3-step strategy to build a complete content operation framework

Content has always been an important system in the Internet ecosystem. Since 2015, the mobile Internet has re-emphasized content: the copyright wars for videos and music, WeChat public accounts took the lead in protecting original works, and IP content was identified as one of Tencent’s two major strategies...

This series of actions shows that the era of "content is king" in mobile Internet has begun. For huge amounts of content, how to sort out the intricate relationships among them and make them an important competitive advantage of a product is one of the most core issues for product operators .

Content management is a traceable chain. Although the content of different platforms varies, they are inseparable from three parts: content production, inventory management, and content recommendation .

1. Content production: the source of fresh water

How to produce content is the beginning of the whole story. Perhaps in the early days, you hire an editor to search for content from the massive Internet and then move it to your own platform, thus completing the content construction of your own platform. However, this model will inevitably fail as business and laws improve. Only by building a content production system legally and compliantly from scratch can sustainable development be achieved.

The sources of content can be roughly divided into three categories: UGC , PGC , and external reprints.

1. UGC

UGC is the abbreviation of User Generated Content, which means content produced by users. UGC is the most important fruit of the Internet, especially web 2.0. It can be said that the history of the development of the Internet is the history of the development of UGC. The blogs, forums, communities, and self-media that we cannot do without are all UGC.

It is very difficult to motivate users to produce content for the platform, but if a UGC mechanism is successfully established, it will also allow the product to enter a positive cycle. Currently, typical UGC products are: Zhihu (community), Jianshu (blog), Lizhi FM (audio radio)... The core issue of UGC is how to build a user incentive mechanism to allow users to produce content. This is another topic that will not be discussed here.

2. PGC

PGC is the abbreviation of Professional Generated Content in English, which means content produced by professionals (institutions) in Chinese. This means that content producers have professional capabilities and knowledge in this field. PGC is also the most important source of content for traditional industries. For example, newspaper reporters, TV production teams and other professional practitioners produce content for readers/viewers.

In the Internet era, the proportion of PGC is not that high, but it cannot be ignored. For example, the content of portal websites is produced by professional teams. PGC content is suitable for platforms that have high requirements for professionalism and high thresholds, such as video programs, high-quality radio programs, and popular music platforms. Only professionals can produce content that users like.

3. External reprint

This seems to be the easiest way to produce content. In the wild era, many companies easily accumulated a large number of users in this way. At present, copyright protection and infringement crackdown are gradually strengthened, and arbitrary plagiarism and reproduction will no longer be so rampant. If you want to operate a regular product, you can no longer go the route of pirated reproduction.

But this does not mean that this channel is no longer feasible. It is still possible to reprint external content by contacting the author through formal channels and obtaining permission.

However, this is an unstable way of producing content, which can only solve the problem of lack of content in the early stage. In the long run, it is necessary to have its own stable source of content production.

2. Inventory management: infrastructure for managing massive amounts of content

Having too little content is a problem, but having too much content is also a problem. Massive amounts of content cannot be piled together randomly. If there is no good structure to store the massive amounts of content, all the content will be a pile of unusable "garbage". Just like IKEA’s inventory management, content also needs to be managed in the warehouse.

1. Structured information processing

The content of a product is generally similar, which means that structured information can be extracted for machine storage and operator management. Take music as an example. The most intuitive form of music is songs. After further decomposition, the structured information of a work includes: singer, record company, album name, release date... For music works, operators cannot just cram audio files into the hard drive and be done with it. Instead, they need to spend a lot of effort to extract the structured information.

Therefore, before the product is launched, it is necessary to consider the future content structure and then design a reasonable content management system. A good information structure is the basis for future content recommendations, because massive amounts of content are pulled by machines rather than manually selected. Machines are very rigid. If there is content, it is there; if there is not, it is not there. It all depends on what information the operator extracts when the content is entered into the database.

