Start an instance
After you purchase an Artificial Intelligence Recommendation (AIRec) instance, you must start the instance to use AIRec features.
If you have no data uploaded to MaxCompute and are not familiar with related operations, use the Quick Start method to start the instance. If you have uploaded user data, item data, and behavioral data to MaxCompute and are familiar with related operations, use the Historical Data-Based Start method to start the instance.
The following video walks you through the start process.
Use historical data to start an instance
Note: After you select this method to start the instance, you must click Activate to validate historical data. It takes approximately 1 hour to validate historical data.
Push data
Note: You can push data only after the instance is started.
Before you push data, you must prepare the data based on the E-commerce industry for specific industries. Then, you can follow the steps in the following video to push data by using server SDKs. This video also shows common data push failures.
For more information about the sample Java code used to push data, see Push data. For more information about how to install dependencies of server SDKs and how to use server SDKs for other programming languages, see the "SDK for Java" topic and the related topics that are in the same directory as the "SDK for Java" topic.
Create a scene by using the console
Note: After data is pushed, you must create a scene. The following video walks you through the process of creating a scene.
Obtain recommendation results and perform a user experience test
Note: Before you perform a user experience test, make sure that the scene is published. If you do not create a scene, you can still use server SDKs to obtain recommendation results, but the experience test is affected.
The following video walks you through the process.
For more information about the sample Java code used to obtain recommendation results, see Obtain recommendation results. For more information about common causes of empty recommendation results and related information, see User experience tests.
Compare the recommendation results with real traffic to verify the recommendation effect
Take note of the following points: 1. Make sure user randomness when you use real traffic to verify the recommendation effect. 2. Avoid inclusion of traffic from other sources and use control variables for recommendation effect comparison. 3. Do not reorder or filter recommendation results. 4. Properly use the instance dashboard to view information. 5. Reserve QPS resources and configure a flexible billing policy in advance.
The following video provides detailed information.