Hacking Book | Free Online Hacking Learning


du xiaomeng: my big data "mind reading skill"

Posted by patinella at 2020-02-27

Du Xiaomeng: my big data "mind reading skill"

In literary history

I'm afraid Du Xiaomeng is an expert on "mind reading".

This is not because she is a beauty with strong intuition, but because he is a Ph.D. in marketing modeling in Peking University. She believes in numbers more than intuition.

To introduce you, Du Xiaomeng is the chief data scientist of percentile group.

Du Xiao Meng

She described a future world full of big data and intelligence for me. However, if you want to understand what she is talking about, you may need to "brainwash" - I mean to clean your mind and build some background knowledge that may be anti common sense.

I guess there are two things that may be important:

1. We have been busy for thousands of years, and the most basic direction is to improve the efficiency of resource allocation.

2. The effect of big data and artificial intelligence to allocate resources has been able to surpass human experience in some fields.

Although many people outside the industry are not familiar with percentages, most people have used percentages of products, but they don't know it.

Founded in 2009, before Du joined, the company had a reputation. At that time, the trump brand product was "e-commerce personality recommendation system". When you log in to e-commerce, you will always see the recommendation of "guess what you like" in a conspicuous area. In its heyday, percentage points won most E-commerce markets except Jingdong, Taobao. Hundreds of millions of people used the system.

(say, guess you like this kind of thing, I will do it too. For example, I guess you like this

A guy who's stuck with data

In principle, a person's behavior can show his preference, so-called body is honest. Therefore, what a person has bought, seen or wandered in front of which shop window may hide his preferences and future choices, which is the principle of personalized recommendation.

In other words, this is a relatively simple big data application scenario.

The reason why the recommendation system understands you is that artificial intelligence algorithm occupies a very important position in it. In fact, although it sounds like a very commercial system, its creators are all academic bulls. For example, Su Meng, the founder of percentage points, is a doctor of management from Cornell University, who once taught in Guanghua School of management, Peking University; Liu Yijing, the chief architect, is a doctor of Applied Mathematics from Peking University, etc.

From the founder's background, we can see that around 2010, the barrier of "personality recommendation" is very high, only doctors can handle it. However, with the dawn of AI and big data approaching, percentage points began to face some interesting changes in 2013.

Since 2013, we have found that the traffic of many e-businesses is seriously declining, so we speculate that the industry may be "oligarchic". Now it seems that the judgment at that time was correct. Many e-commerce customers we served before are now gone... So at that time, we had the opportunity to make a transformation.

Du Xiaomeng said.

The transformation she refers to is to shift the focus of business to the enterprise level big data market. Although it's a transformation, the core technology hasn't changed - by analyzing the data, find ways to improve resource allocation.

The first one to find them is TCL. In 2013, TCL also felt the severe impact of e-commerce on the traditional sales system, and the team began to be poor. TCL decided to "save itself" and look around the market. The "personalized recommendation" system provided by percentage points seems to fit their imagination of the future world. So, for percentage points, this business is actually a door-to-door one.

"Enterprise data decision making" is a simple way to popularize science, which means that enterprises use their own customer information from all aspects to "guess" the customer's ideas after algorithm calculation, such as who will like this new TV better, whose mobile phone has been bought for a year, may buy another one, what model of mobile phone TA will like, and so on. Through these guesses, targeted production arrangements, marketing, adjustment of online and offline channels and so on.

In other words, this system is much more complex than an e-commerce recommendation system. From the beginning, we have faced the problem of data fragmentation. Take TCL as an example. Some user data comes from offline sales stores, some registered user data comes from official forums, and some data comes from third-party e-commerce such as Taobao. To analyze these data, first of all, we need to get through the data of each module and standardize them, which requires many marketing postures. Fortunately, it's not a problem for Du Xiaomeng, a PhD in marketing modeling in Peking University.

At the beginning, TCL and percentile both thought it was an attempt, but from the result, the achievements of all previous promotions were beyond imagination. Percentage points are booming in the field of enterprise data decision-making.

Through the analysis of data connection, users can be divided into different types.

The awakening of traditional industries

Percentage points embarked on a path of "saving" traditional industries.

Many people have more or less misunderstandings about traditional industries, and they feel "the sun is falling", but the real situation is not so. According to the forecast report released by IDC, at present, more than 76% of the domestic digital economy is contributed by traditional industries. No matter from the volume or the position in the national economic structure, traditional industries can not be absent.

In Du Xiaomeng's view, the "awakening" of traditional industries stems from pain. For example, the reason why financial institutions embrace big data is because of the huge impact of Internet finance; the reason why retail malls attach importance to "fine marketing" is that e-commerce led by Alibaba has cornered them.

So, according to the order of pain, there is an interesting order for traditional industries to embrace big data and artificial intelligence.

1. At the beginning, it was TOC's manufacturing industry, because if they want to produce products suitable for the market, they urgently need to establish a connection with their own users and "draw" users.

2. The next thing that started to wake up was the financial industry. In fact, the innovation spirit of the financial industry is very strong, so although they don't have great business losses in front of P2P, "third-party credit investigation" and consumer finance, their values have been greatly impacted, so they embrace new technology keenly.

3. Almost at the same time as finance, the retail fast-selling industry has also begun to wake up. Because the panic brought by the impact of e-commerce is very big, they need to fight back.

