The current recommendation method of Douyin is like a waiter who is very observant. The system This “waiter” no longer stares at the menu categories, but secretly observes your eating habits:
Which dish did you pick up with your second chopsticks? (Predict whether you will like it)
Which dish do you move in front of yourself? (Predict whether you will collect it)
Do you come back to ask for the menu after you leave the store? (Predicting whether you will watch it again and again)
It doesn’t care whether the dish is called “braised pork” or “popular knowledge”, it only cares about “how likely you are to pay for it (interaction)”. In the end, the dishes that you are “most likely to eat repeatedly (high-weight behavior)” will be served first.
Document: “Interpretation of the latest The system
This is why, once you start what are the differences between landbot’s chatbot builder and landbot’s ai agent watching TikTok, you can’t stop.
02 What user behaviors does the algorithm care about?
The algorith attempts to understand users by after the reduction of swift operations in russia? predicting the probability of their key behaviors. The main user behaviors are as follows:
Document: “Interpretation of the latest
When recommendation algorithms were first created
They only focused on a single or small number of goals, the ao lists waiter such as completion and likes. However, a single goal will lead to the “Matthew effect of traffic” – the top content is monopolized and the mid- and long-tail creators are lost. Multiple goals achieve ecological balance through dynamic weight control.
Combined goals, such as “Collect + Revisit”, “Follow + Follow”, and “Open + Search”, help the system find the long-term needs of users;
Cocoon and prevent users from leaving the platform before they get bored or tired of it;
Douyin understands human nature very well. It wants to resolve the contradiction between users’ immediate pleasure and long-term value, and meet users’ deeper needs by supporting in-depth and high-quality final videos.
