Abstract: Ant Financial product and service technology expert Fu Haitao made a wonderful sharing on the theme of intelligent Road of A New generation of Mobile R&D Platform mPaaS in the financial special session of Alibaba Cloud Summit.
On July 6-7, Ant Financial mPaaS team Fu Haitao, Luo Qiping and Lu Dan will be lecturers at the 43rd MPD Workshop in Beijing. For more details, please click here.
Fu Haitao, technical expert of Ant Financial
Outline:
Development history of mobile development platform mPaaS
Mobile development platform mPaaS 3.0 product system
The evolution of Alibaba’s financial business
MPaaS integrated mobile intelligence scenario
I. Development history of mobile development platform mPaaS
In December 2016, mPaaS released version 1.0, which was mainly intended to continue the financial properties of Alipay and serve the financial industry. At that time, the mPaaS team conducted in-depth communication with many financial institutions, and we found that most of them had developed their own apps. However, the difficulty was not App development, but how to solve the performance problems of App and improve user experience. Therefore, mPaaS 1.0 gives priority to opening the underlying development framework, UI library, message push, gateway service and mobile analysis capability of Alipay, and provides services in a component-based way, so that users can choose components suitable for their own needs and quickly build App infrastructure and general capabilities like building blocks.
As we get deeper into the financial industry, we find that some leading financial institutions are gradually maturing and entering the stage of digital transformation, hoping to carry out refined operations for customers. During this period, Chongqing Rural Commercial Bank put forward the concept of “smart bank”, focusing on the construction of data collection and analysis platform. At the same time, due to the rise of Internet finance, the pace of product research and development, release and update of financial institutions is more and more like that of Internet companies. They hope to be able to rapidly expand and update, and carry out dynamic update in response to emergencies. Therefore, mPaaS 2.0 gradually opens up the release platform, hot repair, offline package, data synchronization, custom analysis and other capabilities, further changes the mode of enterprise mobile development, helps enterprises to do digital transformation, and creates dynamic super App.
Over time, financial institutions have gained a deeper understanding of users and demanded more from technology. In order to use data more effectively and improve the ROI of operations, apps need to become smarter. In addition, as a hot spot in the technology circle in 2018, small programs have also attracted the attention of the financial industry, and financial companies generally choose small programs as a sharp weapon to seize the market. Therefore, Ant Financial takes the small program framework out and outputs it as a product. Financial institutions can build their App ecosystem based on this.
Second, mobile development platform mPaaS 3.0 product system
After three years of intensive work, mPaaS has not only accumulated hundreds of paying users, but also greatly enriched its product system. MPaaS product system is mainly divided into three layers:
First, dynamic and flexible front-end capabilities. At present, mPaaS can provide three development frameworks: Native, H5 and Alipay small program. 100+ UI controls; As well as 20+ functional SDK including scan code, local cache, client burying point, etc., developers can quickly access the basic capabilities needed to build App.
Second, it is a solid mobile platform capability. In addition to client development, mPaaS also provides mobile medium platform capability to manage the entire life cycle of App, including App development, testing, release, analysis, operation and other links.
Finally, it is the ability of stable background connection. MPaaS provides customers with mobile gateway and large file channel to serve different scenarios, and provides users with a high stability, high reliability and high efficiency of background connection service capability for APP development.
Third, the evolution of Alibaba’s financial business
Similar to the development of Alibaba’s financial business, mPaaS 1.0 mainly helps to improve the compatibility and stability of financial apps, emphasizing service availability. Next, mPaaS 2.0 advocates refined operation, using data management services to establish a digital system within the system and realize the big data platform. So how to use data to achieve fine, intelligent operation, how to complete personalized decisions and recommendations for different users, mPaaS 3.0 to achieve intelligent platform to support decision-making. “Intelligent upgrade” is the natural transition after two iterations and upgrade of mPaaS, which is the result of market development and customer demand drive.
Data access + Analysis and decision engine +mPaaS scenario
As for the intelligent platform, mPaaS mainly focuses on constructing an integrated mobile AI scheme of “data access + analysis and decision engine +mPaaS scenario”. Ant Financial’s core AI technology, as well as the technology applied in the construction of “Thousand people and Thousand Faces”, has been stripped out and formed a decision engine. In the two directions of “operation” and “experience”, combined with mPaaS perfect business application scenarios, the output of mobile analysis, intelligent delivery, intelligent prediction, OCR recognition and other integrated vertical solutions, so that users can really enjoy artificial intelligence services that can be applied to the ground.
Data: Built-in data in standard format
MPaaS 2.0 has implemented a complete set of data acquisition mechanisms for data-based transformation, including model environment, user behavior, flash-back lag, component usage and custom events. Based on these data, the intelligent prediction model can be predicted.
