This topic describes how to use the instance operation rules in the Artificial Intelligence Recommendation (AIRec) console to improve the diversity of recommendations. Instance operation rules take effect on instances. All the scenes of the current instance must comply with the instance operation rules. If a scene has an operation rule of the same type as another instance operation rule, the operation rule specific to the scene has a higher priority.
An item that has been browsed or clicked will not be recommended again within a specified period of time.
Recommendation blocking in a single scene
By default, an item is recommended based on the scenes configured for the item. If an item is recommended in scene 1, it is not recommended again in both scene 1 and scene 2 within the specified recommendation blocking period. The default value of Rule Level in the Deduplication Rules section for an instance is In Instance. If you set Rule Level to In Single Scene, the item is recommended based on its actual launching scene. That is, if an item is recommended in scene 1, it is not recommended again in scene 1 but may be recommended in scene 2 within the specified recommendation blocking period. We recommend that you do not set Rule Level to In Single Scene.
Log on to the AIRec console and modify the settings of Rule Level in the Deduplication Rules section.
Note: If the number of recommended items is small, you can shorten the recommendation blocking period to prevent the situation where no items are recommended.
AIRec supports only category-based scattering. You can specify the category level in an item table and the scattering window size. If only one category level exists, AIRec scatters all items. If multiple category levels exist, AIRec scatters the items based on the specified category level. The following figure shows the scattering window logic. If you set the window size to 5, the categories of all items in the window sliding direction are different in the recommendations. This avoids excessive recommendations of items of the same category and improves user experience.
Recommended results list:
Log on to the AIRec console.
1. If items do not have the category property, the system processes the items based on the default settings.
2. If items have the category property, make sure that category_level and category_path for each item match. Otherwise, the system processes the items whose category_level and category_path do not match as abnormal data and filters them out.
3. If the window size is set to 1, no scattering is performed.
4. If the configured category level is invalid, no recommendation is returned.
5. We recommend that you set the window size to the number of items on a single screen of the launching page plus 1. The window size can be a maximum of 1.5 times the number of items on a single screen.
Sorting is performed based on the item_type field. You can specify the sorting percentage for each item type. For example, you can set the sorting percentage to 50% for the article type and 50% for the image type.
Configure sorting rules for the items returned by AIRec. In the Sorting Rules section of the Instance Operation Rules page for the selected instance, select an item type in the Name column for each rule and set the sorting percentage in the Percentage column for each item type. The system returns recommendations based on the specified percentages of specific item types.
1. If the returned result or a sorting percentage is not an integer, the value is rounded to the nearest integer.
2. If the specified item type does not exist, no recommendation is returned.
3. If you do not specify the sorting rules, the system returns a result based on the default settings.
4. The sum of percentages of all item types must be 100%.