Learn how to create recommendation scenarios in AIRec by configuring selection policies for different service types and industries.
1. What is a scenario?
A scenario is a recommendation setting tailored to a specific functional policy and user experience goal. For example, you can create a "Guess You Like" scenario for your home page based on all products, a "Guess You Like" scenario for a channel page based on a specific product category, and a "Related Recommendations" scenario for a product page.
Note: Create all your scenarios (scene_id) in the console. This ensures that you can use features such as experiment management for each scenario.
2. Difference between Guess You Like and Related Recommendations
An instance supports both the Guess You Like and Related Recommendations services simultaneously. You can create scenarios for different service types in the Scenario Management section of the console.
Guess You Like:
This service applies when the end user's current browsing intent is unclear. AIRec analyzes each user's long-term and short-term behavioral interests and provides diverse content recommendations to help the user explore their interests. Common placements include the home page, category pages, and sales promotion pages.
Guess You Like scenarios focus on recommendations based on end user behavior. You do not need to pass an item when you call the recommendation API.
Related Recommendations:
This service applies when the end user's interest is already clear. AIRec identifies the user's focus, such as specific products or articles, and recommends related content. Dynamic associations are discovered through user behavior analysis, while static associations are derived from item properties and features. Common placements include product pages and shopping cart pages.
Related Recommendations scenarios focus on recommendations based on a user-specified item. This requires that items have complete properties, such as title, content, and tags. If you do not pass an `items` parameter when requesting recommendations, the scenario falls back to a Guess You Like scenario.
The following sections use the E-commerce, news, and content industries as examples to show how to create dedicated recommendation sections.
Note: Before you use the scenario management feature, you must report your data according to the data specifications for your industry in the official Artificial Intelligence Recommendation documentation. Then, you can start your instance.
3. Custom scenarios for the E-commerce industry
The following example demonstrates how to create a dedicated recommendation section for the "Apparel" category in a general E-commerce business.
Note: You can follow a similar process to create recommendations for all categories, other specific categories, or shopping cart and order pages.
In the console, under Business Customization, click Scenario Management.
Select the service type for this scenario (Guess You Like or Related Recommendations) and configure basic information, such as the Scene ID (which corresponds to `scene_id` in the data specifications).
Note: The filtering method for a scenario can be changed only before its first publication. After a scenario is published, you cannot update its filtering method. Proceed with caution.
Method 1: Use console filtering rules to select items for the scenario
After you report data for the entire product pool, such as category orchestration and time-to-live (TTL) information, you can filter the items. For more information about data specifications, see E-commerce Industry.
The products are selected based on the following three conditions:
1. Product category: Click the drop-down list and select the appropriate category.
Note: This option corresponds to the `category_path` field in the item table.
2. Products published within the last 15 days.
Note: This option corresponds to the `pub_time` field in the item table.
You can then select "Confirm and Publish" or "Customize More Policies".
Publish directly
Note: The preview is for style reference only. To test the recommendation results for a specific user, go to Recommendation Test in the console.
After you click Confirm and Publish, the scenario will be published in about 2 hours.
Customize more policies
The Customize More Policies section provides modules for business customization, experience testing, and performance reports.
Method 2: Use the SDK to specify items for the scenario
When you create the scenario, set the filtering method to Specify items for the scenario through the SDK. After publication, push data through the server-side SDK to select products for the scenario, then manage it in the console.
Note: If a scenario was not created in the console but was created by reporting product selection data through the server-side SDK, you must create a scenario with the same Scene ID in the console to use the scenario management features.
3. Custom scenarios for the news and content industries
The following example demonstrates how to create a recommendation scenario for the "Entertainment News" section in a news app's navigation bar.
Note: You can follow a similar process to create recommendations for other news channels or different modules in the content industry.
In the console, under Business Customization, click Scenario Management.
Select the service type for this scenario (Guess You Like or Related Recommendations) and configure basic information, such as the Scene ID (which corresponds to `scene_id` in the data specifications).
Note: The filtering method for a scenario can be changed only before its first publication. After a scenario is published, you cannot update its filtering method. Proceed with caution.
Method 1: Use console filtering rules to select items for the scenario
After you report data for the entire product pool, such as category orchestration and time-to-live (TTL) information, you can filter the items. For more information about data specifications, see News Industry.
The items in this example meet the following 3 conditions:
1. Item channel: Click the drop-down list and select the appropriate category. For example, "1" in the figure represents the entertainment channel.
Note: This option corresponds to the `channel` field in the item table.