2. Classification system

Classification is a kind of logic, which has actually always accompanied the development of human beings. Zhang Xiaolong once said that making products is actually a process of logical sorting. The most mature classification system in traditional industries is book classification, and almost all libraries in the world use a unified classification system. The most mature ones in the Internet era are e-commerce platforms. If you have the patience to disassemble their structures, you will find that there are thousands of categories of all sizes.

Classification is a set of rules that makes it easier for users to filter content. When there is a lot of content, users can only choose the content they are interested in to consume, and classification provides an entry point for such a selection. It is best to use a unified classification rule for similar content. For example, e-commerce and music platforms have almost the same classification. On the one hand, it respects user habits and reduces user learning costs. On the other hand, it complies with industry regulations and reduces the workload of operators.

3. Labeling system

Classification systems are mostly stable and commonly agreed upon by all parties in the industry, and some people may react that this is not a flexible system. For example, in a classification system, it is best for a piece of content to belong to only one category, so that users can find it by following a specific route. Imagine that a book in a library can be placed on two different bookshelves. The administrators and readers will go crazy. Labels are more flexible. Wang Xiaobo's novels can only be placed on one bookshelf, but N labels can be attached to it: Wang Xiaobo, Chinese, novels, contemporary literature... When the computer searches, it can find the book as long as one of the labels is matched.

Currently popular personalized recommendation systems, such as Douban FM's "Guess What You Like" and Amazon 's "Recommendations for You", are almost all based on tag systems rather than classification. In order to improve personalized music recommendations, the foreign music platform Pandora has hired a large number of music professionals to label each song with N tags to help Pandora optimize the recommendation results.

3. Content recommendation: Whether users like it or not determines success or failure

The two strategies mentioned above are completed behind the scenes and users cannot feel them. However, content recommendation is a process that lifts the veil and faces users directly. From a result-oriented perspective, no matter how well other processes are done, if the content recommendations to users are not done well, everything else will be in vain. Next, we will break down the common content recommendation methods one by one.

1. Popular recommendations

The most common form is the "ranking" format. As a content platform, this is an indispensable module. When a user uses a product for the first time, the beginning for each user is almost the same, so there is no way to talk about personalized recommendations at this time. Then the best way is to show users the most popular things on the platform. This method will never go wrong. Users who come to the same platform are generally similar (unless you are a platform with over 100 million users). Showing other users' favorite content to new users also has the greatest probability of hitting users' preferences based on the law of large numbers.

Popular recommendations are a labor-saving and pleasing way of recommendation, but they are not a good way of recommendation for a large number of users and massive amounts of content. Its biggest drawback is the " Matthew effect ", which leads to almost unchanging content. As mentioned earlier, user behavior is converging, so the result is that popular things are always similar. Due to the influence of popularity and rankings, a lot of new content cannot be presented to users. Therefore, it is only suitable for the early stages. After the product matures, it should be reduced to a product module.

2. Editor's Recommendation

As the number of users and the amount of content increases, the form of "editor recommendation" will gradually become important. The logic behind this is based on the platform's confidence in the capabilities of its own operators, and the belief that the content recommended by official editors is what users like. Of course, the recommended content is not entirely based on the subjective judgment of the operator, but is based on data analysis and user feedback.

Editors' recommended content is generally fresh and high-quality, making up for the shortcomings of popular ranking recommendations. For example, on Zhihu, the most popular content is still about gender and emotional topics. If only popular content is displayed, it will be filled with emotional topics, which is a result that most users and operators do not want to see. Therefore, the operators of Zhihu Daily will choose more innovative content to display on the homepage.

An ideal positive cycle is that editors recommend high-quality content, which increases exposure and then becomes content on the popular rankings. In this cycle, the content on the platform is fluid, fresh, high-quality content is exposed, and popular content is constantly rotated.

3. Personalized recommendations

Finally, let’s talk about personalized recommendations, which is actually a technical issue. One of the core issues of big data , which is very popular at present, is to solve personalized recommendations. It is common in major platforms such as music, video, news, and e-commerce. The most representative ones are Amazon's personalized product recommendations and Netflix's film and television recommendations. It can be seen that the common feature of these platforms is that they have massive content and massive users. On such a huge platform, editorial recommendations and popular rankings cannot fully handle all the content and users. If we can recommend content tailored to each user, the efficiency of content recommendations will be greatly improved, which will naturally bring huge commercial returns.

