I have been using Kuaishou for two months. Let me talk about my feelings from the two levels of product design and the dissemination algorithm of the work .

1. Product Design

When you enter Kuaishou, you will see three menus: "Follow", "Discover", and "Same City". These three menus are the three traffic pools that users’ works can enter .

The traffic pool corresponding to "Follow" is your own fans. This page has three elements: video, author's avatar, whether you are friends with the author, and the time of the video. The pages are basically arranged in descending chronological order. "Time" is the core logic of this page. After a work is published, it will enter the "Follow" section, but the page is sorted in descending order by time. The work is more likely to be seen by current online fans, but may not be seen by all fans.

The traffic pool corresponding to "Same City" is the city where you are located. The main element of the page is "geographic location", and the core logic is the distance of the geographical location. The published works can be seen by people who are nearby as well as by people who are far away. Whether a work can be posted to the "same city" depends on the location of the poster. The greater the user density in the city where the poster is located, the larger the traffic pool of the "same city". However, this does not mean that the work can enter the "same city", and entering the "same city" does not mean that it can share all the traffic in the city. This is related to the platform's algorithm.

The traffic pool corresponding to “Discovery” is the entire platform. The element exposed on the page is "number of likes", and "number of likes" is the core logic. The works that can be included in this menu are all relatively high-quality works.

From my own experience in publishing works, the above three traffic pools are of different sizes and have different levels of difficulty in entering. Most works remain at the "attention" stage, works with good performance will gain "same city" traffic, and works with excellent performance can enter the "discovery" stage.

2. Platform Propagation Algorithm

In August this year, I started to publish my works on Kuaishou, and sorted out my number of fans, number of views of my works, etc. according to the platform’s weekly report. He has released more than one hundred works in total, with a cumulative playback volume of over 100,000, and the current number of fans is around 1,400. During the National Day holiday, the number of views of the works has increased significantly.

The weekly report issued by the platform can be viewed in "Private Message":

The algorithm of the weekly report is summarized as follows: playback volume: the new playback volume of all works within the statistical period, including historical works; number of works: refers to the number of newly released works within the statistical period; increase in fans: refers to the number of new fans within the statistical period.

It should be noted here that the number of new playbacks given in the weekly report does not correspond one to one to the number of new works. Play volume refers to the new play volume of all works.

My first feeling during the whole process is that the first few works released in the early stage can obtain a lot of traffic, but this benefit is gradually no longer available in the later stage. Let's look at the data. During the two periods of 08/04 and 08/11, 22+22=44 works were released, and the average number of views were 242 and 370 respectively. However, during the two periods of 08/18 and 08/25, 17+8=25 works were released, and the average number of views were 188 and 332 respectively. The average number of views declined in the following half month.

It is speculated that in order to encourage new users, the platform will give relatively more traffic to the works posted by new users. Based on this, new Kuaishou users should seize the first wave of traffic benefits provided by the platform, publish high-quality works, and exchange them for more exposure and fans more efficiently. The second feeling: Every time a work is released on weekends/holidays, the total amount of playback and the rate of increase are better. In the table above, the statistical period of 10/06 happened to be the National Day, and during this period the number of views of my single work exceeded 10,000 for the first time. But the type of content in these works is different from previous ones, which does not explain the problem.

The weekly report issued by the platform will tell the top three friends in terms of playback volume. I have summarized the changes in the total playback volume of the TOP in different intervals. The results showed that during the 10/06 period, the total number of views for the top three programs did not increase but decreased.

The following is the presentation of TOP information in the weekly report and the summary of data for different intervals:

As you can see from the "Total" column, the total number of views for the week of 10/06 was the lowest, which contradicts my previous idea that there is more traffic on holidays.

I checked the homepages of the top three in each issue and found that the top three in each issue are basically the same. There are 15 people in the five issues that were counted, and the list has been dominated by five people for a long time. The ones with the same color in the picture are the same person.

The owners of the green and yellow colors that appear most frequently usually post works related to their work. Perhaps because of the National Day holiday, they posted fewer works, which in turn affected the total number of views. It is also possible that my idea is not valid.

In order to further observe the dissemination mechanism of the platform, I recorded the number of views of several works at different time points after they were released, as shown in the following figure:

In addition to the monitored works mentioned above, there are two works with relatively large playback volume. Because there are no monitoring process figures, they can only be described in words. One has a final playback volume of 10,000, and the other currently has a playback volume of 48,000, which is still growing.

The life cycle of a video with 10,000 views is 2 days, and the video with 48,000 views has been sent out for five days, with the current number of fans being around 1,400.

