The TikTok recommendation algorithm works through a kind of information flow funnel algorithm. When a new ** is uploaded to Douyin, the **** will first go through a double review, if the review is passed, the system will assign an initial traffic pool to the **, the size of this traffic pool is usually around 200-300** users, and it may also reach thousands**.
In this initial traffic pool, Douyin will analyze the quality and popularity of the ** based on a series of factors, such as the completion rate, the number of likes, the number of comments, the number of **, and the depth of the tour. The importance of these factors is fine-tuned in real-time based on the overall algorithm.
If the ** performs well in the initial traffic pool, i.e., the completion rate is high, and there are more likes, comments, and **, then the system will allocate a larger traffic pool to the ** in the following recommendations. This process will be repeated, and each time the recommended traffic pool will be larger than the previous one, until the ** reaches a certain level of popularity or the feedback is not good.
In the process of expanding recommendations, Douyin will also distribute traffic based on crowd tags. It's like guessing what you like, ** will be tagged and the user will be tagged at the same time. TikTok will recommend ** to the users who are most likely to like it by matching these hashtags.
When a ** reaches a certain popularity, it is possible to be recommended to the boutique recommendation pool. This usually means that the ** has received a lot of attention and love. Once you enter the boutique recommendation pool, the amount of the recommendation will increase significantly, and it may reach millions or tens of millions.
At the same time, Douyin's recommendation algorithm will also consider other factors, such as users' behavior Xi, interests, activities, etc., to recommend ** that meets the user's preferences as accurately as possible. That's why some** may be similar to the user's preferences in terms of theme, style, or content, but still get recommendations.