TikTok’s video algorithm is one of the most praised inventions in the tech world and has become a cornerstone of the social media’s viral growth. It keeps amazing users with its ability to suggest videos that match the user’s preference. It is so good that there have been talks of the parent company licensing the technology.
However, the secret of the technology is gradually coming into the open, with a report published by the New York Times. The news outfit got access to privileged internal documentation that lays out how the video prediction engine works and is designed.
According to the document, the software is built around these key things: keep users engaged, keep them coming back, retain the creators, and ensure they make money from their content. However, these key points are summarized under four main goals; user value, long-term value, creator value, and platform value.
The document NYT accessed was titled TikTok Algo 101 and was created by the Beijing-based engineering team and has been confirmed for its authenticity. However, a company source claimed the document was meant for a non-technical audience within the company.
TikTok charted a different path for itself, as its goal is not for keeping you connected to your friends but finding entertainment. This is why the app offers a seemingly endless flow of videos.
One of the major differences of TikTok is the ease of creating content. The app gives creators background music to base their performance on. This is why it is so popular with low or no-budget entertainers. Other social media platforms are scrambling to respond, with Instagram rolling out Reelz while attempting to tempt users on board with lavish cash rewards.
Previously, TikTok has shared how the engine works without going into the specifics of its trade secret. It admitted the algorithm considers user information like likes, comments, etc. It also looks at the videos watched by a user, like captions, hashtags, and sounds.
Users have attested to the effectiveness of the algorithm, with claims that TikTok can push you to a side politically or in the sexuality debate, attract you to celebrity videos, trap you into foreign cultures, etc.
Not satisfied with TikTok’s simple explanation of how its tech works, people outside TikTok have long tried to figure out how the video suggestion algorithm works. For example, a recent report claims TikTok uses how long users spend on videos mainly to serve up recommendations. It also points out that a major disadvantage of the method is creating rabbit holes that users can disappear into. Some of these rabbit holes are dangerous and could lead to negative things like suicide or self-harm among the young. TikTok has clarified that it is aware of the problem and counteracts it by quickly deleting inappropriate content.
What TikTok describes mildly as user retention has been called a less flattering name of addiction by analysts, who have long believed that TikTok’s algorithm poses a social problem. They are using this leaked document to back their claims.
After going through the TikTok document at the request of an NYT writer, Guillaume Chaslot, founder of Algo Transparency, said, “This system means that watch time is key. The algorithm tries to get people addicted rather than giving them what they really want.” He touched on the dangers of the platform, especially on young ones, “I think it’s a crazy idea to let TikTok’s algorithm steer the life of our kids. Each video a kid watches, TikTok gains a piece of information on him. In a few hours, the algorithm can detect his musical tastes, his physical attraction, if he’s depressed, if he might be into drugs, and many other sensitive information. There’s a high risk that some of this information will be used against him. It could potentially be used to micro-target him or make him more addicted to the platform.”
However, not everybody is impressed with TikTok’s recommendation engine, proving it is impossible to please everyone, after all. Julian McAuley, a computer science professor, describes the algorithm as “totally reasonable, but traditional stuff” that has the benefit of machine learning combined with “fantastic volumes of data, highly engaged users, and a setting where users are amenable to consuming algorithmically recommended content (think how few other settings have all of these characteristics!). Not some algorithmic magic.”