New Development of Big Data Business Modeling —— From Prediction to "Field Testing"

From Prediction to Current Testing: New Development of Big Data Business Modeling

Where is the new development direction of big data business modeling? How to predict the unique advantages of big data? How is it achieved from prediction to "field measurement"? Professor Chen Yuxin, the chief model scientist, gave us a more in-depth explanation, from forecasting to "on-the-spot testing": the new development of big data business modeling.

The following are excerpts from Professor Chen Yuxin's speech at the 20 15 percentile big data operating system (BD-OS) and the D-round financing conference:

Thank you, thank you! Today, I am honored to share with you some thoughts on the new development of big data business modeling, which is also a frontier of current percentage research and development.

As we all know, forecasting is a core of big data business applications. Big data prediction needs a lot of high-quality data and very advanced models. What are the new hot spots or new development directions in the field of forecasting under the current situation? How to predict the unique advantages of big data? That's what I want to share. Before that, I'd like to introduce two recent news reports in the media, from which some new development trends may be seen. Both news items were published in the Wall Street Journal in August.

First, Apple and Google are developing technologies to know what users want before they want it, and to tell you what you want before you know what you want. Everyone has heard of these two products. Apple calls them "active assistants" and Google calls them "Google Modern". These two companies speculate on what you want to do in the near future by finding out what you will definitely do in the future. Google can know by email that you have a flight at six this afternoon. At three o'clock in the afternoon, according to your current position and the traffic flow in Beijing, it tells you that it's time to go now. If you leave, which bus should you take, Didi or Uber? This is its prediction idea, which is to predict what to do in the future by grasping some known future events.

Another news that seems to have nothing to do with this, but has something in common, the name of the macroeconomic indicator forecasting company is "Xiance", and the founder is a doctoral student who won the Nobel Prize in Economics at Columbia University. Big data predicts macroeconomic indicators, such as price index. Why is it called "now test"? It is not a forecast, but a quick summary of all the price changes that have just happened in the whole United States and a description of what has just happened, rather than a real forecast of future prices. Why does this matter make sense? Because usually the index released by the government is a month ago or a quarter ago, and it can be done a minute ago.

Two news stories have the same keyword "now". Now, this is a very important trend of big data application, that is, from prediction to current measurement, prediction is a core of big data modeling technology, but it is also the biggest difficulty of big data modeling. Everyone says that big data is very powerful and can predict many things. Can you tell me that China's stock will go up 100 tomorrow? But the advantage of big data is that many times what we really need is not prediction, but the so-called current measurement, a description of the very near past and a prediction of the very near future.

What do you mean very close? A day ago or an hour ago? This actually depends on the amount of data we have now and the development of technology. The trend is that this is getting shorter and shorter. It was originally the economic data a month ago, but now we can know it a minute ago, and we need to know the data a second before high-frequency trading. This is the definition of "field measurement" that we think of. Now it is a dynamic process. Field measurement refers to the description or prediction of our present situation. This is actually a prediction model that is really used by a large number of big data applications, such as high-frequency trading. In fact, we already know the market situation, but we only know what happened recently before others react.

It is a very important recommendation engine for big data applications, and it is also the most primitive DNA in percentage points. According to the current situation of consumers on this page, give him a current recommendation immediately. This recommendation may only be RTB advertisements and taxi software put in real time within one second or even half a second. Everyone has used the Didi special car. The principle behind them is the principle of current measurement. I know where the car is, where you are and the traffic conditions, so I can guess the next minute. There is a very important scientific principle behind what just happened and what will happen. Nature has given us a very important means of prediction. Everything in the world has so-called inertia, and when this inertia efficiency occurs, our prediction can be more accurate. When an object slides down an inclined plane, it is very accurate to predict where the wood block is in the next second. Why the current measurement accuracy comes from the inertia behind everything in the world.

In the social field, inertia is scene-driven, such as buying a birthday present for my wife tomorrow. With this goal, according to the principle of inertia, prediction becomes a reality test.

One advantage of current measurement is that it uses the so-called inertia principle and time difference. This is the characteristic of big data, because big data is often real-time data and massive high-frequency data. As you can see, if I take a picture like this every ten minutes, I may miss this shot. If I take a picture every microsecond or every half second, I may see this shot. When I saw this shot, there was an application of inertia. The lady picked up this orange. If there is no time difference, for example, shooting this shot every five minutes will miss it, but if the frequency is high, we will know that she took this orange to see it. Although the child took it away, she was very interested in this orange. In this case, we can have some marketing tools.

Speaking like a prediction is actually a descriptive problem. The description effect of high frequency is similar to that of prediction. We are not predicting whether she likes apples or oranges, but our description has changed from prediction to actual measurement, which is very advantageous, so we have turned a difficult prediction problem into a description problem. That's why I said that we have the advantage of big data field testing.

