After an algorithm model is successfully trained, you can use it to recommend products. You can retrieve the top N recommended products for users or a user's preference score for a specific product.
Create a recommendation task
A recommendation task uses the sole successfully trained algorithm model in the workspace. Ensure that a successfully trained model is available before you create a task.
Recommendations can be made only for users who are included in the behavioral dataset for the algorithm model and have made a purchase within the last year.
Users who will receive recommendations must be saved as an audience. The audience ID type must match the user identity type in the behavioral dataset. Users in the audience who are not in the dataset are excluded from recommendations.
Recommendations can be made only for products that are included in the product tag dataset for the algorithm model and have been purchased within the last year.
Products to be recommended must be saved as a product pool. Products in the pool that are not in the product tag dataset are excluded from recommendations.
The result of Audience size × Product pool size must be less than or equal to 50 billion. Otherwise, the recommendation task fails.
If the product pool contains only one product, the recommendation result changes from the user's top N recommended products to the user's preference score for that product.
A prompt above the list shows the number of used prediction tasks out of the total number of purchased prediction tasks. This number represents the total for all workspaces in the organization. Creating a new task consumes one available prediction task. Failed tasks are not counted.
Procedure:
Go to Workspace > User Insights > Product Recommendation > Product Recommendation > My Tasks.
In the upper-right corner, click Create Task. The configuration page appears.
Enter a task name, select an audience as the user source, and select a product pool as the product source. Then, enter the number of products to recommend. The value must be an integer from 1 to 10.
NoteWhen you set the number of recommended products (N), refer to the accuracy and recall rate in Model Validation.
Select the checkbox to confirm that creating the task consumes an available prediction task, and then click Confirm. The system then starts to execute the recommendation task.
Manage recommendation tasks
The task list is shown in the following figure.
The task execution statuses are as follows:
Pending: If the number of model training and product recommendation tasks running in the organization exceeds five, the excess tasks are queued for execution.
Running
Successful
Failed: A task fails if it does not produce a result within 24 hours. A task also fails if the source audience contains no recommendable users or the source product pool contains no recommendable products. Hover the pointer over the
icon to view the reason for the failure.Offline: A task automatically goes offline 180 days after its creation. This means the task data is deleted, but the record is kept in the list.
NoteWhen a task goes offline, if the corresponding model is also offline and has no other associated tasks, the model is also deleted.
You can view the result details of a task or remove a task.
View result details
For a successful task, click the 
icon to view the recommendation results. For more information, see the View result details section below.
Remove a task
For any task that is not offline, you can click the 
icon to delete it. This action deletes both the task data and its record from the list.
When you delete a task, if the corresponding model is offline and has no other associated tasks, the model is also deleted.
View result details
For a successful task, click Result Details to go to the details page. On this page, you can view the task information and detailed results.
Task information
The task information is displayed at the top of the details page.
The details are as follows:
Actual Recommended Users: The number of users from the source audience who were eligible for recommendations. Only these users were included in the recommendation process. Other users were considered invalid and excluded.
Task Products: The number of candidate products for the recommendation. Recommended Products is the top N value that was set for the task.
Detailed results
The Detailed Results tab displays different results depending on whether the product pool contains a single product:
If the product pool contains multiple products, the Detailed Results tab shows:
The top N recommended products for a sample user:
If the product pool contains only one product, the Detailed Results tab shows:
The preference score of a sample user for the product. The value ranges from 0 to 1. A higher value indicates a higher probability that the user will purchase the product.
The distribution of user preference scores for the product. Hover the pointer over the score graph to view a pop-up window that shows the number of users who received the current score and the total number of users who received a score greater than or equal to the current score.
Click Create Audience to save the recommended users as an audience for a specific product that you plan to recommend:
Determine which users to save to the audience:
If the product pool contains multiple products, you can save users for whom the specific product is ranked among the top N recommended products.
If the product pool contains only one product:
You can save a specified number of users with the highest preference scores for the product.
You can also save users whose preference scores for the product fall within a specified range.
Select the ID type for the audience. You can use the user identity from the behavioral dataset that is used by the algorithm model. You can also use other ID types from the behavioral dataset or the source audience of the prediction task. Select an ID type that is convenient for future marketing activities. If a user does not have data for the selected ID type, that user is not saved to the generated audience.
Enter a name and description for the audience. Select a folder in which to save the audience and associate it with a sub-campaign. For more information, see the Marketing Campaign document.
Click OK to create the audience, or click OK and Go to go to the Audience Management page to view the audience.
On the Audience Management page, the source of this type of audience is People-Product Analytics. You can perform operations on these audiences, such as analysis, deletion, moving, downloading, pushing, and sending them to applications. For more information about these operations, see Basic audience features and Audience push feature. In addition, you can create different marketing tasks for audiences with different ID types in the User Marketing module. For more information, see the User Marketing document.
