Personalized Content Recommendation

Help media companies build a discovery service for their customers to find the most appropriate content.

Overview

With an ever-increasing content library, it is increasingly difficult for people to find the content they really want to read, listen to, or watch. Building a recommendation engine based on a user's habits is likely to produce a better experience for these users. The inference is based on their behavior. To gather, store, secure, process, and display all of the data required from a large number of data sources can become complex, time-consuming, and costly, as data is ingested from different systems each, with specific means of gathering the logs and source data.

Alibaba Cloud's Personalized Content Recommendation solution allows you to build and launch a personalization solution, on a global basis, in the shortest possible time. It brings together fully or partially managed large-scale data warehouse systems to support massive datasets. New data sources, features, and learning models can be tested, measured, validated, and deployed, so an inference model can be sought quickly, and iterated and improved continually. The solution will also show how easy the “preferences” result can be exposed to Front-End systems immediately. End users can have the best experience based on their real-time actions.

Solution Highlights

  • Data Security

    This solution provides you with an easy access to China’s largest wholesale marketplace (B2B), and let you connect with over 120 million enterprise users.

  • Flexibility

    By leveraging Alibaba Cloud’s global infrastructure footprint and secure global network, Alibaba Cloud offers you an efficient and safe way to connect your enterprise systems with 1688.com.

  • Scalability

    This solution integrates API and CDN services to support a global scale, providing the result quickly to each end-user irrespective of where they are and how many other requests have been made.

  • Cost Effectiveness

    The solution leverages Archive methodologies to reduce costs without risking anything needing to be re-ingested and processed. Only when data is considered exhausted will it be removed to reduce costs.

How It Works

Our Solution

This Personalized Content Recommendation solution enables you to derive insight from many data sources such as Users, Content, and Interaction History data sources. Much of this data may already be stored in an existing database. Alibaba Cloud DataWorks makes it easy to connect to existing data-sources or to ingest batched data that can be synchronized each hour (or day) to a common repository.

Alibaba Cloud DataWorks can also merge data from many disparate sources, with ingest scheduled periodically or on other triggers. Function Compute is an alternate solution, with ingest processes enabled based on workflow logic. As a file arrives within OSS (Object Storage Service), a notification can fire a predefined function that can perform Quality Control against the logs and process accordingly.

For Real-Time data sources, Alibaba Cloud's Message Service can be configured to retrieve this data source, queued requests, and notifying upstream services. Connecting to MaxCompute is one example.

Our Solution

Once the data is in OSS or MaxCompute, queries can run within MaxCompute to understand the dataset and identify possible correlations among different attributes.

To associate textual information MaxCompute or Machine Learning Platform for Artificial Intelligence (PAI) offers various embedding and encoding algorithms to convert textual information into numerical representations like word2vec and TFIDF.

In DataWorks, you can use a full set of SQL grammar to transform the data in a flexible method. If SQL is not sufficient, you can use Python or Java to extend the functions. You can also work directly in PAI, where it is possible to do more advanced feature extractions like encoding, embedding, and vectoring.

Note: You can only associate one dataset in PAI directly.

Our Solution

After the recommendation has been created within PAI, the inference needs to be made available to the audience.

Within PAI, getting output to an audience is relatively simple since PAI-EAS (Elastic Algorithm Service) can expose the recommendations within a native API. Alternatively, DataWorks can also do this in DataService Studio, which creates a data bus to publish to an API. DataService Studio allows you to quickly create APIs based on data tables or register existing APIs with the DataService Studio platform for centralized management and release. This way PAI and/or other systems can be brought together in a single managed service.

Whether PAI or DataWorks is used to create an API, it can be auto-scaled and secured by leveraging Alibaba Cloud API Gateway, which provides permission management, traffic throttling, monitoring, and alarms to ensure the solution can be easily managed and scaled to a global audience.

For global reach and further scalability, Alibaba Cloud CDN can integrate into the API Gateway to accelerate dynamic and static content to help your customer access inference results globally. With 2,800 nodes covering six continents and over 120 Tbps of connectivity with major ISPs worldwide, global access to these recommendations is fast and reliable.

Our Solution

Once a Personalized Content Recommendation model is in production, it is important to monitor its prediction accuracy regularly. It is common for a prediction model to deteriorate over time due to environment changes. To enable this long-term evolution, DataWorks allows periodical aggregation tasks to be established to calculate the prediction accuracy over various control groups. DataV can also show findings in a monitoring dashboard with constant visibility of results for non-technical team members.

Once defined, a customer needs to see where an issue might be and resolve it by retraining the model with the latest data and models. PAI is constantly improving to include the new modeling tools and features, to ensure the best data science is supported.

This solution makes it easy to establish a new version of the algorithm, test this in lab conditions, further test with a small sample of users, and scale it to a wider audience. The solution allows real data to be measured (with the new model rolled out or not) based on whether it is considered to better and by how much.

Related Resources

Webinar

How to Build a Personalized Recommendation Engine with MaxCompute

Learn how Alibaba Cloud's MaxCompute can help build an end-to-end recommendation engine to boost online retail.

Solution

Broadcast Live. Broadcast Now.

This solution provides tools and best practices to ensure a live stream is ingested, processed, and distributed to a global audience.

Solution

Secure Media Archive

This integrated solution uses Alibaba Cloud services to help organizations to securely store content into the cloud.

View All

Start with Alibaba Cloud Solutions

Learn and experience the power of Alibaba Cloud with a free trial.

Contact Sales