Background information
During interactions between users and the recommendation system, some recommendations may not meet user expectations. This may be caused by diversity policies that deliver unexpected results, interest shifts that come with a change in the life stage of a user, or topics and images that make users feel uncomfortable.
The negative feedback feature enables the interactions between users and the recommendation system and provides more personalized recommendations based on the negative feedback of users.
Description
Artificial Intelligence Recommendation currently supports negative feedback in the following three dimensions:
1. Block a single item
Example: An item such as a product, content, or news appears in the recommendations for Alice but Alice does not like this item or some images or descriptions of this item make her feel uncomfortable. The negative feedback feature can block this item for Alice.
Effect: This item is blocked and no longer appears in the recommendations for Alice. The block operation takes effect permanently.
Required operations: Track the negative feedback and report the behavior data in compliance with the related data specifications.
Note: If you have multiple negative feedback options, you must map them to dimension 1, 2, or 3 based on the scope of impacts.
The following table provides the specifications for data reporting.
item_id | item_type | bhv_type | bhv_value |
Required | Required | Reported value: "dislike" | Reported value: "dislike_item" |
Note: For more information about the specifications of behavior data, see data specifications of a specific industry.
2. Dislike specific characteristics of an item
Example: A dual-shade contouring palette from a specific brand appears in the recommendations for Alice. Alice does not like this type of item and blocks it by selecting the highlighter characteristic.
Effect: Such items are blocked for Alice, and items that carry the highlighter characteristic no longer appear in her recommendations.
Required operations: Track the negative feedback and report the behavior data in compliance with the related data specifications, and enter correct characteristics in the product list.
The following table provides the specifications for data reporting.
item_id | item_type | bhv_type | bhv_value |
Left empty | Left empty | Reported value: "dislike" | Reported value: "dislike_class:key=value". Example: "dislike_class:category_path=12_4_5" |
The following table describes the enumerated values of keys in bhv_value.
Industry | Key | Example of bhv_value |
E-commerce | category_path; tags; band_id; shop_id | dislike_class:category_path=12_4_5; dislike_class:tags=Lady; (You do not need to enclose the value with quotation marks.) dislike_class:band_id=12; dislike_class:shop_id=123 |
Content | category_path; channel; author | dislike_class:category_path=13_4_5; dislike_class:channel=xxx; dislike_class:author=Alex |
News | organize; channel; author | dislike_class:organize=xxx; dislike_class:channel=xxx; dislike_class:author=Alex |
3. Dislike items similar to or related to an item
Example: A dual-shade contouring palette from a specific brand appears in the recommendations for Alice. Alice does not like items similar to this item and selects the option that stops recommending similar items.
Effect: Items similar to this item are blocked for Alice and no longer appear in her recommendations.
Required operations: Track the negative feedback and report the behavior data in compliance with the related data specifications.
The following table provides the specifications for data reporting.
item_id | item_type | bhv_type | bhv_value |
Required | Required | Reported value: "dislike" | Reported value: "dislike_related" |