THE TIKTOK FORMULA
Is there really a secret sauce behind TikTok’s algorithm?
I s there really some secret sauce in the tiktok’s recommender algorithm? I believe not. Here is why I say that. The New York Times received an internal document from an employee who was disturbed by the way how TikTok can induce self-harm. They received it during an internal knowledge-sharing event at TikTok. The main goal of the company is to increase and retain daily active users which happens to be 1B+ today. Activeness is counted in the way users interact with the product — like, share, playtime, etc. According to the document, the formula for recommendation goes as
Plike x Vlike + Pcomment x Vcomment + Eplaytime x Vplaytime + Pplay x Vplay
P is the number, E is the time and V is the value of that activity that a company gives to an activity. The formula looks pretty standard to me and feels like anyone would use it for their recommender algorithm for short video platforms. However, the secret lies in the value of V i.e. how much the company values that particular activity for a particular type of video for a particular user and user demography. The ML models can easily predict the value of other variables but V is something that remains unique based on the product’s understanding of the user.
There are other ways how the formula is refined, for example, boost the rank if the video is from the same producer but penalize if the video is disliked. Boost it if the user has watched that ‘type’ multiple times before but penalize it if the user watched the same video earlier in the day to keep delivering fresh content.
NY times actually went one step further to create 100 bot accounts to reverse engineer the algorithm and found that most heavy weightage was given on the ‘time’ of watching the video, if the video was paused, and if there were repeat viewings.
So, I believe, one of the reasons for its success is its short video format, due to which, it can show more clips per hour and enhance engagement due to continuous fresh content from the producer. In turn, the product is able to collect more and more data about the user in very less time. This makes it easier to consume and makes it habit-forming. Some real Product Mindset went in there.
Isn’t this good that you are shown what you like? How can it harm?
I said, it makes it habit-forming. Is that a good thing? Well, it can be harmful in many places. Other than just consuming a lot of time from teenagers and the young population, it can actually lead a user into a self-harm state. There are multiple instances in which it can be harmful.
- It can make users more prone to phishing attacks
- It can take away a lot of precious hours from a teen’s life which could have been gone on sports or other extracurriculars.
- It can unleash a ton of data to scammers who can very specifically target a person with ill intentions
- It can definitely bring stress and anxiety and helplessness when an active user is suddenly not able to maintain their charm due to a competition or online bullying.
What are ways to avoid it and extract benefits from it?
Along with their demerits, they have some good merits too when used in a controlled manner.
- Very effectively used for small and large businesses and we have seen this, especially in peak COVID times. Many businesses have adopted it as their long-term strategy
- It can be used to spread positivity in times of grief and economic crisis
- It can introduce us to a lot of similar-minded people through community building which can be later utilized by in-person connections.
- It can be used to get yourself updated with updated trends in the industry if your jobs need them.
- We can dislike the stuff that we don’t like so that only relevant content shows up
- It should be only allowed for folks over a certain age like 21+
- We can use its creativity tools to express our ideas and get free validation very quickly.
- And last but not the least, it is definitely good at what it is supposed to do — entertainment!
Eventually, I think it has its good and bad and if we want to continue using it, we can always choose to control it.
So, to summarize, the algorism behind it is very much common and it is the value of the parameters which is unique to every short video format company. If we use it with some individual restrictions for us and applied by the company, it can prove to be an ethical and positive impact-building product ever.