Editor's Guide: If operators or creators want to get more attention to their content, they can learn how the platform recommends content. In Zhihu, perhaps you can combine Wilson's formula to understand Zhihu's algorithm recommendation. In this article, the author has made an interpretation of Wilson's formula and zhihu operation, let's take a look at it together.
We all know that self-media people output content on the platform to attract user attention and complete traffic monetization, so many people think that the platform is a bridge between self-media people and users.
This view is actually correct, but more accurately, self-media people are never communicating with users, but playing games with the platform.
Why game?
Because any platform has its fixed mechanism algorithm, what self-media people have to do is to constantly let their content get closer to the platform algorithm and constantly trigger the recommendation mechanism of the platform.
I take Zhihu as an example, many people should have heard of Zhihu's "Wilson formula", but they do not particularly understand it, and they do not know how this formula is calculated.
Here, I will give you a detailed popularization of science, I hope to help some friends engaged in zhihu operation.
First of all, we need to understand that if we compare the operation of zhihu to a monster fighting game, the algorithm is a detailed strategy cheatbook, which can tell us what the key points of the monster are, and how to go next, this strategy allows us to know in the process of fighting monsters and upgrading.
Simply put, the figurative manifestation of zhihu algorithm is the personal content search ranking.
The larger the value obtained by the operator through the algorithm, the higher the search ranking, the higher the exposure, the more feedback we get, and the essence of our operation is to send the value that we get as large as possible.
Secondly, before discussing the specific formula, we need to think about the question, if you are the operator of the platform, which users would you tend to keep?
The answer is very simple, anyone who can have a positive effect on the construction of the platform, the platform naturally gives corresponding returns and rewards.
This is one of the basic requirements for a platform to want to grow in the long run, and it does not need to be proved by data.
For users, exporting professional knowledge, screening high-quality content, enhancing community activity, increasing the influence of knowledge, maintaining platform order, etc., all belong to the platform construction operation that can be landed.
Therefore, in the process of our continuous generation of these behaviors, we are constantly empowering knowledge, and we should get higher algorithm values.
After understanding this concept, let's analyze the official algorithm mechanism of Zhihu, the values mentioned in the algorithm, and its directing significance and pit closure guidelines.
Here is a formula, known as the Wilson formula:
where u is the weighted number of votes for approval, v is the number of weighted votes against, and za is the parameter.
The following diagram can be used to visually show several important features of Wilson's formula.
In order to facilitate the discussion, the up-vote, down-vote in the left figure, the corresponding axis of the score is the x, y, z axis, and the right figure is the contour chart of the left chart.
The overall surface shape of the left figure is consistent with the correspondence between the approval vote, the negative vote and the answer quality in the common understanding, which is an officially recognized algorithmic mechanism.
Weighted approval votes refer to the numerical value given to content by other people's likes, but be aware that the impact of each person's likes is different, depending on the weight of the like in the current field.
Many people carefully wrote an article, and then published it in the Zhihu answer, thinking that they could get thousands of likes, ten thousand likes, and become a Zhihu big v, but after a period of time, they found that the answer had few likes, not even a comment against it.
At this time, they began to be disappointed, thinking that they did not have the so-called self-media talent, and even did not fit into the operation of Zhihu.
This is actually a misunderstanding, your answer feedback is insufficient, does not mean that your content is problematic, and does not mean that this answer will remain unattended.
In fact, from the perspective of long-term operational experience, a valuable article, even if there is no feedback at the beginning, will suddenly explode in the future for some time, and the reason why it has not caused a certain degree of influence now is because there are still some problems with your account.
For example, if you write an answer, a big v of 100,000 fans and a trumpet of several hundred powders give you a thumbs up, and the impact gap is very large, which is the concept of "weighting".
Similarly, opposition is weighted, and the higher the weight of the opponent, the greater the decline in our score.
This formula looks more troublesome, everyone can't understand it at first, it is normal, I will simplify it, you can judge your weight through another formula.
i.e. s = like * favorite * like * initial weight * comment * oppose.
Among these influencing factors, in addition to the sub-domain weights, other factors are achieved through the interaction between users, that is, how high the ranking of an article is ultimately determined by other users.
In addition, likes, favorites, and likes increase the score of the answer, while disapproval decreases the corresponding score.
At this point, some people will say, you said this earlier, I understand, what is the effect of putting this Wilson formula?
It is also useful because in addition to expressing the correspondence between various factors, Wilson's formula can also concisely express the development process of factor changes.
Specific manifestations are:
Fixed no votes, the more approved votes the higher the score;
Fixed yes votes, the more negative votes the lower the score;
Fixed the ratio of likes to dislikes, and the higher the total number of votes, the higher the score.
