@ Wang Zhijie
Big data architect of Luo Ming Science and Technology;
Graduated from Peking University, majoring in computer science and technology.
Looking back:
Know you better than you, talk about user portrait (1)
02? Why do you want to do user portraits?
The content mentioned above is the theoretical definition and intuitive understanding of portrait.
After we know the portrait, why build it? What's the use of building it? Let me share my opinion with you.
First of all, we analyze it theoretically and divide it into two aspects: business and technology. At the business level, we can establish a specific cognition and a strategic and tactical direction through user portraits. At the same time, it can also tap the user's footprint and form a user-oriented direction.
What the hell does that mean? It is to construct specific cognitive and strategic and tactical directions. In fact, that is to say, in order to reach a unified and specific understanding with the core users at this level. After reaching an understanding, it is convenient for us to be targeted in the follow-up investment, at least knowing which direction to invest.
When designing products for users, we must clearly know what users look like, what behavioral characteristics and what attributes they have. Only in this way can we be user-oriented. Therefore, we design products according to the information of users' portraits, which is the strategic and tactical guidance of our company.
Exploring the user's footprint to form the user's orientation actually means that we can further deepen the product after we have a detailed understanding of the real user and how he interacts with the relevant content of the product. When we paint a portrait of a user, we must start from the business scene and aim at the business scene to solve an actual business problem.
For example, we have to make a portrait, either to gain new users, or to improve the user experience, or perhaps to save a lost user. In a word, I must have a very clear goal. These are some efforts to build user portraits from enterprises.
From a technical point of view, we can help build the underlying data foundation and serve the upper application through the construction of user portraits. At the same time, it is also convenient for information processing at some levels.
Why? User portraits can not only be displayed intuitively, but also serve some upper-level applications. In fact, as mentioned by several teachers just now, for example, in the recommendation system, the user portrait exists as a very important part of the recommendation system, which greatly improves the recommendation effect.
In addition, like the application in finance just mentioned, the user portrait can also be applied in the application of risk control, so that some regular characteristics exist and the system level can be quantified.
The so-called convenient information processing is actually that after we mark it, the computer can conveniently handle some quantitative requirements.
For example, do some classified statistics, a video website has a very hot talk show recently, and I want to know how many users are watching the talk show conference, and what is the ratio of male to female; Or you can do data mining, and what clothing brands do users who like to buy durian usually like; Or users who often buy coffee and garlic, what is their age distribution? Wait a minute. Can help us do some quantitative analysis.
In short, user portraits can perfectly abstract a user's information panorama, which is the basis for enterprises to apply big data.
User portraits can help enterprises and provide users with personalized products and services. We are always talking about thousands of people. The ultimate goal of every enterprise that provides services to customers is that when a user opens a product, an APP or a website, the content he sees and the experience he gets are designed for him or conform to his tonality. Only in this way can his experience be really improved.
For example, combine the user portrait with the scene of marketing automation, and give a simple example of portrait support business.
When we want to hold a marketing activity, it may not be aimed at all user groups, but at a certain type of users with certain characteristics.
First of all, we will use the user's portrait, according to our portrait of the user, or according to the label established by the user to circle the crowd.
According to the business situation, to determine the user group, we can circle it in the group through conditions. For example, in this example, users who have not purchased within 360 days are circled by a label of consumption behavior.
When I really do marketing activities, in marketing automation products, when planning and implementing marketing activities, I can use the newly screened user groups as standard groups, which is the part I draw red circles. We can easily achieve targeted marketing.
Generally speaking, what is the role of the portrait as a whole? There are several connections between the portrait and the label mentioned just now. Let's talk about labels in detail. What is the function of labels? Why is it so important?
Tags can transform data and the information it contains into guidance with clear decision-making behavior. The more people participate in decision-making, the more they need to tag data/information, so as to improve people's understanding and processing efficiency of data and realize man-machine cooperation.
What exactly does this summary mean? We can talk about it in detail.
Let's take a look at these so-called data products, such as DNT and CPP in the marketing field, or if you are not familiar with them, you can give some general examples, such as BI. The characteristic of these data products is how to use data to make people understand quickly.
