Zhihu's content distribution strategy and operating mechanism!

As an excellent content question-and-answer community, Zhihu’s many practices in content distribution are worth learning. This article will explain from three aspects and is recommended to those who are interested in content distribution.

Zhihu has become the largest content question-and-answer community in the Chinese Internet world.

After ten years of hard work, Zhihu has accumulated a huge amount of high-value content, which is its broadest moat. But precisely because of its massive volume, how to efficiently distribute the content to those who need it becomes a key issue.

The traditional solution is search. The content is associated with the user's intent through clear user search terms. This is an era where people look for content.

Zhihu’s content distribution strategy goes beyond search, and also includes recommendations, ideas, and hot lists.

Why does Zhihu use them as a content distribution strategy? What are the operating mechanisms behind these strategies?

1. Algorithm-based distribution-recommendation

I believe everyone is familiar with the recommendation system. Nowadays, whether you are doing e-commerce, social networking, or content, you would feel embarrassed to go out if you don’t get personalized recommendations.

But to build a recommendation system, three steps are necessary: ​​understanding the content, understanding the users, and building rules.

The same is true for Zhihu.

Whether it is a search system or a recommendation system, the purpose is the same, which is to achieve efficient and accurate matching between people and content.

To achieve this goal, understanding content and users is a must.

So, how should we understand the content in the first place?

1. Understand the content

A more common way to understand content is by classification. For example, when uploading a video on Bilibili, the up-host needs to fill in the classification of the submitted content, which is the most basic classification of content.

The same is true on Zhihu, where channels, categories, and topics are all classifications of content. Content classification often presents structured characteristics. For example, programmer Xiao Li feels that the song list recommended by NetEase Cloud Music today is really bad while on the subway after get off work. He may open Zhihu and browse other people's good song recommendations under the NetEase Cloud Music subcategory of the music app category.

Therefore, from the music app category to the NetEase Cloud Music sub-category, the structural characteristics of content classification actually reflect an inclusion relationship. This kind of classification refinement is certainly not infinite, it will eventually point to a specific content.

So how do we continue to understand the specific content at the end of the classification? The most common way is labeling. For example, label a song with genre, duration, album, singer, etc.

There is no inclusion relationship between these tags, but an equality relationship. Bringing them together allows you to understand content from multiple dimensions.

2. Understand your users

Similarly, since we can label content, we can also label people. Labeling users also has a professional name, called user portrait.

When it comes to user portraits, it is easy to think of basic demographic labels such as age, gender, education level, and region. But what is more important for building a recommendation system is user behavior labels.

Every time a user follows, likes, shares, comments, or collects something on Zhihu, it is a behavior label.

Although these behavior labels are not inclusive, they have different weights. The weight of sharing is definitely greater than that of commenting, and the weight of commenting is definitely greater than that of likes. This was confirmed a few years ago when the source code of Bilibili was leaked.

With these multi-dimensional labels, we can build a user portrait system. Then we can do something, such as using big data to target old customers?

3. Recommended Rules

After understanding the content and the users, the next step is to find a way to accurately connect user intent and content.

Zhihu uses collaborative filtering that combines users and content.

There is a term here, called collaborative filtering. What is collaborative filtering?

Let me give you an example.

Let’s talk about user-based collaborative filtering first. In addition to listening to music, programmer Xiao Li also likes drinking. He feels that he is more inspired when writing code after drinking, so he likes to study various flavors of wine. Therefore, the recommendation system will find users who like to study various flavors of wine like Xiao Li, and then recommend the wine-related content that these users are interested in to Xiao Li.

Let’s talk about content-based collaborative filtering. Under the influence of programmer Xiao Li, designer Xiao Wang and product manager Xiao Wu also developed the habit of working after drinking. Once, they both browsed the two answers of how to drink healthily and why pineapple beer does not make you drunk, so the recommendation system thought that these two answers had a stronger similarity. After Xiao Li browses the answer to why pineapple beer doesn’t make you drunk, the recommendation system may recommend this answer to Xiao Li.