Judging from the works I have experienced:

1. More than 90% of the works are relatively mediocre, only a few works have more views than fans, and very few have more than 10,000 views;
2. The life cycle of most works is only about two days, and a few works can last up to five days (the longest one so far);
3. In terms of playback rate, the rate is fastest right after a work is released, then gradually slows down and finally stagnates. Sometimes, it will get a few more views every few days.

I selected several works with different playback volumes and recorded their likes and comments:

In the list above, the work with the largest number of views has the highest like rate and (likes + comments) rate, while the work with the second highest (likes + comments) rate is not the second in terms of views. Likes and comments will affect the number of views of a work, but the impact is not a simple one. In addition to monitoring the number of views, the “identity” of people who clicked on the works within different time frames was also monitored. Monitoring logic: Through the likes and comments received by the works, reversely check the homepages of these people (the platform does not provide the function of "Who has viewed my work").

It was discovered that in the early days of the work, the people who liked/commented on it were basically those who followed me. When the number of views of the work exceeded seven or eight times the number of fans, approaching ten thousand, I began to receive attention from people in the "same city". When the number of views of the work approached 20,000, I began to receive attention from people who "discovered" me .

Click on the number of fans on your homepage to view your fan list, which will display the source of the fans, as follows:

Starting from monitoring “people” as above, the intuitive feeling and experience is: most works stay in “attention”, works with good performance will gain “same city” traffic, and works with excellent performance can enter “discovery”. However, because this conclusion is inferred by reversely checking the behavioral data of "likes, comments, and follows", the conclusion itself is also affected by these data. The above are some of my feelings about publishing works. I saw a speech by Yan Zehua, Zhihu's Marketing Product Director, on the Internet. He talked about the flow model of content platforms, which can help us understand the platform's dissemination mechanism. Let me tell you my understanding. The main process of the model is as follows: content understanding → cold start → user feedback → spread or die → long tail

• Content understanding: The program will understand the works sent by users (by identifying the title description, video content, etc.), label the works and classify them.
• Cold start: After the program understands the work, it will actively push it to users who may like it. This involves another problem, which is that the program itself also labels and categorizes its users. After a new user releases his work, even if he has no fans, it will be promoted to a certain group of people. In my own experience, the first wave of people who view my works that I can directly monitor are my fans, but as a new user, even if I have no fans, I will still get traffic.
• User feedback: After the work is pushed to users, the platform will collect user feedback, including reading, likes, collections, sharing, comments, etc.
• Diffusion or extinction: The algorithm makes a choice based on the feedback it receives: to proceed to the next diffusion step, or to die.
• Long tail: In the dissemination of long tail, the timeliness and periodicity of the content will be taken into consideration. "Entertainment content is actually very easy to be forgotten by users and will be pushed once and for all; content such as food guides and travel guides will be pushed periodically."

The above model is a macro model, and there are the following ideas for some links: User feedback : After the work is released, the feedback indicators that the platform can collect include: open rate, number of likes, number of comments, number of reposts, completion rate, attention rate, etc. (referring to the ratio of people who follow the author by watching the work).

These indicators, together with the dimension of time, form a model for calculating "work performance". The final score of the model is used as a benchmark for the next wave of traffic distribution.

Among these five indicators, the “completion rate” is more interesting, and the “attention rate” is very strong. Kuaishou emphasizes "old irons". There is a group of people active on the platform. As long as their friends release a work, they will click on it or comment 666, but they do not have the patience to read the entire work.

"Following" is recognition of a person, while likes and comments are recognition of a work. The threshold for following behavior is higher, and it is believed that the "following rate" is very rigid. For the previous works with 48,000 views, likes: comments: follows = 3216:105:120. Spread or die : draw an evolutionary model. After the work has passed the test again and again, the user feedback it receives gradually approaches its true level, just like the praise rate on an e-commerce platform. The higher the sales volume, the more accurate the praise rate will be.

When the number of views of your own works is not good, you may sometimes have the idea that "it is gold that has not been discovered". The platform provides promotion services. I have purchased them twice. After the promotion, the final value of views of one work remained at 2200, and the final value of views of another work remained at 7107.

The conclusion is: the platform algorithm is very effective. If the work is freely disseminated but does not get good playback volume, then there is no point in purchasing promotion services. After all, the amount of delivery provided by the promotion service is very, very limited. If a product wants to spread well, it must be able to start on its own.

Author: A product dog

Source: A product dog

Related reading:

Kuaishou Operation: Inventory of techniques to increase followers!

How to create a Kuaishou account from scratch? Here’s a how-to guide!