TikTok’s algorithm has many advantages compared to Meta and Google’s products, causing it to often fall into the sights of technology circles and US officials.
TikTok’s content recommendation algorithm continues to become the focus after the US government passed a law forcing ByteDance to sell the application within 9 months or be banned from operating in the US.
Reuters Quoting four sources, ByteDance will not divest TikTok to any other company and is ready to close the application in the worst case scenario. They emphasized that the Chinese company agreed to do this because it did not want to give up the core algorithm – the “secret formula” that created TikTok’s success that no other social network has.
Algorithm and application design
Experts and former TikTok employees say the app’s global success stems not only from the algorithm, but also has to do with how it coordinates with the short video format on the platform.
Before TikTok appeared, many people believed that connecting users’ relationships was the secret to creating success for social networks, like Facebook and Instagram did.
TikTok logo on a user’s phone screen. Image: Reuters
However, TikTok shows that operating based on knowledge of user preferences and interests brings a much greater advantage. Instead of creating an algorithm based on social relationships like Facebook, TikTok leaders, including CEO Shou Zi Chew, built an algorithm based on “interest signals”.
“Many social networks have developed algorithms based on user preferences, but TikTok maximizes its effectiveness thanks to its short video format. Their recommendation system sets it apart in design and content.” “, commented Catalina Goanta, lecturer at Utrecht University in the Netherlands. In particular, the TikTok algorithm is capable of tracking small changes in user preferences over a long period of time, even capable of determining what they want to see at certain times of the day.
Collect data quickly
Jason Fung, former head of TikTok’s gaming division, said the short video format helps the app learn user preferences much faster than its competitors.
“We can gather interest data faster than YouTube, which has an average video length of nearly 10 minutes. Imagine the ability to collect data every few seconds, versus nearly 10 minutes at a time.” , he said.
TikTok started out as an application for mobile devices, helping to gain an advantage over platforms that had to find ways to convert the interface from computer to smartphone.
Entering the short video market early also gives TikTok many pioneer advantages. Instagram launched Reels in 2020, and YouTube’s Shorts in 2021. Both are several years behind TikTok in terms of product development experience and user database.
Allows exploration
TikTok also regularly recommends content beyond user interests, which the company’s leadership believes will play an essential role in each person’s experience.
According to a report published last month by US and German researchers, based on data from 347 users and 5 automated bots on TikTok, the platform’s algorithm exploits user preferences in only 30-50% of videos. propose.
“This finding suggests that the TikTok algorithm recommends a large number of exploratory videos, to better understand user needs or maximize retention through interesting unrelated videos,” the researchers said. research said.
Include users in groups
Ari Lightman, a professor at Carnegie Mellon University in the US, points out that one of TikTok’s effective tactics is to encourage users to focus on public groups through hashtags. This method helps them learn more about users’ behavior, preferences and thoughts.
Lightman said that leading US technology corporations have the ability to copy and replace TikTok’s position if this application is completely banned, but will face many challenges if they want to recreate the user culture TikTok has built. .
China’s advantage
TikTok’s recommendation algorithm is developed based on the Douyin application launched in China in 2016. ByteDance has repeatedly said TikTok and Douyin are independent applications, but a source with knowledge of the matter said the algorithms of both there are many similarities.
Douyin’s AI is boosted by cheap labor in China, allowing the company to recruit large numbers of people to label content and users on the platform. “During 2018-2019, Douyin sought to label each person. They manually marked each video, then marked users based on videos watched. This method also applies to TikTok,” Yikai Li, a former director at ByteDance, revealed.
Hiring data labelers is now a common practice among AI businesses, but ByteDance is one of the pioneers in this field.
“Statisticing and organizing such labeling data takes a lot of effort. Chinese companies have an advantage due to cheap and large human resources. The cost of this activity in China is much lower than in other countries.” North American businesses,” Li said.
Diep Anh (According to Reuters)