Iv. MPaaS integrated mobile intelligence scenario
MPaaS provides management of the whole life cycle of App development, testing, release, analysis and operation, and naturally provides many intelligent application scenarios.
By using the AI technology deposited by Ant Financial and the business data collected by mPaaS, we can dynamically create crowd classification according to users’ behaviors, which is the intelligent prediction product. Intelligent prediction can also be combined with grayscale publishing, message push, intelligent delivery and other products to provide customized operation activities for people with the same behavior, improve retention and promote transformation.
Intelligent prediction technology model
Predicting whether a behavior will occur is essentially a supervised learning model. We extracted the data of the last 28 days, marked the data of the last week, and classified the data for the users. The data of the remaining 21 days were used to generate the feature sequence, and then all the data were sent to the machine learning platform for model training. Predict user behavior in the coming week to form a crowd.
In the training process, recall rate, specificity, accuracy and other key indicators will be used to evaluate the prediction accuracy of the model. Of course, the prediction model has different risk tolerance in different scenarios. Intelligent prediction has three levels of “low confidence, medium confidence and high confidence”. The higher the confidence, the lower the misjudgment rate, but the fewer users the model can target.
For the scenario of “financial recommendation”, we can choose “low confidence” as the standard to circle people, because even if users do not want to buy financial products, the push of marketing information is still acceptable. Instead, we push coupons to even lost users to increase retention, and “high trust” is the best choice.
Smart forecast has two prediction tasks built in, one is “users will be active in 7 days”, the other is “users will lose in 7 days”. At the same time, the product supports the setting of “custom event”. We can customize different App experience for different groups by combining grayscale publishing, and also carry out targeted marketing push by combining message push. Even if we are not sure which marketing strategy is the best choice, combining ABTest can be used to conduct in-depth testing for the same group of people.
ABTest technology model
Through ABTest, we can know what users like and don’t like, so as to provide more data support for App experience optimization. As shown in the figure, Alipay provides different interface styles for different users, thus helping the product team find the optimal interaction scheme more directly.
ABTest can not only provide support for client experience optimization, but also participate in server algorithm and strategy experiments. Combined with mobile Gateway Services (MGS), ABTest can easily support back-end algorithms and policy experiments. Combined with the mobile Analytics Service (MAS), ABTest helps customers make the right decisions based on data on user attributes and behaviors.
Intelligent delivery technology model
Intelligent products can be in accordance with user attributes, the actual needs of the real thousands of thousands of faces, targeted advertising. Intelligent materials, intelligent circle people, intelligent recommendation and intelligent monitoring are mature application intelligent modules in Alipay at present:
Intelligent material module through intelligent algorithm to copy, picture assembly and rendering to the user, to solve the problem of single content, lack of strategy;
The intelligent circle people module solved the problem of difficult target group classification by model prediction of specific events and target group delineation of seed user portraits.
Intelligent recommendation module can sort the content, control the advertising display, control the fatigue of advertising display; In view of the characteristics of banking and other financial businesses, common recommendation algorithms such as LR,MAB and GBDT have been integrated into the engine. Combined with the unified data collection and standard processing process of THE SDK of mPaaS client, customers can realize the intelligent recommendation of basic marketing content without algorithm engineers.
The intelligent monitoring module can provide early warning in combination with data analysis to reduce operational risk of launch activities.
At the same time, ant Financial is constantly pursuing the goal of lightweight AI. AR red envelope is a popular game during the Spring Festival in recent years. 70% of the scanning and recognition tasks are performed on the client side, while less than 30% of the tasks are performed on the server side. The main reason is that the ant can generate the client recognition module through the background training model, and can directly complete most of the recognition on the client.
Based on the specific practice of AR red envelope, mPaaS launched a lightweight client intelligent solution. Mobile analysis service (MAS) in mPaaS provides client data collection capability, and the underlying intelligent platform contains AI model and decision-making capability matching MAS, so mPaaS can make accurate prediction based on its own data. And for the groups that may have similar behavior, gray publishing, message push, intelligent marketing, ABTest and other operational means.
In addition to the content in the article, I will share with luo Qiping and Lu Dan, two other members of the mPaaS team, in depth for three hours at the 43rd MPD Workshop in Beijing on July 6-7. Welcome to live communication. The topics of the architecture workshop are as follows:
In addition, the workshop we invite to ali, baidu, tencent, sina, where to go, drops, netease, VIPKID companies such as a line of experts and technical Daniel, they will be sent to you by speech, speaking, group discussion, group between PK, landing process implementation theory, can make participants from speech to excavate, summarizes the ascension.