eclicktech CTO Wang Yizhou: Federated Learning + Image AI + Public Cloud Helps the Revolution of Digital Advertising Accuracy


The sudden attack of the new crown epidemic has shrouded the global economy with great uncertainty, but the "new infrastructure" represented by 5G, deep application of the Internet, big data, and artificial intelligence has revealed its tenacious vitality different from traditional industries. Become a new engine of economic development.

According to media reports, advertising and marketing companies, as intermediary service providers, have always been known as the "barometer" of the national economy. They are also weak in business operations, especially offline advertising companies. But on the contrary, the digital marketing market on the main "battlefield" online has been less affected by the epidemic. In particular, digital marketing companies whose main customer groups come from industries such as short videos and games have increased their revenues instead of falling.

How can "new infrastructure" integrate with the digital marketing industry to produce a greater boost to promote the internationalization of Chinese companies?

In the topic of "Cloud Infrastructure" at the 2020 Alibaba Cloud Online Summit “Continuously Helping Internet Companies Go Global”, eclicktechCTO Wang Yizhou shared the technical challenges and successful experience faced by digital marketing companies in the process of globalization services based on the cooperation process between the company and Alibaba Cloud.

He mentioned: "Currently, there is still much room for improvement in the accuracy of digital advertising. The addition of public cloud not only saves a lot of costs for enterprises, but also allows massive data storage, flexible calling, and mining efficiency to maximize data value. With the improvement, we also have more energy to improve the analysis model and help more companies accurately reach potential users around the world."


1. Globalized service brings technical challenges of globalized digital asset management

eclicktech is an enterprise international intelligent marketing service provider. It provides customers with global marketing promotion services. Through integrated services such as effect marketing, brand building, and vertical industry solutions, it helps companies efficiently acquire users and enhance brand awareness on a global scale. , Realize commercial realization.

As the world's leading third-party independent advertising service provider, eclicktech has a wealth of global media resources. Platform data covers more than 200 countries and regions around the world, and currently serves more than 2,000 customers, including well-known companies such as Alibaba, Tencent, NetEase, Bytedance, Kuaishou, and iQiyi.

The characteristics of globalized business have brought three major technical challenges to eclicktech:

First of all, facing the relevance and similarity of the needs of the same advertiser in different global markets and different advertisers in the same global market, eclicktech must integrate upstream and downstream data to meet its marketing needs. At the same time, the service process must also maintain consistency, and the difference in service cannot be caused by market differences.

In addition, native data born in different regions of the world needs to complete cross-region data collection and management in the unified data governance process, which puts forward higher requirements for multiple links such as network and storage. Wang Yizhou mentioned: “The global network users have exceeded 4 billion. During the peak period, we have to withstand millions of business requests per second. This requires very high concurrent processing capabilities for each data center and node. At the same time, users in different countries around the world Because time zone and behavior differences have led to staggered peaks and valleys, when data demand is low, it is necessary to scale down in time to control costs."

Third, data-driven business is not aimed at data collection, but aimed at computing, feeding back business decisions, and truly helping advertisers acquire users at a lower cost. Therefore, under the huge data scale, how to effectively implement offline computing and provide auxiliary support to the business is also a very important link.

2. Activate Alibaba Cloud Global Data Center to complete infrastructure deployment

In order to achieve the consistency of global services, eclicktech must complete localized deployment in different countries and regions in terms of technology. At the same time, the selection of data centers must also meet two major requirements at the same time:

For product and operation and maintenance requirements, eclicktech requires data centers in different regions to maintain the same capabilities of cloud-native services, and ensure that business resources are sufficiently stable, and network quality is efficiently radiated to surrounding users.

Due to the requirements of business indicators, such as millisecond-level attribution statistics and minute-level real-time reports in attribution scenarios, eclicktech has very strict requirements on the reliability of data communication between dozens of technology centers launched.

In these two aspects, Alibaba Cloud satisfies the actual needs of eclicktech very well. In the process of cooperation, eclicktech has almost enabled all Alibaba Cloud data centers around the world to complete the deployment of global infrastructure and meet data requirements.

3. Data placing is only the easiest step, the focus is on how to use it

Because advertisers are distributed in different industries and countries, and have different goals, such as e-commerce's goal of placing orders and game customers' goal of installation, how to manage and analyze different log data is very important.

Under different scenarios, it is necessary to support the operation team to make aggregated reports in different time dimensions, e-commerce remarketing scenarios to update user bidding lists and bid prediction models in real time, and to analyze user relevance in new application scenarios . Building different services in different scenarios will also result in data redundancy and waste of resources.

eclicktech builds a global unified data computing center based on Alibaba Cloud, with data multi-party access, unified aggregation, and centralized computing.

After the multi-source hybrid computing based on Alibaba Cloud DLA+OSS is enabled, the operation and storage separation of most data computing scenarios is first realized, most of the original logs are placed on OSS, and different cloud-native services or services are used behind the scenes. Self-built services to retrieve these original logs can reduce storage costs on the one hand, and improve data consistency and effectiveness on the other.


In order to better release the cost of operation and maintenance management, eclicktech gradually transitioned a completely self-built independent cluster to a hybrid self-built reserved cluster + elastic EMR. At present, most tasks have been implemented on-demand to run on the EMR cluster. The daily data volume reads more than one hundred terabytes, which supports the data calculation and analysis business of the global business. At present, with only 1-2 operation and maintenance personnel, hundreds of business reports can be completed every day, hundreds of business models are reanalyzed and updated, hundreds of features are screened, extracted and evaluated, and more than 10 billion traffic forecasts and evaluations can be completed every day. Advertising recommendation.

4. Federated learning + image AI + public cloud to help improve the accuracy of digital advertising

eclicktechCTO Wang Yizhou mentioned in his speech that the current accuracy of digital advertising is still very low from a technical perspective: "To give an example that is not completely accurate, netizens now have a click on almost every 100 advertising materials. Only one ad click can have a single order. There are many reasons for this, and we have been committed to increasing the proportion of the conversion funnel. On the one hand, the higher the proportion means the less interruption to users, and on the other hand, it also contributes to the efficiency of business realization. Increase and decrease in customer acquisition costs for each advertiser."

In the future, eclicktech will establish a broader and more secure data analysis platform, such as cross-customer and cross-industry data analysis and modeling through Federal Learning (Federal Learning) and blockchain technology, in order to protect customer data security. Next, improve the accuracy of customer performance marketing.

In an era where pictures and short videos are king, eclicktech plans to use NLP and image AI technology to realize the processing, assembly, splicing and display of creative materials on the premise of understanding user needs, avoiding the high cost of manual production, and truly can do Thousands of advertisements.

To achieve the above goals, a large number of computing resources and IT resources must be required. The current public cloud provides a very good base. eclicktech will continue to invest in flexible computing platforms, increase the proportion of container / serverless computing / cloud-native big data platforms and other technologies used, improve data processing and model training capabilities at unit costs, and provide effective underlying resources to increase the breadth and depth of data processing Support to improve our resource utilization in different scenarios, improve service efficiency, and reduce service costs.



Subscribe to our Newsletter, you can get more corporate information and real-time updates