I have several doctoral students doing model building and model testing. According to the concept of field testing, we can develop a series of technical models, which have a series of commercial applications. One is collaborative filtering and time series analysis. Usually, in the field of computer science, we all know the algorithm of collaborative filtering. If we consider the current measurement, just like the time series analysis of econometrics, time series analysis is widely used in high-frequency trading. Now we can combine collaborative filtering with time series analysis to make a corresponding application.

The second is the portrait of the user. Combined with fast iterative Bayesian learning, we know what kind of people users are and what kind of things users will do, but we can look at the previous requirements in specific scenarios and make corrections. This correction must be completed in real time and iterated quickly.

Thirdly, some dynamic data visualization and human-computer interaction products have been developed. The advantage of the human brain is that a comprehensive grasp of a scene can get better predictions than computers, and big data can intercept such information quickly and in real time. If some data are displayed dynamically, it can be judged through human-computer interaction. Under the condition of supermarket staff, it is not necessarily accurate to judge whether a lady likes oranges or not, but we can combine people's prediction of some panoramic views with computer's capture of data through dynamic data visualization.

At present, some scholars focus on anti-time series prediction based on operational research and behavioral science. In other words, predicting the present with the future sounds a bit suspenseful. Many times in our distant future, you may buy a plane ticket and fly tomorrow. This is the exact future. We have a clear future and use it to help predict the present. For example, if you are in a department store or Wangfujing department store, I at least know that you must go out and can't stay in it all your life. If you want to get out of this department store, I can tell you several choices, which shelves each choice has passed, and how many possible goods are recommended by the logistics optimization method. It has quite a few applications to turn several very difficult forecasting problems into logistics optimization problems pushed back from known endpoints, which is also a very close combination of what we have done before and what we are doing now. A few percent of DNA has been recommended in real time, which coincides with the current measurement.

Percent has done a lot in modeling, that is, drawing portraits of users and refining scenes. After careful deliberation, we refine the scene further, not only to study the user scene, but also to study what the user's goal is in this scene. My purpose in this venue is to communicate with you. Any user has a purpose in any business environment and any scene. The purpose of going to a restaurant is to eat, and the purpose of going to school is to study. We found this purpose. Through the technology I just talked about, we know that the user portrait pushes his current thoughts and behaviors. At present, the established data system and scene segmentation, including the user portrait, have many applications in the current measurement range. Therefore, accumulating a large amount of data has a very good prospect.

Like big data forecasting companies, many financial-related information and real-time economic indicators predict the future by describing the past. There is also an area related to financial information and personal credit reports. Judging from the current measurement, the so-called demographic indicators, indicators of past behavior, now have more practices. When a person buys a big commodity and a car, this information will be immediately fed back to your personal credit information system and the changes in your cash flow level. Such a real-time monitoring method is similar to personalized enterprise marketing.

In the same example just given, personal finance and credit cards can calculate how much money you can use this month, the total amount of consumption each month and the amount of bank deposits according to the changes in your real-time consumption. According to this, you can push back what you bought, otherwise the users you recommend may not have the financial resources to spend.

The third application is tourism management. For example, your flight, I know the weather will change tomorrow, so I can know that you will have a meeting in Beijing. If the weather is bad tomorrow, I can remind you in real time whether to change your ticket to a train ticket. This is the concept of current measurement. Knowing some uncertainties in the future will, in turn, help you solve the uncertainties.

Personal health management, the goal is to lose five pounds in three months. Based on this, we can infer how much to lose and eat every day, and adjust our health management plan in real time.

Finally, by putting some things together, you can build a so-called digital life assistant to manage all aspects of your life and provide you with the best experience.

Why do you repeatedly emphasize the importance of in-situ measurement, and the application of big data has great advantages? In the past, big data often mentioned this point, the so-called three V or four V, but usually when people discuss big data, this V is often speed, which can quickly process massive data. When you know the past and the coming future, in addition, the relative advantages of big data have special advantages for short-scale applications and no obvious advantages for long-scale applications.

In the long run, I can say that I don't need big data either, and I can succeed with small data. But short-term prediction, I'm not sure what everyone will have for dinner tonight. Big data often knows this because it knows what you ate at noon, where you came from, what you like to eat, where you are now, and what restaurants are around. Often you can infer what you want to eat, when you eat lunch, your height and weight, and when you will be hungry. We can push back and make a series of introductions. This is a very big advantage of big data, so we should pay special attention to it in business. Mining and using behavioral inertia through big data user portraits and detailed scene modeling. This is a new development trend to enhance the value of big data and foster strengths and avoid weaknesses. I hope you can correct me and communicate more. thank you

The above is what Bian Xiao shared with you about the new development of big data business modeling from forecasting to "field testing". For more information, you can pay attention to the global ivy and share more dry goods.