Expert: TikTok algorithm is not unique, but buyers can’t wait to develop it themselves
The source said that if the recommendation algorithm is not available, the rapid sale of TikTok’s US business is “unlikely.” Although many experts do not believe that Bytedance’s recommendation algorithm is unique, users and investors may not want to wait for a new algorithm to be developed .
When ByteDance acquired Musical.ly, a karaoke app, in 2018 and rebuilt it into TikTok, the industry generally believed that this was just another ordinary short video app for American teenagers.
Today TikTok is the world’s most downloaded application, and its popularity is so high that it has become the focus of the US government.
Prior to this, the US government required ByteDance to divest TikTok’s US business. Musical.ly may have allowed ByteDance to find a foothold in the US market. What makes TikTok’s business take off is ByteDance’s artificial intelligence recommendation algorithm, which can provide relevant content after screening based on users’ interests and activities.
Since its establishment in 2012, Bytedance has always been a supporter of content recommendation systems and has widely adopted this algorithm in other products such as Toutiao. According to the information disclosed by ByteDance in June, TikTok mainly considered three factors when developing the recommendation algorithm:
- User interaction on the application, such as liking a video or following an account;
- What is the content of interest? In the short video, it is information such as music and hashtags;
- And the environment the user is in, such as language preference, country and region settings, and device type.
At the same time, the TikTok application will also push some video content that users are not directly interested in.
The source said that if the recommendation algorithm is not available, the rapid sale of TikTok’s US business is “unlikely.”
Why is this algorithm so important? Is it because others cannot imitate it?
After ByteDance acquired Musical.ly and merged it with TikTok, it introduced the recommendation algorithm to the platform, significantly increasing the time users spend on the application. Product expert Eugene Wei said on his personal blog that this change is “subtle.”
According to data from market research company App Annie, last year, TikTok users on the Android mobile platform spent a total of 68 billion hours on this app, more than three times the amount in the previous year. According to the lawsuit filed by ByteDance against the U.S. government at the end of August, as of June 2020, TikTok’s monthly active users in the U.S. market are close to 92 million, which is more than 8 times that of January 2018.
According to data from the market analysis company Sensor Tower, TikTok was the world’s most downloaded non-game application in the first half of 2020, with more than 596 million installs, excluding Douyin.
Huang Jinhui, a professor of engineering at the Chinese University of Hong Kong and an artificial intelligence expert, said that although the basic algorithms used by TikTok are similar to those in other technology company applications, each company will add special features to the artificial intelligence engine, which is different.
Huang Jinhui doesn’t think TikTok’s artificial intelligence engine is unique. He said that it may take about a year to build a new recommendation system for TikTok based on new user data, but the loss of existing tools will have a “very significant impact” on TikTok’s current valuation.
“This technology is only effective when the algorithms and user data are running well. The reason why byte-beating applications have an advantage in the competition is partly because of their user data.” Technology blogger Hao Peiqiang said. He used to be a software engineer and now provides consulting services for companies.
Huang Jinhui said that some users and investors may not want to wait for the development of new algorithms. He said, “You can’t wait for the TikTok team to re-develop the algorithm, because TikTok is already very popular.” “It’s like your favorite TV show is stopped due to technical issues… I don’t think users will accept this. “
“For bidders like Microsoft and Wal-Mart, they want to acquire this application and make it work immediately,” Huang Jinhui said. “But if they need to wait a while for it to work well, maybe they won’t want to buy it again.”
“Tiktok wouldn’t exist without its own recommendation system, but that doesn’t entirely mean that this system is special,” said Julian McAuley, an expert in related fields and an associate professor at the University of California, San Diego.
“Early adopters of recommendation systems are also e-commerce companies. For example, Amazon has used recommendation technology for nearly 20 years, but the early system only involves simple product-to-product similarity matching, rather than any machine learning-based Stuff,” Macaulay said.
“In the first decade of the 20th century, Netflix was also a major driving force for recommendation technology, and in 2006 established the Netflix Prize algorithm competition, which also sparked interest and research on recommendation technology in academia,” said Macaulay.
However, in the modern smartphone era, recommendation technology has been criticized for the so-called “information cocoon room” problem, that is, users will lock themselves in content that promotes their own prejudice and reject all information that is inconsistent with their own worldview, thus preventing people from understanding the truth. world.
Macaulay said: “Companies want to optimize user engagement metrics. They don’t want to inject diversified or balanced content because doing so will harm their key metrics.” He added that the company has no incentive to solve this problem. We live in an era of unprecedented demand for preferred information.”