4. Then it is the traditional media that awakens. They are facing the impact of new media and are eager to establish closer relationship with readers.

5. The latest awakening is the government. Because many big data application models have been prepared, which can effectively improve the efficiency of government affairs.

Du Xiaomeng said.

The big data system of cities can accurately guide specific policies and plans.

"Preacher" Du Xiaomeng

Huawei is a milestone customer.

They need to be able to deal with a large number of complex data, while strictly ensuring data security.

However, the hardest thing is not hard work, but "preaching".

In 2014, grassroots employees didn't have a clear understanding of the role of big data, so each department at the beginning didn't understand why they shared their data to other departments. Du Xiaomeng's heart, of course, has a blueprint. Intelligent decision system is like a brain, data is like knowledge. If one learns more knowledge, one can be more creative. The sharing of data is a huge potential return for enterprises.

At that time, however, Du Xiaomeng, as a data expert, and various project teams worked hard to popularize the principle of big data, listening and persuading at the same time.

Du Xiaomeng has experienced many things like this.

Every time before, I emphasized that I was a doctor of algorithms, talking about data-driven, business efficiency, potential customers, and algorithms. Most people listened with a smile out of respect for the doctor from Peking University. However, there is no later...

She said.

Three years ago, China's marketing was still the era of "personal heroes". A big bull CMO (Chief Marketing Officer) has the supreme right. With a budget of 200 million yuan, one can decide how to spend money. For example, buying a billboard and betting on a certain advertising channel. Ask him why he chose it, and he'll say it's experience. Their trust in themselves is far greater than their trust in data.

However, Du Xiaomeng has seen that in the past three years, there are fewer and fewer CMOS that have achieved success through personal creativity. How much money does an advertising channel need to put in? It needs to talk about data; whether a product should be promoted to 30-35-year-old people, or to 25-30-year-old people, it also needs to talk about data. She recalled that it was slowly changing. Start to eat crabs from the industry leader, and only when you taste the sweetness will you slowly drive the whole industry to respect major data, and use big data.

Of course, these years of painstaking "preaching" have also created Du Xiaomeng, who is now eloquent...

Back to Huawei's project. The final result is that Huawei, after a / B test, sends promotional messages to a group of users generated through analysis, and also sends the same promotional messages to a group of users selected through manual analysis, with a 50% difference in click through rate. At that time, the efficiency of the system was so high that even Du Xiaomeng himself could not help marveling.

A / B test, by sending the same information to different groups of people, or sending different information to the same group of people, can test out more advantageous communication methods.

Application of "wonderful flower" in intelligent decision

It's 2017.

Over the years, Du Xiaomeng is more and more confident. In her opinion, the reason why many government projects will find percentage points is that they are also interested in percentage points' understanding of specific industries.

Objectively speaking, the data in the hands of the government is very large and of high quality. But it has always been criticized: the government has so much data, why not open to society, promote economic development.

According to Du Xiaomeng, the data of various governments have one common feature, which is "chaos". Legal person storehouse, resident storehouse, geographic information storehouse, economic information storehouse, industry, commerce, civil affairs, taxation and legal affairs storehouse all fight for their own. Many local governments may not know what data they have. At this point, someone should come forward to help them get through the data.

What will you see after you get through the data?

Du Xiaomeng said that from the perspective of existing customers, after getting through the data, we can see the development status of each area of the whole city and the specific characteristics of each key area through a panorama. For example, what is the situation of agriculture, forestry and animal husbandry in a certain area, and what is the population aggregation mode in a certain area. According to these, we can formulate industrial policies to guide population flow and industrial development.

You can imagine to what extent the government will promote the reasonable development of the city if it adjusts its industrial policies according to big data.

In addition, Du Xiaomeng also introduces some interesting application fields of big data.

For example, a radio and television service provider has 48 base stations around the world. However, the equipment in these base stations will be damaged unpredictably. In order to ensure the smooth operation of the service, the damaged equipment needs to be replaced constantly.

There are two ways:

1. No matter whether it is damaged or not, replace the equipment regularly every 3-6 months. This will cause waste.

2. Repair the damaged equipment. This has resulted in a decline in the quality of service.

In fact, there is a third solution to this problem, which is to use big data intelligent decision system. By analyzing the real-time parameters of each device, such as voltage, current, high-frequency electricity rate, water temperature, etc., a health judgment model is established. By using the method of artificial intelligence, the parameters can automatically predict whether a device is about to be damaged. This method has been proved to save a lot of cost.

Similar to this, intelligent decision system can also do "bridge railway health inspection". Simply put, a detector is placed every 5 meters on a 300 meter long railway bridge to measure the tension at each point after the train passes. It can predict how many years the bridge will collapse.

Recently, Du Xiaomeng is working on a new project.

That is to establish a mathematical model by collecting the data of signal attenuation of the operator's signal tower under various weather conditions, such as heavy rain and blizzard, so as to help the operator automatically adjust the power of the signal tower under various weather conditions, so as to ensure the communication quality of the telephone network.

There are countless applications like this. Du Xiaomeng can't stop talking about these things.

She felt that "intelligent decision-making" itself was of great significance to society.

Introduce yourself again. My name is Shizhong. I'm a technology journalist who loves stories. My daily life is to chat with gods. If you want to be friends with me, you can add my wechat.

If you don't want to lose, you can also focus on my official account of the media.