Shulou(Shulou.com)06/03 Report--

As we all know, online merchants can easily obtain a great deal of basic information such as purchase rate, collection rate, conversion rate, turnover and so on through a large number of front-end "burial points" of Taobao platform, even at a more fine level. obtain the data of the changes in the flow of commodity keywords or new / popular styles, and even the traffic structure of the competing products. In order to constantly adjust their own commodity structure and business strategy.

So how to make physical merchants have the same data operation ability as Taobao? Next, let's take a large-scale complex as an example to see how we can use visual intelligence engine and data decision engine to build a business "intelligent sand table" system in order to achieve our great mission of making business smarter. Thus help offline merchants in the "intelligent sand table" through data analysis to deduce calculation, plan strategies, win thousands of miles away.

1. How does the data come from?

For online merchants, their data comes from a large number of front-end "buried sites" of e-commerce platforms. And how should the data of offline complex come from? Through our self-developed AIoT+ cloud capabilities, we create an AI+IoT+ cloud environment to complete the data characterization and collection of "people" and "fields" of the complex. Through the previous technical introduction (Face ID, Re-ID, identification, etc.), we have a powerful visual intelligence engine, and the visual engine combined with cloud computing platform to build a set of AIoT+ cloud data collection platform.

For the data collection, through the IoT intelligent terminal + AI algorithm, the people are portrayed and the field is portrayed. Among them, IoT intelligent terminal includes singularity recognition, singularity magic box and so on.

Through our visual intelligence engine, we can abstract attribute portraits, behavior portraits, relationship networks and so on. Attribute portraits, including basic gender, age, height, etc., more complex attributes include dress, appearance, occupation, such as hairstyle, face shape, lip gloss, coat style, underwear style, necklace, satchel type and so on. Behavior portraits include entrance position (parking lot or gate), on-site consumption, on-site preference area, on-site movement trajectory, whether there is dangerous behavior, and so on. The relationship network includes intimate relationship, kinship, peer relationship and so on.

(photo source network)

Portraits of the field, including regional portraits and shop portraits. Among them, regional portrait refers to the analysis of regional heat, regional flow of people, regional value and other aspects of a specific region in multiple dimensions of time and space. The shop portrait, by combining the portrait of people, describes the popularity of a store, the consumer group of the store, the conversion rate of entering the store, the repurchase rate of the store, the attributes of the store industry and so on.

2. How to deal with the data

In a large number of IOT intelligent terminals, a large number of video and image data are produced every day, so in the face of such a large amount of data, how can our system achieve multi-algorithm cascading, timely response, fast processing and stable operation? Here, we put forward the video structured technology, which relies on the powerful cloud computing platform of AI algorithm, intelligent devices and singularity cloud, thus forming the intelligent data processing platform of our AIoT+ cloud.

Here, let's make a self-thinking definition of structured technology:

Video structuring is a technology of structural extraction based on video content (Content-based). According to semantic relations, video content adopts techniques such as spatial quantization, time quantization, target recognition, target tracking, feature extraction, image processing, coding and decoding to produce high-dimensional data with low storage, high value, easy to analyze and understandable.

The structure of video is based on the structured processing of video content, the so-called video content is the understanding of the semantics in the video, the so-called semantics simply means what is in the video and what is happening in the video, which can be accurately described by our AI algorithm. In addition, time quantization and spatial quantization can more accurately describe when, where, what is in the video and what is happening. "

For example:

"there is a camera, and through the space calibration tool provided by our system, we can complete the mapping between the camera image and the CAD image, that is, every position in the camera can be accurately projected onto the CAD image. There are shop An and advertising space B in the coverage area of the camera, and dozens of customers pass by the camera in a certain period of time. Then we take the individual as the dimension, through Re-ID, Face ID, attribute, tracking and other algorithms, we can produce the following relatively simple structured data:

(click on the picture to enlarge)

Taking the store as the dimension, through pedestrian detection, Re-ID identification and weight removal, target tracking analysis and other technologies, we can produce the following relatively simple structured data, and through these simple data we can quickly calculate the store conversion rate, through the stay time can initially calculate the consumption conversion rate, major consumer groups and other data.

(click on the picture to enlarge)

Taking the advertising position as the dimension, we can generate data such as the conversion rate of advertising space, advertising attraction groups and so on. Taking the camera shooting area as the dimension, we can analyze the basic data such as regional heat, regional flow of people and so on. "

From the above example, we can see that through the rough processing of IoT intelligent terminal + AI algorithm (we define it as first-level structure), we initially extract more granular information from massive video data, thus greatly reducing storage and network bandwidth transmission.

For the first-level structure, which is mainly processed on the end, we need to save as much video content as possible, while taking up as little storage and network bandwidth as possible. The first-level structure completes multi-model cooperation through intelligent hardware, quickly completes the rough processing of video content, and then uploads the first-level structured data to the cloud for fine secondary structured processing.