There are three common ways to provide personalized recommendations:

1. User-based recommendations : find similar users to see what content they consume, and then recommend it to the user.
2. Item-based recommendations find similar content based on the current content consumed and recommend it to the user.
3. Recommendations based on item features (model-based) extract features based on the consumed content to find more similar content.

The first two are the most widely used. Neither is better or worse. Each has its own advantages and disadvantages in different scenarios. It is generally believed that item-based methods have better effects for new users. For in-depth learning, you can read Xiang Liang’s book "Recommendation System Practice".

4. Practical Case: Application of Content Operation Framework from QQ Music

QQ Music is one of the largest music platforms in China. Its DAU exceeded 100 million in mid-September 2015. It is a typical platform with massive content and a large number of users. Research on its content management has strong reference value.

1. Content production: Produce content in a PGC manner and adhere to the path of legalization

The music industry is the industry most affected by piracy and has been almost destroyed. Musicians basically cannot get the due returns from their music works and can only earn income from other places, so the entire industry is deteriorating. In this context, several major music platforms, as leaders, have gradually raised the banner of copyright, and QQ Music is one of the better ones.

The foundation of a music platform is content, that is, musical works. The first problem that QQ Music needs to solve is to establish a stable source of content production. Since the threshold for music production is very high and the content production method is almost all PGC, QQ Music has signed copyright agreements with several major record companies at home and abroad to pull the music industry towards legalization. In fact, this is also the necessary foundation for a positive cycle. Legalization allows musicians to get profit returns so that they can continue to produce high-quality content. In this way, musicians, platforms and users all benefit.

2. Content storage management: the largest music library management practice in China

Currently, QQ Music’s music library has reached 15 million songs, making it the largest music library in China. Faced with such a huge music library, QQ Music is equipped with a dedicated warehousing team and automated processing programs. As mentioned earlier, dozens of structured information such as a song's singer, album name, release date, style, category, etc. will be processed when it is entered into the library, using both manual and program methods. The perfect structured information is the basis for subsequent content recommendations.

3. Content recommendation: editorial recommendation, popular recommendation, and personalized recommendation complement each other

In the QQ Music APP, the recommendation module of the Music Library is editorial recommendation, with a high degree of manual intervention, and mainly recommends new songs and seasonal playlists. The rankings belong to the popular recommendation module. They are all results calculated by algorithms and reflect the platform users' favorite songs. Guess You Like is a personalized recommendation product. After several revisions, it can now be regarded as the leading personalized recommendation system in China and has a very good reputation.

V. Conclusion

Newcomers often ask, what does content operation do? "Old people" who have worked for a period of time will also be confused about what content operations should do next?

The content operation framework mentioned in the article can solve these questions very well. It is like a work map, which constantly allows operators to understand where they are and where they should go in the next stage.

(1) First of all, when you first accept the operation of a content platform, you must first solve the problem of content production: who is producing the content? Are they satisfied with the input and output? Is this production mechanism sustainable?

(2) Next, in the face of increasing amounts of content, we need to start building a complete content management system backend: Is the content inventory managed in an orderly manner? Can you quickly sort all the content into categories? Can it provide sufficient support for subsequent content recommendations?

(3) Finally, continuously adjust the recommendation strategy based on user feedback: At the beginning, it must be editorial recommendations and recommendations based on the popularity of the entire site, but as the number of users and content grows simultaneously, it is necessary to adopt decentralized personalized recommendations to activate the large amount of content in stock and recommend it to the most suitable target users .

The three-step strategy of "production-management-recommendation" of content is a process that all content platforms need to go through. Therefore, understanding and applying the content operation framework is the only way for content operators to upgrade from "editor" (writer) to "operator".