Case analysis: 4 key points of content platform operation!

The theme of this article is: 4 key points of content platform operation from the perspective of “Daddy Framework”. Let me share with you some content platform operation methodologies that I have summarized based on my personal work experience.

1. What is the “Daddy Framework” for content platform operations?

1.1 Background of content platform

First, let me introduce the background of the content platform. In the current market environment, content platforms are the king of duration. In the era of the red ocean of traffic, user attention is actually equal to duration. We can use duration to fit user attention. Duration is the indicator we use to explain user attention, that is, duration is approximately equal to income.

There are two core data:

• The stock is large, and nearly 34% of the total user time on the entire network is concentrated on content platform products. Online videos, short videos, comprehensive videos, browsers, Weibo social networking, etc. basically belong to the category of content platforms, and the rest actually also include content platform modules.
• The growth is rapid, with more than 65% of the total user time growth on the entire network concentrated on content platform products. These include short videos, comprehensive information, online videos, Weibo social networking, browsers, and online film, all of which belong to content platforms.

In general, content platforms are very important and they occupy most of our time on the Internet now.

The picture above shows the products with the largest number of users on the entire network, and basically all of them contain content platform modules. Some are pure content products, such as iQiyi, Douyin Short Video, Tencent Video, etc. The products themselves are content.

Others, such as Moments and official accounts in WeChat, highlights and interest groups in QQ, Weitao and Haohuo Live in Taobao mobile apps, etc., actually belong to the content module. Alipay also has a community, and Baidu is actually a content product that distributes content through search. Although it is essentially a tool, it is actually also a content product.

Ubiquitous content platforms account for a large amount of traffic on our networks.

1.2 The composition of the “Daddy Framework”

Let’s look at this picture. There are three main participants in the entire content platform: ordinary users, creators, and content platforms. The content platform itself has four major elements: content, distribution, growth and commercialization. These four elements keep the content platform running normally.

Then let’s look at content flow, capital flow and user flow. Content flow refers to the content created by creators. After entering the system, we will review and identify the content, and then distribute it to our users through recommendation algorithms, social networks, etc. to form a content flow.

After watching the video, users will click on ads, watch live broadcasts, give rewards, etc., which returns the funds to our platform. We then use the funds to buy users, thus forming a cash flow. We can use funds to acquire more ordinary users and creators, which in turn forms user flow. This cycle keeps our content platform running.

This picture can be further sublimated into what we are going to talk about today - the "Daddy Framework" for content platform operations. This is a framework for a large content platform product. For content-based products, the product will build a large framework, operations will add flesh and blood, and technology will provide energy.

Operations play a very important role in the entire work life cycle. In an environment where content platforms are unprecedentedly prosperous, as an operator of a content platform, it becomes very important to know what the focus of work is and how to promote the development of the entire platform and keep the entire gear running.

There are four key points in content platform operation:

• First point: Define. When we do operations, we must first understand the content ourselves, help the machine understand the content, and conduct qualitative and quantitative analysis of the content.
• Second point: Acquire. We need to introduce content, that is, acquire content and ensure the sustainability of the content.
• The third point: Auxiliary distribution – Deliver. Machine distribution has now become the mainstream. As operators, we play a more auxiliary role, helping machine algorithms distribute content and improve distribution efficiency.
• Fourth point: Activate the user - Activate. The last step is to attract and retain users and creators, allowing more users to use, become active, and retain the app, thus expanding the scale of users and creators.

To sum up, the first letters of these four words are combined into DADA, which we call the "Daddy Framework".

1. Definition of "Daddy Framework"

2.1 Characterize the content

The first step of the "Daddy Framework" is to define the content. First, let me share with you how to characterize the content.

Please take a look at this picture. At first glance, some people may think that this is related to sports, entertainment, or table tennis (Zhang Jike), while others may think it is related to variety shows (Xie Na). From this we can see that even humans can have different understandings of the same content, let alone machines.

So how do we resolve differences? We need to define a standard for the content and reach a common agreement internally. How can we define the standards of content so that everyone can understand it more easily?

There are 6 points here: clarify goals, qualitative description, quantitative description, focus on the big and let go of the small, highlight cases, and clarify boundaries.

