“This system means that viewing time is key. The algorithm tries to get people addicted instead of giving them what they really want, “said Guillaume Chaslot, founder of Algo Transparencia, a Paris-based group that has studied YouTube’s recommendation system and has a dark view of it. Product effect on children in particular Mr. Chaslot reviewed the TikTok document at my request.
“I think it’s a crazy idea to let the TikTok algorithm run our children’s lives,” he said. “Every video a child watches, TikTok gets information about him. In a few hours, the algorithm can detect your musical tastes, your physical attraction, if you are depressed, if you could be high, and a lot of other sensitive information. There is a high risk that some of this information will be used against you. It could potentially be used to micro-target him or make him more addicted to the platform. “
The doc says that viewing time is not the only factor that TikTok considers. The document provides a rough equation for how videos are scored, summarizing a prediction driven by machine learning and actual user behavior for each of the three bits of data: likes, comments, and watch time, as well as an indication that the video has played:
Plike X Vlike + Pcomment X Vcomment + Eplaytime X Vplaytime + Pplay X Vplay
“The recommendation system scores all videos based on this equation and returns the videos with the highest scores to users,” the document says. “For the sake of brevity, the equation shown in this document is greatly simplified. The actual equation in use is much more complicated, but the logic behind it is the same. “
The document illustrates in detail how the company modifies its system to identify and suppress “lookalike baits” – videos designed to play with the algorithm by explicitly asking people to like them – and how the company thinks of more nuanced questions.
“Some authors may have some cultural references in their videos and users can only understand those references better by watching more videos of the author. Therefore, the total value that a user watches all those videos is higher than the values of watching each video added together, ”says the document. “Another example: if a user likes a certain type of video, but the application continues to present him with the same type, he will quickly get bored and close the application. In this case, the total value created by the user who watches the same type of videos is less than that of watching each video, because repetition leads to boredom. “
“There are two solutions to this problem,” the document continues. Make some assumptions and decompose the value into the value equation. For example, in terms of repeated exposure, we could add a value ‘same_author_seen’, and for the boredom problem, we could also add a negative value ‘same_tag_today’. Other solutions besides the value equation may also work, such as forced recommendation on users for feeding and dispersing, etc. For example, the problem of boredom can be solved by dispersion. “