The content recommendation algorithm that powers online short-video platform TikTok is back in the spotlight after the United States ordered its Chinese owner, ByteDance, to sell the app's U.S. assets under penalty of a national ban.
Here's how it works and why it's generated more discussion than technologies used by rivals, such as Meta's Instagram, Google's YouTube, and Snapchat:
ALGORITHMS
The algorithms are considered essential to ByteDance's overall business, which would prefer to shut down the app rather than sell it, Reuters reported, citing sources.
China changed its export laws in 2020, giving it the right to approve any export of algorithms and source codes, adding a layer of complexity to any attempt to sell the app.
Academics and former employees of the company have said that it is not just the algorithms, but also the way they work with the short-form video format, that have made TikTok successful around the world.
THIS IS ALSO THE APPLICATION
Before the emergence of TikTok, many believed that technology linking a user's social connections was the secret sauce of a successful social media app, given the popularity of Facebook and Meta's Instagram.
But TikTok showed that an algorithm, based on understanding a user's interests, could be more powerful. Rather than basing their algorithm on the “social graph” like Meta, TikTok executives, including CEO Shou Zi Chew, said their algorithm was based on “signals of interest.”
While its rivals have similar interest-based algorithms, TikTok is able to increase the efficiency of the algorithm through the short video format, said Catalina Goanta, an associate professor at Utrecht University.
“Their recommendation system is very common. But what really sets TikTok apart as an app is its design and content,” she added.
The short video format allows TikTok's algorithm to become much more dynamic and even track changes in user preferences and interests over time, going as far as what a user may like for a certain period of time. time of day.
QUICK DATA COLLECTION
Additionally, the short-form video format allows TikTok to learn user preferences much more quickly, said Jason Fung, former head of TikTok's gaming unit.
“Because it's a short video, you can gather data on user preferences much more quickly than on YouTube, where the average length of a video is perhaps less than 10 minutes,” he said. he declared.
The fact that TikTok positioned itself from the start as an app designed for mobile devices also gave it an advantage over competing platforms that had to adapt their interfaces from computer screens.
TikTok's early entry into the short-form video market also gave the company a significant first-mover advantage. Instagram only launched Reels in 2020, while YouTube launched Shorts in 2021, two platforms that lag behind TikTok in terms of data and product development experience.
ALLOWS EXPLORATION
TikTok also regularly recommends content that does not match users' interests, something the company's management has repeatedly said is essential to TikTok's user experience.
A study published last month by American and German researchers found that TikTok's algorithm “exploits user interests in 30 to 50 percent of recommended videos” after examining data from 347 TikTok users and five automated robots.
“This finding indicates that TikTok's algorithm chooses to recommend a large number of exploration videos in an attempt to either better infer the user's interests or to maximize user retention by recommending many videos that are outside of the user's (known) interests,” the researchers wrote in the paper titled “TikTok and the Art of Personalization.”
MOBILIZES USERS IN GROUPS
Ari Lightman, a professor at Carnegie Mellon University, said another effective tactic TikTok employs is encouraging its users to form groups publicly through hashtags.
By encouraging users to form public groups, TikTok can better understand its users' behavior, interests, alignment, and ideology.
If TikTok ends up being banned in the United States, Mr. Lightman believes that American tech giants certainly have the ability to replicate TikTok with their own products, but that replicating the user culture enabled by TikTok could be the most difficult task.
THE CHINESE ADVANTAGE
TikTok's recommendation algorithm was also largely carried over from its Chinese sister app Douyin, launched in 2016. Although ByteDance often insists that TikTok and Douyin are separate apps, a source with direct knowledge of the question stated that the two algorithms remain similar to this day.
Douyin's AI was boosted by the company's ability to take advantage of low labor costs in China, allowing it to hire numerous content annotators to carefully label the entire content and users of the platform.
“In 2018 and 2019, Douyin worked to label each user. So they labeled each video clip manually. Then they would label their users based on the video they watched,” said Yikai Li, director of the advertising agency Nativex and former director of ByteDance. “They also applied this tactic on TikTok.
While hiring annotators to label data is now a common and important practice for AI companies, ByteDance was the first to adopt this strategy.
“It's a lot of work to sort through these labels. It's very laborious,” he said. “So Chinese companies have an advantage. You can afford to employ a lot more people. The cost is lower than for North American companies.