To give a concrete example, under the current epidemic situation, there may be a label that everyone often encounters. When we take our temperature, if it exceeds 37 degrees, we may have a fever, so we need to see a doctor, and we may be taken away for isolation. This point is an important label, that is, a hot spot.
For example, when driving, if the speed exceeds 120 yards per hour, the display on my navigation will turn red. This is also an important label. It will tell me that I was speeding.
Therefore, in our daily life, we will always encounter some particularly low-level labels. For example, when it comes to a label, I can know what to do or what to do next. For example, when you see speeding, you know it's time to slow down.
Let's look at it from another angle. Why do we need to label it? For example, our current news or short video app has a very good recommendation behind it, and these recommendations may not need labels. For example, if it knows you like this thing, it will push it to you, and then you will see it. It may recommend another one to you, and you may watch it again.
When these data are processed by the machine, it may not need to understand what it is, because it has a lot of feedback data from users to help it make decisions, so it can constantly refresh and train the model.
But if people need to participate in decision-making in some scenarios, that is to say, the more places people participate in decision-making, the more we need to label the data. Why? Because people can't process information through a large budget like machines, after all, the information they process is limited.
Therefore, in order to make people understand the data quickly, we can improve the processing efficiency and finally realize man-machine cooperation. The purpose of using tags is to turn the results of checking a large amount of data into the form of quick understanding and quick decision-making through tagging information.
As we just mentioned, in data products, if you can turn the data into a clear label to remind users that you need to take a break now, or remind users that you have a fever, it's time to see a doctor. This is a very good label, because it speeds up our processing and directly helps us make decisions.
Let's take an example and take a closer look at the role of labels.
The function of this label is also related to the flow direction of the label to be mentioned later. In this place, there are four steps. Let's see how it is transformed into a label step by step through these steps, and finally guide us to make decisions.
The first step, the leftmost column is called data online. What does data online mean? That is to say, I want to make our business process online through digital transformation, and the data generated in the business process will naturally go online, which has the condition that people will not deal with it.
For example, we used to buy things in the supermarket, and it was difficult to count what each user bought. In this case, your subsequent analysis will be very difficult. What about now? Many people buy things on e-commerce, and the data of shopping links are online. In the future, as more and more business processes go online, we will analyze more and more content.
Back to this example, a piece of data or a purchase process I made online has been recorded in my system through online shopping.
After the data is online, the second thing we have to do is to carry out a process of transforming data into information. For example, we now see that the user's name is Wang Erni. What information is this? In fact, this information is, for example, what kind of information does the user convert into information? This information means that it can be explained in my business scenario.
We see Wang Erni in this example, which is a typical purified text. This may not be a very direct change. But this is also a process from data to information, and it is also based on our understanding of business scenarios. We not only need to directly analyze and transform data. In fact, we can attach some new information to these data, which is what we call information conversion and a controversy about our information.
The third step is to turn our information into labels. For example, we can make some rules. When I read this message, we can judge that this person is a woman named Wang Erni with a high probability of more than 90%. So I can put a label on her at this time, which should be regarded as a gender label. For example, in this example, I gave her a label called woman.
The fourth step, how do I make a decision based on this label? In this tab, I just make a decision. When I communicate with this user in the future, I am more likely to use a mother-like communication method and address. Why is there such a decision? It is because we found out that she bought baby milk powder.
In the e-commerce scenario, one of our users bought a baby milk powder. How should I interact with her next? In the process of making this decision, we may judge the probability of being a mother by combining her gender tag, so we should use her communication method and address to communicate with her.
Will there be some mistakes here? Of course, there must be some mistakes. For example, if I buy this thing for someone else, this may happen.
But we will do more in-depth analysis. For example, we saw that she bought three pieces of milk powder, which means it may not be newborn milk powder. Generally, once a child is two years old, no one will visit him again. Usually when he is born, it is more likely.
Therefore, combining a variety of labels can help us decide what kind of communication mode it is.
In this link, we don't ask this decision 100% to be correct, because in most businesses, we just need it to recommend us, at least let us try.
To be continued ... please pay attention to "Know you better than you, talk about user portraits (3)"