A "buyer" who doesn't want to be a "chef" is not "good to operate". For a content product, in the daily work of operation, especially the "content operation" classmates will often have a lot of work cooperation with the recommendation algorithm classmates. .
The operating classmate is like a buyer of a restaurant, responsible for the purchase of ingredients, while the recommended classmate is like a chef, combining the menu (preference) ordered by the user with the corresponding ingredients to make dishes that the user is likely to like.
In this chain, the operating classmates are upstream. If the imported content/creators are not high-quality, just like the purchased ingredients are not fresh and high-quality, it is difficult for recommended classmates to make delicious dishes no matter how hard they try.
At the same time, there is another problem. Even if the operating classmates buy the best ingredients, if there is a problem with the way they recommend the classmates to cook, the latter does not use the ingredients in the most reasonable way, nor does it maximize the value of the ingredients, and it will be a waste.
Therefore, for the operating students, it is necessary not only to do their own upstream work, but also to know the recommended related work, so that when the dishes made are not delicious enough, we can timely find out whether it is the problem of the ingredients or the method of doing it. question? Make the next adjustment faster.
1. Must know 1: How is the content recommended?
For operation students, the first thing to understand is: how is the content recommended? How do the creators we introduce and their content go through a layer-by-layer process to decide whether to be recommended and how much traffic they are given?
After the content enters the system, the overall processing flow is different for different products and those that do not pass the company, but the overall logic is basically the same. The business logic of the large module is basically as follows.
Operation must know "recommended" two or three things
As shown in the figure above, when a user uploads a piece of content, the content will first go through the security review process. The security review mainly removes some illegal, yellow, violent and bloody content, and the unreviewed videos are basically permanently blocked or directly deleted.
After passing the security review, most of the content communities will have original review to filter out some content that has been uploaded or moved repeatedly. , or private domain display such as fans' follow pages.
After passing the original review, the video will enter the first quality review. The quality review is mainly to filter out some meaningless, unsubjective and messy content. After passing the first quality review, the content will be included in the recommendation system. The candidate pool is recommended, and then it will recommend the most basic traffic of the work. The purpose is to preliminarily judge the quality of the work through the data generated after the basic traffic.
If the data feedback after the basic traffic is good, it will continue to add more traffic recommendations. After getting more traffic recommendations, if the data performance is not good, the recommendation will be stopped; if the data performance is good, it will be recommended again. Enter the second content quality review or report review.
The main purpose of the second quality audit is to prevent the previous audit from being missed, or some content that does not conform to the tonality of the community content. Report review refers to reports that users who consume content actively click on. Content that has received too many reports must have potential risks and need to be reviewed manually.
After passing the second quality review or reporting review, the work will continue to receive more traffic, enter a cycle of recommendation, and become a candidate for key recommendation by the content platform.
However, during the entire continuous recommendation process, there will be some more detailed review processes, such as high-profile review, reviewing the most popular videos on the entire platform to ensure that there is no risk, and at the same time, users will continue to report and review to detect potential violations in time.
During the continuous recommendation process, if the data feedback of the content declines, the recommendation cooling will be slowly country email list carried out until the recommendation stops.
In all the above processes, the works that have been stopped recommended will also be reactivated due to some accidental triggers or other recalls in the subsequent process, giving more traffic to be recommended. For example, when it comes to festivals, the content of past festivals will be recalled and recommended.
After understanding the above recommendation process, operation students can have a clear understanding of the flow of the overall content, which can be combined with their own products or business logic to refine the overall process, so that when encountering problems, they can Know in time what stage the.