TikTok has enjoyed tremendous success for several years now due to three main factors. First, it allows freedom and ease of creation that no other social platform has yet to match. Then, it is highly addicting thanks to its short contents, which are linked one after the other.
Finally, and this is undoubtedly the most important point, TikTok has an ultra-powerful algorithm that allows targeting with incredible precision the centres of interest of its users. But then, how does the latter work, and above all, how does it manage to be so precise? The Wall Street Journal sought to discover this through a survey published on July 21, 2021.
TikTok reached 3 billion downloads
In September 2020, TikTok revealed some details about how its algorithm works. The social network explained that interactions and viewing time were the two main factors that allowed the algorithm to know its users.
It also specified that other elements were taken into accounts, such as the captions of the publications, the hashtags, or the language, the country in which the user is located, and his type of device.
If this information allowed us to know a little more about TikTok’s algorithm, it was besides insufficient to drill it entirely up to date, in particular, because it was not specified to what extent each factor weighed in the balance compared to another. This is why the Wall Street Journal decided to conduct its investigation by testing the social network algorithm in actual conditions.
Viewing time would be the most important factor.
To do this, the journalists in charge of the investigation created dozens of automated accounts, each with its personality and areas of interest. All also had a separate IP address assigned to different geographic regions and their profile was completed with their name, age, but not gender. Note also that at no time were their areas of interest specified within the application.
Arriving on TikTok, each profile was confronted with a series of popular videos (on average, 6.31 million views) and various themes. As the navigation progressed, the content offered was less and less popular.
but more and more precise on the recurrence of a particular theme.
According to the WSJ’s findings, this was made possible by one factor in particular: viewing time. When a piece of content is viewed entirely, paused, or replayed several times, the application belonging to ByteDance considers that this is the signal that the user is particularly interested in. It will then take care of offering similar or similar content.
And for a good reason According to Guillaume Chaslot, the data scientist who worked on the YouTube algorithm and who campaigns for the transparency of algorithms, TikTok wants above all to target the content that makes you stay on the platform by capturing your attention and not those that have a positive impact on you, or that you particularly like.
You have to see TikTok content as a gigantic tree. The trunk would represent the most popular content on the platform, dealing with topics that can affect more or less everyone. This is where all new users start.
The algorithm understands which content mainly captures their attention and will direct them to videos with increasingly specific themes.
Imagine the case where a user is susceptible to animal content. When the algorithm detects him, it offers him a majority of videos with animals. Then, he understands that it is the dog videos that are particularly interesting in the user’s eyes, so he offers him content mainly presenting canines. Eventually, the algorithm ends up detecting that the user likes a particular Bulldog breed. So he will give him a majority of Bulldog videos.