Two-level structure, relying on strong cloud computing capabilities, to complete more refined data processing and analysis on the cloud. In the two-level structured system is equipped with our intelligent decision engine, through the analysis and mining of big data and the precipitation of industry prior knowledge, so as to produce high-value, easy-to-analyze, understandable digital information.

The whole data processing platform has the characteristics of hierarchical processing, multi-level linkage, dynamic node allocation, multi-node load balancing and so on, so as to ensure that the whole system can be stable and fast data processing.

The system architecture of the data collection system is shown in the following figure:

3. How to apply the data

If the above groundwork mainly introduces the visual intelligence engine, which is the eye of the platform, then we will use some examples to introduce how the intelligent brain of the platform, the data decision engine, allows merchants to perform deduction and calculation under "what the eyes see." so that merchants can be uninformed, discerning, and overall planning.

3.1 basic report

Through the analysis of customers, shopping malls, regions, time periods, shopping malls and shops and other dimensions, thousands of dimensional data reports are generated every day, including simple shopping mall passenger flow, shop passenger flow, male-to-female ratio, popular stores, regional heat and so on. Through these data reports, we can quickly see the recent operation of the mall, as well as the focus of attention.

3.2 layout of shopping malls

On the e-commerce platform, stores need popular styles for drainage, and after entering the store, customers are guided to shop deeply through strategies such as recommendation and combined sales. Offline is the same, we hope that customers come to the mall is not to buy something to go, but to be able to go shopping in depth to eat, after all, let the flow to produce its rich value.

Through our "intelligent sand table" system, we have obtained the digital store data and learned the basic data of each store, such as out-of-store flow, entry conversion rate, consumption conversion rate and so on. through these data, we can know which stores are the "drainage stores" and which are the "drain stores" of the shopping malls. Then through our higher-level mining and analysis of the data, we can find that there is a relationship between the store and the store. For example, 60% of people have been to store An and will go to store B and store C.

Well, for the drainage shop, we can put it a little deeper into the mall, so that it will attract more passengers to flow to the mall. At the same time, there are strongly related stores, so that they are at a certain distance, so that on the way from store A to store B or store C, they must pass through other stores, so that customers can flow better, resulting in other shopping conversion rates.

When the shopping mall is laid out according to the data analysis, our "sand table" will soon be able to feedback its positive and negative effects. Through further carding and analysis of the data, we can further fine-tune and optimize the structure of the whole shopping mall.

3.3 Smart recommendation

Online, we are often able to receive product recommendations close to our needs because of the recommendation system of "thousands of people and thousands of noodles". So how can we recommend intelligent play in brick-and-mortar stores?

Let's start with a combination of offline and online recommendation, which is based on the premise that the recommended object is the registered members of the mall. Through the above example of data collection, we can learn that through the AIoT intelligent terminal, we can get data such as where members go, whether they produce shopping, repurchase rate and so on. So if we find that member An often produces consumption in store A, then when there is a promotion or activity in that store, we can recommend or remind the member through SMS or online App.

The other is direct offline recommendation, which is based on the premise that there are multiple electronic advertising spaces in the mall. When a customer stays in the advertising space, we can quickly recommend the corresponding products to her according to the portraits of her users in the mall, so as to achieve the same recommendation effect as the online "thousand people with thousands of faces".

3.4 advertisement delivery

For the merchants in the mall, it is necessary to further publicize or attract customers to spend money in the store, then the merchants will place advertisements in other parts of the mall to attract passenger flow. So where do you put the ads? How much do you charge for advertising space?

For advertising, we can put in more points, and then we can quickly analyze the conversion rate generated by changing the advertising space through the system. For example, if customer A stops next to advertising space An and finally enters the store where the advertisement is placed, there will be a transformation. Then through the analysis of the data for a period of time, we can analyze the conversion rate of each advertising point, so as to select the advertising sites with high conversion rate.

For shopping malls, how much do you charge for advertising space? Shopping malls can charge advertising space pricing according to the number of passengers, regional heat, regional value and the historical conversion rate of the advertising space.

3.5 trend Forecast

Through the capabilities of AIoT+ cloud, we further quantify the data from time, space and other dimensions. Then through the historical data analysis, we can basically analyze the flow changes in the next period of time, the next stage of user consumption habits, so that we optimize our business strategy, so as to optimize the value generated by customers.

Above, through some simple examples, the "intelligent sand table" lies in how to assist the strategic layout of shopping malls, optimize management, predict the future and other aspects of analysis, but the value of the data is far more than these. It is believed that more and more refined data analysis will generate more and more data value, and offline merchants can operate their own stores as flexibly as online merchants.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.