Hong Times: How to use large data to play digital media

2014 Zhongguancun Large Data day on December 11, 2014 in Zhongguancun, the General Assembly to "aggregate data assets, promote industrial innovation" as the theme, to explore data asset management and transformation, large data depth technology and industry data application innovation and ecological system construction and so on key issues. The Conference also carries on the question of the demand and practice of the departments in charge of the government, the finance, the operators and so on to realize the path of transformation and industry innovation through the management and operation of data assets.

In the afternoon operator @big Data link, from Admaster's flood times to make a keynote speech about how to use the data to play digital media. The following is the full text of his speech:

Hong Times: In fact, we speak of the entire digital media this biosphere is in fact the whole field of large data may spontaneously play one of the most fire field, and we talk about carrier data is to provide the entire digital media very good grassroots business support, we see the entire digital media ecosystem what kind of role division, just actually have talked about the media, A variety of Web sites is the media, the second refers to advertising, the third is the audience, telecommunications users, mobile users, including broadband users, that is, we speak of visitors, the fourth is of course the electricity business, that is direct marketing channels. The fourth one is audience interaction, the fifth is advertising push. So finally, we say that the Internet biosphere is the accumulation of data on five dimensions. This is a trend that we see, from 1980 home began to popularize television, television is a single transmission channel, and then the Internet, the first is interactive, and linked together, and finally see the Internet more and more, from the beginning of the internet to mobile concept, mobile phone, PAD, and finally we found that back to the original point, Television, Shanghai TV in the field of digital television has been very hard. So in the end you'll find that all of our audience is moving, our media content has been grabbing all sorts of time, and then generating all kinds of data that gives us a wide variety of information, so it forms a very good internet closed loop.

Everyone saw a picture, are eating, are playing mobile phones. Look at micro-letters, micro-blog, including Friends Circle, the company circle, are in the micro-letter communication, so this is a lot of data. So in the end you'll find that the Internet has included mobile internet for the last ten years, and you'll find that the end of it is the right bottom, survive are high viscosity, the highest viscosity is the mail alive very well, and then slowly news, we see a lot of traditional paper media, television, are in transition, they want to become a point of view of the viscosity of the state. Internet marketing methods are also changing, the first is to smash a lot of money in the Electricity business section to buy a good advertisement, cast TV, slowly become advertising precision. And buy a doctor's glasses how to define the audience, also we see the Chinese good voice how to divide the audience, also we can see the advertisement and social linkage. How to linkage? Before we did a case, a car user he has millions of fans, what kind of relationship does these users have? No one knew before, and we used the new trend to compute the graphics, we know that the entire human network is a graph, so the graphics card is naturally calculated, so we analyzed, and found that a fan of such a computing brand naturally reported 18 groups, two of which are zombie powder Regiment, He has two companies in the history of powder, one is related to his celebrity spokesperson, was invited to take part in the test drive, he invited to take part in the shooting Star is a fan group. The other one is the dealer. If you advertise, don't vote for the four fan groups. So the rest of us should be internal, the remaining 14 groups to advertise, social and advertising is also linked, so this is a large data analysis of the comparison has a typical case.

Let's see, the Internet is also starting to do special events, some different seasons different areas of different products to do activities, they through data mining to marketing planning this activity, tomorrow is double 12, one months before the double 11, double 11 is the electric business platform linkage, how to improve conversion rate, I see ya letter colleagues mentioned, By increasing the conversion rate by 17%, it means increasing the conversion rate by 170% in the future.

This is our history with customers to do a very classic case, one is the standard rate, geographical distribution, PINCI control, content repertoire orientation. Everyone looks more than eight times higher, many people will find it very annoying, I believe that everyone watching the video when the front dozens of seconds of advertising is very uncomfortable. The same region, is not able to advertise, we found that there are differences in the region, we and partners are considering how to analyze how many of the errors caused by the routing settings. For example, there are no 4S stores in the area, so they don't advertise for sale. If two media users overlap is very high, is not another small media can be replaced, or the role of reinforcement.