The key is to use qualitative and quantitative descriptions, that is, what kind of content it is and what range of content it is. Furthermore, highlight the cases and use the cases directly to explain what the content is like and what the content looks like.

Let me give you an example: the characterization of disgusting content.

First, determine the standard target. Some content contains content that makes consumers feel disgusted or uncomfortable. Displaying it directly to users will seriously affect users' viewing experience. Some of it will cause strong disgust among most people and should be removed from the shelves directly.

The other part is that some users are insensitive or disgusted by the content, so the exposure range can be limited. The goal of this standard is to identify such content and reduce the chances of users seeing offensive content.

Then list the standard terms (such as standard terms for content that most users would find extremely uncomfortable, has a negative impact on minors, and should be removed from the shelves):

1. Corpses: Close-ups of disgusting human or animal corpses, incomplete human or animal corpses, and large bloody corpses;
2. Wounds: images of purulent infection of human or animal wounds, extremely bloody images such as large-area trauma, dissection images, and images of bloody organs;
3. Women have menstrual bleeding, bloody sanitary napkins, etc.

We will set standards for the content in great detail, and even add pictures when necessary. In this way, we can reach a consensus on internal content standards, what our content is like, and we can give the same answer for each piece of content.

Now that the content standards have been set, how can we better implement them?

We set up a process like this:

First of all, students who set the standards should provide standard training to new students. This face-to-face training can greatly improve communication. During the training process, trainers can also improve the standard based on feedback from trainees.

After the training is completed, we will let these new students take an exam to test whether their accuracy rate meets the standards. For example, if more than 90% is accurate, then this is a student who has mastered the standard. If it does not meet the standard, retraining is required.

This forms a closed loop, and new students must be trained to meet the standards before they can leave, so that they can execute the content to the standard.

2.2 Identifying content

After characterizing the content, how do you identify it? People often ask us, how does the recommendation system work? Everyone may say that it is to provide users with content that interests them by matching content with user tags. But how to type this label? Many people may not be able to answer this question, so here are the four steps of labeling:

• The first step is ideal classification. Everyone reached a consensus on what the ideal classification should be like.
• The second step is content standards. Based on the ideal classification, we set the standards for the content, and we use the standards to carve out the ideal classification.
• The third step is sample labeling. We will have some labeling students who will label our samples according to classification. Usually, tens of thousands of sample data are required to build an NLP (natural language processing) model with relatively high accuracy.
• The fourth step is model training. We will train the model and finally get a practical effect.

With such a funnel, there will be more loss in the middle. The labeling process is a painful and difficult process. The reason why big companies’ recommendations are so accurate is because they put a lot of effort into areas that you can’t see.

There are three stages of content identification: manual labeling, rule-based strategy, and model plus strategy. You can choose according to the actual situation of your company.

If the amount of content is not very large, manual labeling is sufficient, and the accuracy rate is very high.

If the content reaches thousands or tens of thousands, there will be very high requirements for accuracy and recall rate. Given that the labor cost is too high and the accuracy may decrease, we need to use rules and strategies to do some labeling, that is, model + strategy.

Let me give you an example of content portrait. The label we finally typed out will probably look like this.

For example, if I wrote an article, the creator information of this article would be me. The content topic information is determined based on the probability of the article belonging to a certain topic, which may be news, comedy, etc. If the article is written in a messy way, the markings will look strange. The following scores are based on the NLP model.

Content quality information is critical during the review phase. Whether in the United States or China, content security is a very important matter. For example, Neihan Duanzi and Xiaohongshu were taken off the shelves, which are all issues of content security. For pornographic, disgusting and other content, we will adopt corresponding strategies according to different degrees and delete the content, ban the account and other processing.

In the past, we could only identify the theme of an article, but now we can gradually identify all kinds of information in text, images, and videos. We will also label users based on their behavior to determine whether they are interested in something, so that we can better understand them and provide them with a better experience.

2.3 Evaluating content quality

Next, imagine we go to a store to buy something. Stores will pick out some higher quality and better-selling products, place them in more conspicuous locations, and sell them at a higher price. The same goes for rating content.

We need to launch high-quality content so that users will feel that the tone of our platform is very high. If we put out something vulgar and clickbaity, it will lower the tone of our platform, and neither users nor advertisers will buy into it.

