Author: Qiu Canqing (Bai Li)

On October 12, 2018, Alibaba Group announced the establishment of Alibaba Local Life Service Company. Ele. me and Koubei merged to form a leading local life service platform in China, with the mission to “redefine urban life and make life better”. Koubei focuses on in-store consumption service, Ele. me focuses on home life service, Hummingbird focuses on instant delivery service, Keruyun focuses on providing digital and upgraded products and services for merchants, jointly promoting the digitalization of local life market, so that there is no difficult business in the world.

Local life is a very important business map of Alibaba. Based on the business development scenario of Ele. me APP of Alibaba Local Life, this paper describes how to give full play to the advantages of end intelligence in end-to-end computing and decision-making in the process of business development of Ele. me APP according to the actual business scenarios and by combining technology with business.

End intelligence concept

background

With intelligent simply is to do in the end side of machine learning or reasoning and upper application of deep learning model, thanks to the development of hardware, such as mobile phone, at present, the strong CPU and GPU can support part of the higher arithmetic operations on the client (under performance requirements can sacrifice accuracy), can keep the computational work on the client side, This avoids uplink bandwidth overhead and network latency.

Whether in the cloud or in the terminal, intelligence solves the same problem. How to make continuous learning by taking a large amount of data as the input source and adjusting the algorithm strategy to deduce the final or the most appropriate conclusion? Then the end intelligence is to realize the intelligence on the end side, which can carry out engineering standardization of the intelligence of algorithm data. Combined with specific business scenarios, the end side can carry out real-time feedback through the real-time collection of behavioral information and equipment information on the end side, and run artificial intelligence, machine learning and other end algorithms. Adjust the algorithm, operation and product strategy to improve the user’s purchase, experience and other business technical indicators.

features

Compared with cloud-side intelligence, end-side intelligence has significant advantages such as low latency, data privacy protection and cloud computing resource saving. Specific segmentation points are as follows:

trend

In the last two years, end intelligence is a hot research and development direction of major Internet companies. Many end intelligence applications have been used in our daily life, such as AI camera for taking photos on mobile phones, FaceID for unlocking faces, and VARIOUS AR effects in short video apps, etc. Inside Alibaba, there have been a lot of business exploration, such as mobile Taobao Pai Li Tao, AR makeup test, product recommendation, etc., are typical representatives of the effective combination of end intelligence and business.

As a part of Alibaba’s business map, local life cannot be absent. Based on the brief introduction of the concepts related to end intelligence in the previous chapter, the following is a description of local life scenes.

Local life scene

Business attribute

According to reliable data, the overall scale of local life service is roughly estimated to be about 20 trillion yuan, which is a very large market. However, the digital penetration rate of this market is still relatively low, only about 10%-15%. The industry attribute of local life is very different from that of traditional e-commerce. Localization and real-time are the salient features of local life service. In essence, local life is actually a service industry, in which the difference of craftsmanship is very large. At the same time, there are a lot of subcategories in local life, resulting in the service of local life is a non-standard product, which is in sharp contrast to the characteristics of standard and easy digitalization of goods of traditional e-commerce.

Ele. me focuses on home service in alibaba’s local life field. Intelligence on the user side is the only way to better match the demand side with the supply side. Business attribute based on home service, the user to make a decision time is very short, basically peak is in the afternoon and evening meal time, in the limited time period, how to analyze the user’s current needs, and the demand of users at the moment and we supply a better match, with intelligent can play a good role in this field.

The technical architecture

Local life side includes ele. me, word-of-mouth, Hummingbird, business edition, Regulus and other apps, there will be different application scenarios of end intelligence in different apps; Based on the consistency of local life technology architecture, we abstract a layer of local life end intelligence adaptation layer, and make customized development based on the business attributes of local life. At the same time, combined with the rich middleware and platform in the group, we quickly carry out the iterative development of end intelligence related applications.

At present, the architecture of local living intelligent technology is as follows:

Based on the general group middleware and end intelligence SDK, the end foundation and end intelligence foundation are abstracting. The end foundation includes business burying point library UT, switch configuration Orange, high-speed data transmission Highway, real-time monitoring Answer, database operation SQL, network library Mtop, etc. The foundation of end-to-end intelligence includes MNN, end-to-end user behavior data BehaviX, and Walle, an end-to-end intelligence operating environment.

Aiming at the technical and business characteristics of local life, ALSCAdapter of local life is designed. In the adaptation layer, we added many features needed by business, such as application life cycle, automation burying point, etc. At the same time, combining the monitoring system of local life to create intelligent real-time monitoring system. API was opened for different businesses. In order to solve the problem of low efficiency in development, debugging and testing, we developed an end-to-end Debug platform for local life. Meanwhile, we co-built the PLATFORM with MNN workbench of the group to improve the efficiency of algorithm and testing students.

Technology to explore

Based on the terminal intelligence technology architecture of unified local life, we can better and faster develop iterations in business, and also help other apps in local life to quickly access and get started. On ele. me APP, we have made some technical explorations in user characteristics, intelligent recommendation, intelligent touch and other fields, and achieved certain business effects. The following are introductions one by one.

