You have probably heard of it: TikTok. The widely popular app that came to existence after the acquisition of the video startup Musical.ly by the Chinese company ByteDance. Today, the app is used by almost 1.3 billion monthly active users, far more than Twitter and Snapchat combined. The success has been attributed to the platforms ‘For You’ page. A page on the app that provides an endless stream of videos that are uniquely tailored to the user, driven by a mysterious state-of-the-art algorithm, at the pinnacle of data fusion and machine learning. In this article, I will dive into the technology and mechanics of the feature that can make or break the career of a TikToker (i.e. a user of TikTok).
First, I’d like to state that the For You page is in fact a recommendation system, like the ones that we have seen for as long as the internet exists. Recommendation systems are made to gauge a user’s interest in certain content and based upon this provide the user with recommendations. Netflix for example provides you with a list of shows or movies that you can watch with the ‘Because you watched X’ category, where X is a show or movie you have previously watched. Google on the other hand uses a wide set of parameters on which it bases your search results. Unlike the first version of Google, in which the search algorithm was the relatively ‘simple’ PageRank algorithm, the current algorithm uses data from many data points to serve its search results to the user. What becomes clear of these examples is that recommendation systems, recommend content based on a user’s interaction with the software.
Now, for TikTok, there are many factors that contribute to your feed on the For You page. As we delve deeper into the workings of this intricate system, you will realize that a lot of your data is funneled into the algorithm, as to provide relevant recommendations. The For You page thus reflects preferences that are uniquely tailored for each user. The system takes a lot of engagement metrics into account. All these metrics are given a weight based on their value to a user and a recommendation is served in the form of a new video. These engagement metrics and signals can be, but are not limited to:
Interaction, video, and account information:
- A user’s interaction with other videos determines what type of videos are served on a user’s For You page. Interaction can be in any form that is possible within TikTok: giving a like, writing a comment, sending a gift, following another user, producing your own content, adding a video to your favorites, watching a video from begin to end, etc. The system also considers video information of the videos that interest you: hashtags, music, duration, and account information of the user that posted the video: number of followers, posts, etc. These three factors may well be given the most weight when serving recommendations.
Account and device settings:
- What is also important, albeit given a lower weight, is the vast number of account and device settings that the user naturally cedes to TikTok by using the app. These include language preferences, geographical location, sharing of contact information, device type, operating system, and many more. These are given a slightly lower weight, as the user does not actively engage with these to convey their preferences.
Initial For You page:
- When you initially start using TikTok (i.e. after you create an account), you are asked to make a selection from a wide variety of topics which might be of interest to you. This is used by the recommendation system to generate an initial feed of videos which might be of interest to you. Now, the option exists to not specify the topics that you are interested in and thus skip the entire onboarding step. In this case, the system will ‘only’ serve popular videos to the user, after which it will give new suggestions on a rolling basis as the user’s ‘profile’ is slowly ‘learned’ by the recommendation system.
Getting better:
- The system gets better as you start using the app more. Most of the insights are based on the interaction, video, and account information mentioned in the first item of this list. There are also systems in place to ensure that you do not end up in a ‘filter bubble’ in which you are solely served content that you like, hence resulting in an endless cycle of confirmation bias. The system will now and then also mention videos that are not entirely in your own ‘interest bubble’ as to show new insights to the user and foster diversity. Checks are in place to ensure that: no video of a user is served twice, spam is removed, and duplicate content is also not shown again.
As you can see, it is no wonder that this intricate system is able to serve hundreds of millions of users around the globe with so much precision. This remarkable feat of engineering called the For You page is at the benchmark of recommendation systems. One thing that is not mentioned before (at least in the body of this article), is the fact that the system also deploys machine learning algorithms to see patterns in user behavior that might otherwise have been invisible to the eye of an engineer.
To conclude, one of the main questions that you are probably wondering about is: how do I get featured on TikTok’s For You page? And the answer is not as straightforward as you might think. Although in this article I have discussed many aspects that contribute to the recommendations, a lot is still unknown about this behemoth’s For You page. So, to get featured, the best you can do is ensure that users do interact with the content you produce in the above-mentioned ways, grow a loyal following, and produce quality content, which draws interest from all groups of society. Maybe then, you might someday go viral on TikTok’s For You page.
Sources
https://newsroom.tiktok.com/en-us/how-tiktok-recommends-videos-for-you