Therefore, we need to quantify the quality of content, guide the adjustment of our distribution strategy, regulate the tone of our content by labeling the content with grades, and ensure the stable operation of our content platform.

In addition, the level of the creators must be quantified. We work with so many creators, and we need to determine which ones should be given more or less subsidies and how to reasonably evaluate this. We also need to evaluate based on the level of the content.

Let me give you an example of content levels. As shown in the picture above, we divide the content into 6 levels. Each level will have a corresponding level meaning. Based on different themes and carriers, we will also clearly list what the content at this level should be like, and there is also a very clear standard.

For the lowest level red line content, such as blood, violence, and other level zero content, we will directly take the content offline and even ban the creators. Like some low-quality content that has no central idea and is written in a messy way. We will set it to level one to limit the exposure of this content.

Now that we have content ratings, how do we rate creators?

There are five main aspects for evaluating creator level: content level, influence, personal connections, fulfillment history, and content verticality. Among them, the content level is the most important, the influence is the number of fans of the creator, and the personal connection can be measured by the big Vs who have a good relationship with the creator.

For example, there is a creator who has a great influence and has good relationships with many big Vs. He can bring us more big Vs. Moreover, the quality of his own content is also very vertical, and he has never published low-level content in history. Then the level of this creator is relatively high.

In addition, regarding the relationship between creator level and content level.

We will rate existing creators' content. For example, if a creator has 100 pieces of content and we randomly select 20 of them and rate them as level five, then all new content created by that creator will be rated level five. This means that we believe that this creator can continue to produce high-quality content, but we also need to continue to verify it later.

1. "Daddy Framework" content

The second step of the “Daddy Framework” is to obtain content. Starting from the vertical field and then generalizing is basically the path for all products to grow together. In the early stages of a product, if there is no particularly vertical angle to enter, and if there is no something like a scorpion to enter the market, it is basically difficult to grow.

3.1 Clarify the platform’s vision

So this requires us to think clearly about what the platform’s vision is. The direction in which we introduce content should be the intersection of user needs and platform vision.

For example, Kuaishou has a large number of young users under the age of 24, a large number of female users, a large number of users from third-tier cities and below, and users with relatively low educational levels. They prefer interesting and down-to-earth content.

Kuaishou’s platform vision is to “record everyone equally”, so the content introduced and presented by Kuaishou is very down-to-earth, and it feels like what is happening around you (things that happen to everyone in the village).

Tik Tok is different. Tik Tok creates a performance and a show. The content must be beautiful, very good, and something that everyone will envy when they see it. It means that I am so beautiful, life is so wonderful, and everyone yearns to be like me. The platform's vision is closely related to the content we want to introduce.

Let’s take a look at Xiaohongshu. It was originally defined as a guide to overseas shopping, and the team itself produces some knowledge content related to overseas shopping. By 2014, it had become an overseas shopping community, and its slogan was changed to "The best things from all over the world". Users could make purchases directly from overseas through posts, forming a closed business loop.

By 2016, it had become a lifestyle community, the platform's vision began to generalize, and its slogan became "Good Life Around the World." The themes of the content have changed from vertical beauty, overseas shopping sharing, etc. to general life-related fitness, pets, food, etc., which has greatly enriched the universality of Xiaohongshu. The user scale will also grow greatly at this stage.

We will find that the direction of introducing content is not static, but changes with time and product stage. Xiaohongshu has ushered in a new era of Internet content e-commerce in China and is a company worth learning from.

3.2 Criteria for introducing content

After clarifying the platform's vision, we will start introducing content. We have two standards for introducing content:

For content that already exists on the platform. We will judge whether such content is good or not, and whether it is sufficient or not.

For example, for fitness themes, we use three indicators, namely user penetration rate, click-through rate, and per capita consumption, to measure the quality of such theme platforms. If the content is good, then we will consider whether it is enough.

Let’s take fitness-themed content as an example. If we gradually increase this type of content, but the click-through rate keeps decreasing, it means that this type of content is enough and does not need to be introduced again. But the click-through rate is still increasing, which means that this type of content is not enough and can be added.

For content that is not available on the platform. We will judge whether there is such content and whether we want it.