User characteristics

Richer and more real-time user characteristics are the key to improve the accuracy of algorithm decision-making. Based on the ability of end intelligence, the key behavioral data of users are stored on the end, such as the user’s data of entry and exit, entry and exit, purchase order data, browsing behavior data, etc., and very rich metadata can be collected on the end side.

Based on metadata, real-time user behavior change line feature extracting, real-time maintenance user behavior characteristics, latest after reflow igraph delay around 2 s, and make the algorithm of the cloud model can obtain in advance to the user’s behavior characteristics, the characteristics of the data at the same time local persistence, the other end model also can directly obtain relevant characteristic data in the end, do a extraction, use. In order to standardize the process, we specify the production rules of the end feature and divide the function of the data table of the end feature to improve the efficiency of obtaining local features.

At present, the overall process of feature extraction on the terminal is as follows:

By continuously expanding the types of features, the feature center on the end is established to provide feature metadata for the subsequent end-to-end model calculation and cloud model promotion, so that the business algorithm can focus on the implementation of business details without extracting their own features for each business, saving development time and avoiding the waste of resources on the end.

Intelligent recommendation

Search and recommendation is a business field in which end intelligence can play an important role. For recommendation feeds stream, there are 20-30 store data on one page due to pagination loading, and the algorithm can only intervene when the user browses to the next page. Client trigger page request when the cloud recommendation algorithm can make decisions intervention, the result of the recall of the whole process is very long, the user – > client – > project – > algorithm, algorithm level decision depend on the data return link delay in minutes, completely unable to meet the needs of real-time recommended, recommended the results may be associated with the order before the user not big.

In ele. me APP, since users have not been using feeds for a long time, the algorithm can make multiple decisions when users browse feeds and provide the optimal solution for users.

Recommend rearrangement scheme based on intelligent, can browse the shop for the user for which the strategy rearrangement, weighted according to the preamble of the user behavior adjustment, business shows will not interfere with the user’s use, and provides multiple contacts to do trigger algorithm, makes the algorithm model can real-time intervention on end, achieve algorithm; The overall recommendation rearrangement logic is as follows:

With the algorithm, the timing of rearrangement triggers is agreed, and some boundary triggers are eliminated, such as the number of rearrangements to be small, etc. The algorithm model uses the features on the cloud and on the end to make decisions together. AB experiments are carried out through models of different business objectives, such as GMV target, IPV target, etc. At present, there are apps in the home page recommended feeds of Ele. me and food Channel feeds, which have achieved good business effects. At the same time, in order to expand the number of candidate pools and give more decision-making space to the algorithms on the end, we also made an intelligent update on the end to replenish store candidate pools in time. Interactive recommendations and smart weathervanes are already being planned and developed.

Smart touch of

In the field of user growth, how to reach users efficiently and accurately is the key to achieve user growth. Push is the channel that can reach the most users, whether inside or outside the end. Before the smart touch is proposed, only batch push and ring push can be configured for users by the operation. The push configuration in this way is too rigid, resulting in a very low click-through rate of the overall push, and many users will be disturbed, increasing the probability of users closing the push permission. Operation students need more accurate timing and richer ways to reach business goals.

Intelligent touch based on the end intelligence, through the deployment end model on the client side, intelligent identification of the touch point that may be the inflection point of user behavior through business rules and operation configuration, combined with the characteristics of the end cloud, interaction with the push center, cloud algorithm racing mechanism to obtain the optimal push configuration, and finally reach the user. Based on the intelligent touch inside the terminal, since it recognizes the inflection point of users’ behaviors, the large probability of push at this time will not disturb users, but will become an assistant to promote users’ ordering transformation. Based on the out-end intelligent contact, it mainly collects the characteristics of in-end behaviors of users, reports them in real time, and identifies the intention of users to do back-end recall. The overall intelligent touch scheme is as follows:

Through the production of intelligent contact points, the operation students have more and better opportunities to reach users. Combined with the customized configuration of the operation platform, the contact effect of thousands of faces can be produced. Through the launch of smart touch, the click-through rate of push in the end has been greatly improved, and the withdrawal recall also has a very good user retention effect, helping the business operation to achieve the target of user growth.

Other business

In addition to the business technology to explore, we also try the other end of the intelligence, such as hummingbirds blue storm project, carry on preliminary identification by conducting project, to help improve the rider photo experience, in the rider in the sampling process saves the rider’s operation time, also reduced the cost of artificial audit at the same time, improve the service quality of hungry? And the brand image, Ensure safety and brand image during delivery.

future

Mobile equipment performance constantly strengthened in the future, the network infrastructure, improve the environment, with intelligent can play space will be more and more big, in the field of local life, advance the search recommendation real-time and operation of intelligent technology development direction is the end, thank you in the group with intelligent technology team to give support and help, Thanks also to the close cooperation of local living product technology algorithm and other related parties, so that the end intelligence in the local life scene continues to bloom.

The future, we will also need to end the ability of intelligent precipitation for better infrastructure, help the business fast access, let algorithm experiment more flexible and more rapid data collection, and continuously explore new business scenarios, the resistance and long, need to be more willing to participate in local kindred spirit, in the field of life work together to grow together, together with technical power business shine! (PS: You are welcome to join us if you are interested in local life. Your resume can be sent to [email protected])

Pay attention to [Alibaba mobile technology] wechat public number, every week 3 mobile technology practice & dry goods to give you thinking!