A keynote speech on the role of data analysis in promoting mobile order transformation

August 28-29th, the revamping annual meeting in 2014 was successfully held in Beijing Kun Tai Hotel, during the two-day conference, the first day was three special sessions of mobile, channel and brand, the main venue was the next day. The theme of this annual meeting is "broken", that is, "broken, break, break", the aim is to break the plight of the traditional electric business thinking limitations, set up new ideas. Dedicated to create a platform for innovation in the electric business industry, from a professional perspective for the electrical business colleagues to solve the entrepreneurial process encountered in the most practical problems.

Star Wardrobe Song Weirong on the 28th Mobile special session titled "Data Analysis in the promotion of the role of mobile order Transformation" keynote speech.

In his speech, he shared the role of the star wardrobe in using large data to pinpoint the user-approved process, and the star wardrobe is trying to create a need to try to buy a way for users to choose the items he likes to buy. He said: "Why should we pay attention to fashion, pyramid tip is the source of fashion." And we are now from the bottom of the pyramid to the source, to go up to guide. Fashion from the top down Guide everyone shopping guide, since the spread of the downward, we can learn how stars, fashion talent is how to wear. Such large data we have to get the real meaning, to the concept of fashion spread out, so that people wear their own body. ”

The following is the original song Wei Long speech:

Hello everyone, first introduce myself my name is Song Weirong, is one of the stars wardrobe co-founder. Today, we share the theme of the mobile era with large data to enhance the experience of fashion customers. Our side is mobile special, I hope this can be more from the mobile side to do some sharing. The big data is a category that we started 3 years ago and how to make big data work in fashion shopping is the whole that I'm going to share with you today.

How to find 1 billion SKU for apparel products?

We put forward the first question is the apparel products 1 billion SKU, how to find? With the city more and more big, inside SKU clothing category of goods more and more big, how to let better SKU products better let everyone search, this is a big problem. Take Alibaba, Taobao just launched a through train with Diamond booth, investment is not very high, do not need too much money, now want to buy a user to enter their own shop to spend money, need to spend a large amount of cash, in such a huge SKU system, how to dig out better products. Our approach is to use the big fashion data, establish the relationship between the goods, to recommend to users accurate, like, the need for apparel products. Why emphasis is the big fashion data? Here are a few stories, 2011 we just try to do the star wardrobe products, we have done a lot of the role of the class, do a lot of UGC content to do, we ask many users come up to put a different clothing position, spelling into a match to guide the user. Later we have a user is a fashion veteran girl, how do you guarantee that your clothing is fashionable, we like it?

Big Data is a core point, is to do collocation, how to learn collocation.

Behind us from the stars, learning to wear the stars, we believe that behind the stars are image designers, they wear what we think is fashionable, we have a large number of star elements on the map. Finally, at the end of 2012, this data was recognized by everyone, and the reason for recognition is that we analyzed it through data. This is also the way we do analytic diagrams, which can be visually described, there are a lot of stars in the wardrobe, this star map how we choose to do, as we have just said, we add a large number of star elements, there are a large number of star map, are all body, most of the body. These whole body star charts, we disassemble it, for example, this is the picture of sun, we will be dismantled, such as her blouse, skirts, shoes, stockings, we go through the four distribution recommended users this four, the user put up we do not dare to ensure that the stars wear out is the same effect, but can learn her feelings, the essence. Why other people like this, we do not need to understand today why this collocation, but we know that this is with the fashion is detached, OK. This is our big data in a core point, is to do collocation, how to learn collocation.

There are a lot of small red dots, green, this is our separate points, each part of the problem of elements to make up, put on the clothes what it looks like, we find similar sections according to the elements, these are done through the machine.

