watches all of these videos is greater than the sum of the values ​​for each individual video, ”the document says. “Another example: If a user likes a certain type of video, but the app continues to push them the same type, they would quickly get bored and close the app. In this case, the total value the user gets by watching the same type of video is less than watching each and every video because the repetition leads to boredom. “

“There are two solutions to this problem,” the document continues. “Make some assumptions and break the value down into the value equation. For example, we could add a value “same_author_seen” for repeated exposure and a negative value “same_tag_today” for the problem of boredom. In addition to the value equation, other solutions can also work, such as B. a forced recommendation in the users for U-food and distribution, etc. For example, the problem of boredom can be solved by distribution. “