The front is the statistical production of large media data. It's microscopic data, like a user, we look at the website design, from the user access path to see is not reasonable, Sue is a, but your interaction is B, so we pass the path is also a map network, which path is the user's favorite shaft, which path is the easiest way to achieve the user to buy the path, This is our way to help customers improve the efficiency of conversion. We also find that there is a very interesting data, it is really useless to see the ads? No, we found in some telecom websites to see advertising users, after watching the ads three days after the return to the electric platform to buy the advertising rate is not to see the ads three times times. And this user is loyal to the old user, and his loss rate is very low, which is a big data mining out of a very interesting phenomenon.

Of course this can be in the future with a lot of operators, operators will not help false information, how to discriminate? For example, it is very convenient to know that this is a robot, is always playing, this is the letter and operators know that the robot he constantly brush cookies, it is the means of simulation.

Just talked about the anomaly data screening six dimensions, we see many operators inside, we cheat when the biggest cheating is the robot, the machine control people, or call the robot, the operator itself to do security when there is a certain model. and the entire advertising industry is cleaner, so advertisers are willing to spend more money and produce more value for him.

We have just been talking about some cold numbers, we found more is to realize the value of the first advertising is not able to see? In what kind of environment is seen, video ads and ordinary originally we see banner advertising the biggest difference is what? You will find that the video is almost exclusive in space, but the time is 15 seconds, then the original small button ad in space is not exclusive, but in time is always exclusive. So the results of these two possible spaces exist, but the effect is completely different.

Also referred to the secondary media. There are also audio ads, and recently a lot of apps are starting to grab the ear, eyes have been robbed, so some listening to music software began to do, like Simalaya, FM, they are also starting to advertise the way, including this morning to see the news Google has been in the United States to the outdoor LCD screen success can be priced in real time to launch , of course it will refer to a lot of parameters such as geographical location, human traffic, as well as the surrounding advertising interaction between different circumstances.

The most traditional microscopic is research, research is what, we take samples to statistics, do statistical indicators to map, this I think is more interesting things. How to build a better sample, we talk about the biggest problem of the sample is to identify the true and false. We think that the operators of data natural people and data, people and distribution, even the use of packages, consumer simple statistics can do simple consumer behavior, of course, simple statistics to deal with sensitive data. How we used to get this person's income, now you also use to ask, the operator found that every month, the number of text message 20th has been issued 18500, operators can fully know, but he did not know who. Used to say you've seen this kind of ad every year? This time there is a paradox, this time to see ads is not counted ads? Only through real monitoring data, how many times have we looked at ads, in which media to see what form, through this collection process, we collected 2 million of the sample library, next we will do with operators more accurate, covering a wider, because now mobile phone has 9 months to 12 months of the machine cycle, We associate it with anonymous methods to provide some sample durability.

This is our actual in our industry to advertisers to do the brand value of the indicators, these indicators in fact through the research method calculated, when the survey sample is relatively large when it is a relatively large amount of data statistics, we have a database has a branch is tens of thousands of sample library, There were almost a hundred thousand of of them, and that was a very large one, and we saw it as a table of extraordinary forms, and it took some work to analyze it.

This is a real-time case of all media comparisons. I will not talk about the time relationship.

We are back to social, mobile, there are a few things with operators are very related, operators are doing, WIFI, mobile, telecom Unicom are doing WiFi access, recently CCTV also reported not to WIFI, may be stolen information. But with WiFi it's a gateway, and privacy can be extracted. Then we can pass a specific channel code coupon to use the coupon conversion rate, this is not accurate enough. Similarly, we set two conditions, for peacetime consumption of sports products to sell hidden glasses and usually go to the hot pot of the hidden glasses, through the SMS issued coupons and credit card issued to the coupon is not the same. This is an intuitive transformation pathway. Similarly, our advertising in the past is very hard, now slowly softened, the past said APPG virtual props, such as a certain Cola movement expression, will be paired with sports products to allow users to use, the purpose is to accumulate user behavior, I can target your users of the use of records, just we see the user portrait. So the same operators I believe can do a lot of such things.