For example, we will refer to competitors that have overlapping attributes with ours. There are some game categories whose content is very popular there, but we don’t have such content. Should we introduce it and give it a try? It’s a similar concept.

This chart shows some key indicators of the content we introduced. We will use some indicators to measure how well the content introduction work is done, including the daily incoming document volume, the total incoming document volume, the incoming document volume of different themes and different carriers, etc.

1. Auxiliary distribution of "Daddy Framework"

We have defined the content and acquired the content. The next step is to distribute the content to users.

4.1 Identify opportunities

Nowadays, machine distribution is a mainstream distribution method, but the machine itself has no creative thinking and cannot truly understand the content. The best you can do is break down the content, for example, what words are in the content, what the general theme is, and use these small points to connect people and users. That is to say, machines can only function within the framework of creative limitations.

But people can provide the upper and lower limits of the framework, and this is the role of operations colleagues. Operations colleagues can truly understand content, create content, and even create hot topics. It can discover opportunities in machine distribution to increase the upper limit of machine distribution, and it can also discover problems in machine distribution to increase the lower limit of machine distribution.

If operations students want to discover opportunities, there are four directions:

1 Hotspot Tracking

Machines cannot predict content that will become popular soon, they can only track content that has already become popular and present it to users. The machine cannot tell what content is about to become popular, so operations staff need to discover it.

Machines cannot create hot spots by themselves, only humans can. Therefore, our operations colleagues need to be able to operate hot spots and create a hot spot themselves. This is a very important value of operations.

2. Embodying values

Although it sounds a bit empty, in fact our content platform itself is a media, and the media should assume the responsibility of spreading correct values. Because machines cannot understand content, only humans can understand content and distribute content that is truly valuable to everyone.

3. Set the pace

Taking advantage of users' herd mentality to guide community users to participate in the community and create an atmosphere with platform characteristics is commonly known as "setting the pace." The operation may need to disguise that the user has made a lot of comments under a certain post or a certain point and has done something.

4 Traffic support

We need to increase the value of traffic and use it to cultivate new users or monetize traffic. These are two very important directions that can cultivate new forces on the platform and obtain more value.

These are the four directions I think to discover opportunities, which can increase the upper limit of the entire machine distribution.

When it comes to the means of human intervention, it is roughly like this, from mild to coercive. Operators can select some high-quality content and put it into a pool, which will increase the exposure weight and increase the chances of the content being seen by users. The weighting is relatively mild and the effect will not be particularly obvious, but creators who enter these pools can feel the increase in exposure.

In addition, we can also set the pace through topics or activities, and even have some fixed operating positions, such as Tencent News, or fixed positions on the opening screen, to emphasize truly hot content and core content. The most powerful one is push, which is used to wake up App users. This method should be used with caution, because push notifications often result in accidents.

In terms of content distribution, we can use data analysis tools to monitor traffic changes at each entrance at any time, such as traffic at each different entrance, active users, user stay time, etc. We can use data to measure the performance of different content.

For example, when measuring content themes, we will label each different content theme. For example, for sports content, we tag it with sports tags. After distributing and transmitting the data, we can know in real time which content has a high click-through rate and which content has been most exposed. It is very convenient.

4.2 Find the problem

After finding opportunities, the next step is to find problems. There are two main aspects:

First, monitor anomalies

There are two main ways to monitor anomalies: one is to monitor from the source, and the other is to monitor from the result. The former focuses on possible causes of distribution problems, such as changes in incoming documents, algorithms, and new indicators. The latter focuses on the results of distribution problems, such as changes in indicators such as retention, reading volume, duration and user feedback.

Second, get to the bottom of it

The questions behind the indicators test your data analysis and logical reasoning skills. When we discovered changes in indicators, we were able to find the corresponding content through sampling in different ways, evaluate the changes in content based on the standards, and then discover problems. Another way to track down problems is to let users express their dissatisfaction through user surveys.

1. Activating users in the Daddy Framework

The last step of the "Daddy Framework" is to activate users. In fact, it is to acquire new users, recall lost users, stimulate active users and awaken dormant users. We will make different segmentation strategies according to different goals and use various methods to activate users.

Users of content platforms are divided into two categories: ordinary users and creators.