2. News published within the last 3 days.
Note: This option corresponds to the `pub_time` field in the item table.
You can then select "Confirm and Publish" or "Customize More Policies".
Publish directly
Note: The preview is for style reference only. To test the recommendation results for a specific user, go to Recommendation Test in the console.
After you click Confirm and Publish, the scenario will be published in about 2 hours.
Customize more policies
The Customize More Policies section provides modules for business customization, experience testing, and performance reports.
Method 2: Use the SDK to specify items for the scenario
When you create the scenario, set the filtering method to Specify items for the scenario through the SDK. After publication, push data through the server-side SDK to select products for the scenario, then manage it in the console.
Note: If a scenario was not created in the console but was created by reporting product selection data through the server-side SDK, you must create a scenario with the same Scene ID in the console to use the scenario management features.
4. Scenario business customization
Operational policies are configurations related to product design and user experience.
Note:
If a scenario has no operational rules configured, it will use the operational rules of the instance by default.
If operational rules are configured for the scenario, they will take precedence.
I. Fatigue rules:
Configure the scenario to use the same fatigue rules (interleaving and discretization) as the instance, or customize fatigue rules for the scenario.
Note: If you use the same fatigue rules as the instance or use the default setting, items that have been exposed or clicked in other scenarios are also excluded from recommendations in this scenario for a specific period.
II. Diversity rule examples:
Note: Set the window size to the number of items on one screen of your delivery page plus one. The maximum size cannot exceed 1.5 times the number of items on one screen.
You can test and change the switch state to suit your scenario.
1. Increase the diversity of recommended product sellers (to avoid excessive traffic concentration on a few sellers)
This corresponds to the `shop_id` field in the item table of the data specifications.
2. Increase the diversity of recommended product categories (to avoid excessive concentration on certain categories)
This corresponds to the `category_path` field in the item table.
You can click the pin button in the red box to adjust the priority of the discretization rule.
3. Increase the diversity of recommended item types (to avoid excessive concentration on certain material types)
This corresponds to the `item_type` field in the item table.
4. Interleave items based on item type
Interleaving is primarily based on the `item_type` field of an item. You can specify the interleaving ratio for each `item_type`. For example, you can set the ratio for articles to 50% and the ratio for images to 50%.
Note:
① The sum of the ratios for all types must equal 100%.
② Because the `item_type` for the E-commerce industry is typically "item", you do not need to configure interleaving for E-commerce scenarios. The following example is for the content/news industry.
4. Publishing scenarios
Before using other scenario management features, such as experience testing and performance reports, publish the current scenario first.
5. Scenario performance reports
View relevant metrics for both on-site users and items:
Business metrics:
PV_CTR: The click-through rate (CTR), which is calculated as Total Clicks / Total Impressions. Duplicates are not removed.
UV_CTR: The user-based CTR, which is calculated as Users Who Clicked / Total Viewing Users.
PV_CVR (Conversion Rate): Number of Purchases / Number of Clicks.
UV_CVR (Conversion Rate): Number of Purchasing Users / Number of Clicking Users.
User activity:
Daily active users, monthly active users.
Overall traffic scale:
Impressions, clicks, favorites, reviews, cart additions, purchases.
Traffic conversion rates:
Favorite rate, review rate, add-to-cart rate.
Per capita traffic:
Impressions, Clicks, and Total Behaviors
6. Scenario lifecycle
1. Draft status
If you exit the interface while configuring a scenario, the scenario is saved as a draft.
To publish the scenario, click Publish Scenario as shown in the figure to reconfigure and publish it.
Note: You cannot change the Scene ID when you republish a draft scenario.
2. Publishing a scenario
After you confirm and publish the scenario, the process takes about 2 hours to complete.
You must wait for the current publishing process to finish before you can modify the item filtering rules and publish again.
3. Running status
For a running scenario, you can perform business customization and experiment management as described in this document.
4. Frozen status
To stop the service for a scenario, click Freeze Scenario to deactivate it.
Note: Before you freeze a scenario, make sure it is not receiving any traffic.
To reactivate this scenario, click Reactivate Scenario in the upper-right corner of the page and republish the scenario.
5. Delete a scenario
To delete a scenario, you must first freeze it and then click Delete Scenario.
Note:
A deleted scenario cannot be recovered. Proceed with caution. After you delete a scenario, the corresponding scenario tag is removed from the item library. This means that the `scene_id` of the deleted scenario is removed from the `scene_id` field of the selected items.
Also, you cannot use the ID of a deleted scenario when you create new scenarios.