Just now we're talking about some actual application scenarios, our methodology. In fact, I do not want to mention the things you are very familiar with, such as Hadoop, such as data Association analysis, data modeling, data mining. You see, this stuff is all about analysis, you will find that the internal public opinion is very important, we have a consensus with the internal is that advertising is a number, but it is a symbol in front of, or negative is very important, if you have recently been negative, we see the results of advertising, this is a television exposure of the brand. So we spent a lot of effort to do such a laboratory, specialized in public opinion analysis, divided into a few parts, who on which occasions said what kind of words and what brand relationship, said how. Our Sina Weibo is our biggest partner. After climbing the public opinion to do emotional analysis, emotional analysis of our identification rules engine, can quickly set the theme of the argument. In the past how to do emotional analysis, such as So-and-so object is very good, but another brand is a bit poor, I would like to say that the past emotional judgments to determine what? Neutral, because there is a good word has a bad word, but our engine inside two brands, a brand good, B brand bad, so it will be dismantled, because we do with the traditional content. The same good news or a bad message, how to be forwarded, forwarded to the number of people, it is not covered to your fans. Finally, you can count whether you are good or bad, depending on whether you are finally positive or negative. We all know that cola, two years are on the Coke Bottle label, cola These two years bottle is we do, we have to pick out 300 hot words, like what kind of song, like what kind of self-proclaimed, they put 30 more tall on the positive word, and then we can buy a small fresh coke bottle inside the supermarket. Then they gave Leehom a box of Coke, and Wang volunteered to receive a bottle of Coke, then called my singing Leehom, I traded a box of coke cost hundreds of thousands of times for forwarding success. It's not just advertising that we talk about, it's fun, but you have to rely on the data.

This of course is back to the last we talked about the pre's engine, we talked about marketing is advertising, there is an O word, advertisers have their own positions, their own sites, their own stores, these advertisers are their own od, these OD itself does not have any traffic, we need to earn traffic, just talk about social, How many people would like to share you, how many people would like to bask in the list, it is willing to follow your official website with the flow, of course, this can be the only marking. PC Internet cookies are well identified, and the carrier's future ID is not an operator's digital identity. If a family has multiple Internet accounts and multiple Internet accounts, you can tell which phone the family is bound to. This identification allows the orientation to be found under any environment, as the approximate portrait of his behavior can be analyzed by means of a logo. The behavior portrait has many many sides, for example this is HQ1, returns to the human side, many data sources can draw a person's side, but how to draw a more whole side? This is the level of discussion we have with our operators today. How do we get these sides together and advertisers themselves are part of it because you might have bought something for advertisers. We put all this together, we can find the right person at the right time, on the right occasion. The recent double 11 Taobao words, will find you search for things haven't bought, will be in micro-blog and other occasions everywhere to see Taobao about this part of the ads. At that time you will think Taobao how clever, then more you may be disgusted. And then you'll find that if we can get all these tags into IDs, there are so many forms of advertising, may be related to your needs, this time, of course, some people are afraid, I have no privacy, my world has been found out, but you sometimes think my life is very convenient, I go out when may want to call a car, This time will feel a social small assistant constantly reminding you.

Of course, this is how we advertisers with television and the Internet, the future of the digital is better cross, of course, there are curve models and data models, but also based on our many experience mining, do a good integration. This year we did a very interesting model called sponsorship model. We use this modeling method to evaluate large data to make the sponsored things more accurate and can be evaluated.

So finally we go back to the integrated marketing, we talk about the eco-chain inside you'll find a word that's really important, just what I mentioned, the right time, the right place, the right person, the right thing to say. Any operation is accurate, also you can estimate, I should spend how much money, what kind of media, I in the electric business website How to stock, my product should how design, I for different people to develop new products.

This is the result of our double closed loop, which is The advertiser, and you will find that everyone is developing in this double closed loop structure. At first you develop this product, you need to promote this product, these are data to help you provide data decision-making, and then in the integration process will help you do two times, you are in the network channel sales, or online sales, this comment back to the Internet, where comments can redesign the next generation of products, Next-generation products can be promoted at the same time, feedback, and then do the data model. If this ring is done, it will find that the product continues to do, this product will not die, because there are people in the inside to make it better to go down. I'll be here today and we'll have a round table later to discuss more practical things. Thank you!

(Responsible editor: Mengyishan)