5.1 Growth of Ordinary Users

To acquire ordinary users, we mainly have two ways: online and offline.

1 Online

Online mainly includes advertising and new customer acquisition activities. For operations, advertising is actually buying traffic, which is currently the most important and important way to acquire customers. This requires operations staff to find advertising materials that can easily bring traffic to the corresponding products.

In addition, there are some specific activities to attract new customers. Incentive type, such as cash subsidies from Qutoutiao; content type, such as Douyin and Kuaishou attracting users through their own content; game type, such as Alipay's Ant Forest, etc.

2 Offline

If our online traffic reaches a bottleneck, we need offline traffic. The most commonly used source of offline traffic is television advertising. For example, during the Spring Festival, you often see the Spring Festival Galas of Baidu, Tencent, Alibaba, etc.

Here, materials, users and content are related to our overall user growth. Let me share two cases with you:

The first case: a content product we made for Southeast Asia at the time. The country was about to hold a general election, and people's enthusiasm for political discussions began to increase. We will look for relevant political content to increase exposure and improve weight.

We attracted a large number of users through political content, and these users further distributed this content as material, forming a cycle. Our DAU at that time directly doubled.

The second case: India’s WhatsApp. Indians are very interesting. They like to say good morning to everyone every morning when they wake up. They send a picture of either a god or a flower, but pictures like these are not easy to find.

Whats App is very smart. It has collected a lot of such pictures and established many sharing groups to share similar pictures. Users saw that there was a lot of content like what they needed on this platform, so they downloaded the product. WhatsApp gained millions of DAUs through this move, and the first step of growth was taken.

In addition, I would like to share with you some information about the growth of ordinary users.

In my experience, about 2%-10% of users become active through push notifications. 10% is a considerable number, which is equivalent to doing nothing but bringing in 10% more users through push notifications. Just think about how considerable this is.

There are roughly three directions for push: the first is hot content, the second is product features, and the third is interest. Pushing based on user interests will have a higher click-through rate.

Pushing seems to be a simple thing, and while it brings us great benefits, it also carries relatively high risks. Even some large companies have their share of mistakes. For example, a video mistakenly said that Typhoon Lekima has caused deaths in the entire province, which caused a very serious impact. When it comes to push, we need to pay special attention to the security of the content.

5.2 Creator Growth

In addition to the growth of ordinary users, there is also the growth of creators. The things that are most attractive to creators are nothing more than traffic, subsidies, and some growth-related things. So how do we help creators achieve growth?

We have levels for creators, and each level has a corresponding gift package. Senior creators should receive correspondingly higher subsidies. If the content created by this creator is in a direction that we are particularly eager to follow, we will provide a little more subsidy.

For example, now we are in urgent need of level 5 sports creators. We will list out how much traffic and forced exposure traffic they should give, how much base salary they should give each month, and how much money they should give after reaching a certain number of readings. We will make different gift packages according to the different levels of creators.

Creators mainly come from two aspects: one is the training of local creators, and the other is poaching. At the beginning, everyone thought that poaching was more important and could bring in some different content. In fact, we should also give our own creators space and train them more. These creators who are very loyal to us can cooperate with us better.

In addition, how to cultivate creators? We can create some creator manuals, hold some lectures, establish a creator academy, etc. Larger companies can directly open an incubator and provide creators with venues and other support, which is equivalent to strongly binding these creators.

1. Summarize

Today I shared with you the 4 key points of content platform operations from the perspective of the “Daddy Framework”. In the new era, content operations have retreated from the open editing to behind the scenes, playing more of a strategic role.

Finally, let me summarize for you that the operation of content platforms has four obvious advantages:

1. Truly understanding the content will enable you to better control the content.
2. Truly understanding users will enable us to introduce more content and better assist our distribution.
3. Be creative, be able to do more work to increase user growth, make the platform's content more distinctive, and keep up with the times.
4. Having insight can potentially bring us more commercial opportunities, help us discover problems in each of our links, and improve the effectiveness of our entire process.

I believe that with the support of all the excellent operations colleagues, the content platform will shine and grow better. This is the end of my sharing today. Thank you everyone!

Author: Zhang Junjie, Product Director of Taqu

Source: Speech at GrowingIO 2019 Growth Conference (Beijing)