This is a product we are doing, this product we will let users start with a commodity, for example, there is a girl, in the middle of a product, in the fashion search application, can search what kind of goods, can produce dozens of of pictures, hundreds of sets of collocation, this is the star map, what others wear is like, According to this to find her favorite star, or star map, I think this girl dressed well, I like. Point in, you can see the girl's hat, bag, shorts, shoes. This solves a very large proportion of users, about the needs of clothing matching, how to go with, how to match. In this search engine, you can see other people blockbuster existing clothes. For example, I want to buy this dress in the middle, and I can see how it matches. Do you really like people to mix and match? Users may not like the views of others, after two years of application data collection, the user is very very recognized, we are also trying to create a demand, try to buy the user to choose the way he likes to buy goods.

Back to just the 1 billion SKU, to do their own fashion engine, to their own point of view, their own needs, to find their own goods, 1 billion SKU display volume, exposure, exposure to the opportunity far more than the search out of the same section of the garment is much larger.

Spread the concept of fashion with big data

Just now when it comes to big data, one is a big data source, and if the source chooses the wrong big data analysis it will become meaningless. Just now we have mentioned, why we should pay attention to fashion, pyramid tip is the source of fashion. And we are now from the bottom of the pyramid to the source, to go up to guide. Fashion from the top down Guide everyone shopping guide, since the spread of the downward, we can learn how stars, fashion talent is how to wear. Such large data we have to get the real meaning, to the concept of fashion spread out, so that people wear their own body. Then to these fashion large data collection, learning rules, every star is our learning rules. Like this rule our database has accumulated 70多万条, enough to let everyone search their own collocation.

The second is big data accumulation, and there's nothing special about the way we accumulate I am doing technology, this circle everyone will use this, data accumulation, manual training, accumulation of learning, we realize is automatic matching, we do with a small amount of manual browsing audit, you can push the large data content to the line, For your reference.

The last one is a large data processing method, in order to achieve better for the user to do with the recommendation, we introduced a depth of learning, is actually a branch of machine learning, everyone will simulate the logic of the human brain, thinking, neural way to concept concepts. For example, in-depth study of image scanning to learn, so that the machine to guide the public to spend more intelligently. Another is natural language processing in the fashion Search Center application. For example, I need a dress that looks loose and we'll explain it to you as a cake skirt.

We have many predecessors are also later, there are many things to learn, there is a long way to go. This is simply to say, this is our current entire backstage processing logic. One is our data source, which is a very important part of a single product data sources, part of the cooperation from the electric business, for example, now the largest proportion of the entire electrical quotient is Taobao, cat, Jing Dong, Fank, etc. are also in the gradual development of the big electric business.

The application of big fashion data in Taobao

Below the rough data, the data source comes out, our simple one screening, will become the fine density, the structured commodity data, is used to provide the user with the collocation. Talking about a guy who's got a lot of machines to do things. In fact, we attach more importance to machine learning in the whole process, more attention to machine recommendations, we are not ignoring the UGC is not. In the first half of this year, we have done a lot of things to bring up the talent, we think that in the domestic real UGC content is less quality, deviation, as far as possible to give everyone the right guidance.

This is our big fashion data in the application of Taobao, this is the trend of women and fashion mister, there is a beauty in the middle, there is a cool ride, the two data provided by the graph is our data provided in the past. We do the method of Taobao single product data processing in the feedback to Taobao. We will feed back to Taobao, Taobao users to see again, this very obviously improve the conversion rate and the number of passengers. At the beginning of our own statistical data, the same user in this mode, the purchase of more than 2 items of clothing is very high probability. A small percentage of users, they are shopping very fast, directly to all the matching items, five full income bag, wearing the same as the star map, this result proves that our way is to reach the theme of the operation, how to the current low conversion rate of the electricity business how to break the bureau.

The last one to share our fashion search output results. Through a large number of user behavior analysis, the study of fashion data, we can accurately grasp the style of this user, he likes what kind of challenges, like what kind of clothes, we pass large data to different customers, each has its own different style.

Behind the challenging recommendations is also one of the things we often say, because all the recommendations are for you to push the things you like, as clothing, as a match, as a style, we encourage you to try to change the style, so that you can quickly enter the fashion reform.

That's all I need to share with you today.