Diversification rules

The Diversification rule enables automatic reordering of products to ensure a balanced and varied product display, while preserving the overall ranking logic.

Overview

Diversification helps avoid repetitive product sequences (e.g. same brand or similar items) and improves product discovery without sacrificing performance.

Key benefits include:

• Improved product discovery
• Reduced visual repetition
• Increased engagement and CTR
• Maintained business control and performance

How it works

When applied, Diversification does the following:

• Runs on the final returned set: It executes after all other custom rules, filters, and maximum display limits (e.g., a cap of 8 items) have already been applied.
• Reorders existing results only: The algorithm does not fetch replacement items from your catalog. It only shuffles the existing list of products that successfully passed your prior rules to maximize variation across the selected attributes.
• Functions on a best-effort basis: If the final list of products contains a high concentration of identical values (for example, if 6 out of 8 products share the same brand or fiche_id), the algorithm will spread them out as much as possible (e.g., A, B, A, B, A, A, A, A). However, visual repetitions will inevitably remain because there are not enough distinct values in the final subset to completely avoid consecutive duplication.
• Modifies display order only: Product scores and underlying strategy logic remain unchanged.

Where it applies

The Diversification rule is applied on the overall strategy, after all ranking and rules have been executed. This ensures global diversity across all combined rules.

Example: avoiding multiple products from the same brand across a full recommendation list.

Attribute support

Diversification can be configured on any product attribute available in the catalog, regardless of type:

Type	Examples
String	brand , category , color
Numeric	price , rating , margin
Boolean	on_sale , new_arrival

Up to 3 attributes can be combined simultaneously.

The order of attributes matters

When you select multiple diversification fields, they are not treated as an unordered set. Instead, they follow a strict hierarchy based on the order in which you list them (e.g., [ "category", "brand" ] behaves differently than [ "brand", "category" ]):

1. Primary Field: Diversification is executed first on your first attribute.
2. Secondary Tie-Breakers: If the algorithm evaluates multiple candidate positions that are equally valid under the first attribute (i.e. "ties remain"), it will use the second attribute to break the tie and further reorder those items.

Example: Consider three items already sorted by recommendation score:

• Product 1 (Category: Shoes / Brand: Nike)
• Product 2 (Category: Shoes / Brand: Nike)
• Product 3 (Category: Shoes / Brand: Adidas)
• If you diversify only on ["category"], there is no secondary rule to resolve the category tie, and the display order remains unchanged: [P1, P2, P3].
• If you diversify on ["category", "brand"], the category tie remains, but the system uses brand as a tie-breaker, shifting the Adidas shoe to break up the consecutive Nike products: [P1, P3, P2].

Typical use cases

Common scenarios where diversification is useful include:

• Avoiding showing multiple products from the same brand consecutively
• Mixing different product types within a category (e.g. a skincare routine)
• Balancing price ranges within a recommendation list
• Alternating between promotional and full-price products