Therefore, when you open the Zhihu homepage and see the content recommended to you, it is calculated based on your past browsing, likes, comments and other behavior tags. Every action you take is a vote for the information you may receive.

2. Social-based distribution - ideas

1. How to distribute

Similar to algorithm-based content distribution, Zhihu’s idea page can also provide personalized recommendations based on users’ active attention. This is called social-based content distribution.

This is easy to understand. Everyone is a self-media and you can choose to follow any user you like. Then the content posted by these users is aggregated and displayed in front of you in the form of a feed stream. Compared with algorithmic distribution, users have greater autonomy in choosing when using social content distribution. It also places more emphasis on the content output capabilities of the users being followed.

The process of sorting out social distribution content can be roughly summarized as: content - follow users - more users.

What connects content and the users it follows is following behavior; what connects content and a wider range of users is the sharing and forwarding behavior of the users it follows.

Therefore, such a content distribution strategy is strongly dependent on the relationship chain. Weibo and WeChat public accounts are the most typical representatives.

2. What content is suitable for social distribution?

This requires that content must have two values: content value and sharing value.

Content value is the value of the content itself; while sharing value is the social value brought by sharing behavior.

So how do we create content value and sharing value? The answer is: the amount of information.

The essence of content is still information, and the smallest unit for measuring information is bit. In short, the smaller the probability of an event happening, the greater the amount of information it contains. Insightful industry observations, the highlights of the lives of top academics, and unexpected short videos; these contents are not routine events in daily life and therefore contain a large amount of information.

3. The dilemma of social content distribution

After all, there are only a few users who can stably produce large amounts of content. They also have a professional name, called KOL. Therefore, social distribution of content can easily lead to the Matthew effect. Most of the distribution traffic is controlled by a few top KOLs.

Therefore, Tik Tok will spend a lot of money to recruit talents, Weibo will support mid-level KOLs, and NetEase Cloud Music will also support independent musicians, all in order to make social distribution more controllable.

3. Hot List

1. Hot list content distribution strategy

Open the Zhihu homepage, and there is another content distribution channel - the hot list.

The launch of the hot list is to have in-depth discussions on the hot topics on the entire network. Depth reflects professionalism, which has always been the community atmosphere of Zhihu, while hot topics reflect timeliness.

According to the official statement on the popularity of the hot list: the content popularity value in the Zhihu hot list is calculated comprehensively based on dimensions such as the number of views, interaction volume, professional weighting, creation time and time on the list of the content in the past 24 hours.

2. Timeliness

The biggest difference between hot content distribution and algorithm distribution and social distribution is timeliness. There is no need to make personalized recommendations for time-sensitive hot topics, so the Zhihu hot list is also the same for everyone.

What is the timeliness at that time? Users want satisfaction immediately, right away, and now. Therefore, the popular answers must be those that are closely related to public sentiment.

This is somewhat contrary to Zhihu's early labels of idealism, depth, and substance. Because it takes time for content to accumulate and it also takes time for valuable issues to be developed. No matter how popular the content is, its heat will eventually fade.

IV. Conclusion

Recommendations, ideas and hot lists have bid farewell to the era when people could only rely on search to find content, and now it is the content that finds people. In any case, they all want to solve the need for efficient connection between people and content.

This demand did not arise out of thin air. The accumulation of Internet content has gone through a stage from scarcity to abundance and then to high quality, and people's requirements for content quality are getting higher and higher.

From this perspective, knowledge payment, recommendation systems, and content distribution are actually all intended to help people better understand the world.

Author: Lcarusd

Source: Lcarusd

Related reading:

Creative analysis of advertising on Zhihu platform!

Practical operation! 4 tips for attracting traffic from Zhihu!

Zhihu account anti-ban rules and product promotion guidelines!

Zhihu’s traffic pool!