This should be well understood, so I will not explain it more.
When the total number of votes is small, the answer will increase rapidly if the vote is obtained, and the larger the total number of votes, the slower the increase, what does this mean?
In the process of operating Zhihu, we will often find that some dozen answers that agree are ranked in a very high position, and the answers after him may have thousands and tens of thousands of approvals.
That's what this rule is all about.
Knowing that for the content just created, the system will give the content a weighted base score according to the weight of the creator's current field, so that it can get greater exposure.
This is reasonable because, relative to the very early answer, if the new answer does not have a certain initial exposure, it is not even qualified to be judged.
Under the premise that the system gives basic exposure, if it can be recognized by the first users, the system will determine that this is a potential content, and it will push it to more people, contributing to an outbreak in a short period of time.
During this time, although its number of approvals is not as good as other answers, the rate of praise is very high, which allows it to quickly improve its ranking and stabilize in the top position for a period of time.
Answers with higher numbers of approvals, when you start to get a negative vote, the score will drop rapidly, and the larger the total number of objections, the slower the decline, which is somewhat similar to the second principle.
How to understand this?
Just imagine, a person in our real society, as long as his heat is high enough, there will be a controversy, because everyone's ideas are different, there will always be different voices, but with these controversies, does not mean that he is a bad person, he is just hot enough.
Therefore, when the number of approvals of a piece of content is high enough that it can be seen by more people, there will be objections, and at this stage, the score of the content will drop rapidly until a stable proportional relationship is formed with the approval.
It should be noted, however, that there will be a large change in score only at the stage when we begin to receive objections, and the more objections we receive in the subsequent development process, the less the impact of each objection will be.
That goes back to the second formula, which is: s = like * favorite * like * initial weight * comment * oppose
From this formula and long-term operational experience, we can get the following six important principles:
All users see the same sort;
While other conditions remain the same, gaining approval will increase the order of responses, and obtaining objections will decrease;
The influence of content created by users in a certain field will increase the user's weight in this field, that is, the initial weight can affect the content score, and the final score of the content will in turn enhance the weight of our current field;
The vote of high-weighted users in the field has a more significant impact on the ranking, this vote includes likes and dislikes, of course, high-weighted users themselves when answering related questions, due to the blessing of the initial weight, their answers will be higher in the initial position at the beginning;
When voting or answering questions using anonymous identities, the user's weight is not calculated;
The fact that there is no good feedback at present does not mean that there is no possibility of becoming a high praise after that.
According to the Image of Wilson's Formula, it can be inferred that a good piece of content will eventually be recognized, but the time will be biased, and it may be affected by the initial weight or instability factors at the beginning, so that it has not been exposed more, but there will always be a point in time that will be discovered and recognized, which is also one of the regulatory roles of the zhihu distribution mechanism.
Therefore, content is very important, and the phrase "content is king" is by no means empty.
After a complete understanding of Zhihu's algorithm, what important information can we get from it?
Or how can we make the most of this official rule?
To sum up a little, there are two points:
After understanding the algorithmic mechanism of Zhihu, before creating a new answer, we must not answer the question as we wish, but first think about what our vertical field is, and then selectively look for the problem.
In the process of creation, we must learn to add some interesting and easy to be complained about on the basis of the original, that is, I often say that self-media people should constantly touch the user's "pain points" and "cool points", in order to actively interact with more users.
Of course, there are also some proactive behaviors, such as directly reminding readers to like at the end of the article, and even inducing readers to like, which are common ways to increase interactive behavior.
I have told some friends before that self-media people have three major realms.
The first is to express themselves, which is a common mistake that many primary self-media people will make, and use Zhihu as a platform for expressing themselves, and even an emotional tree hole.
This is the wrong way of operating, you are right to express yourself, but if you want to be exposed, think of traffic, and subsequent monetization, then you should not blindly express yourself.
The second is to serve users, many self-media people have experienced, when writing self-media articles, not for their own writing, but for the user to write, self-media people are more to act as a media role, the user wants to see the views they want to say to better express and list.
The third is to trigger the rules of the platform, why should self-media people write articles for users?
In essence, of course, it is not to make users satisfied, but to trigger the rules of the platform through the satisfaction of users, so that their answers can be more exposed and enter a larger traffic pool.
This is a good illustration that self-media has never been a struggle between creators and readers, but a mutual game between creators and platforms.
Thanks for reading~
Author: Jiang Han; WeChat public account: Jiang Han's number
This article was originally published by @Jiang Han in Everyone is a Product Manager. Reproduction without permission is prohibited
The title image is from Pexels, based on the CC0 protocol