Recommendation module weights: full parameter reference

What is this article about?

Module adjustment and tuning introduces the 6 weights most people adjust. Several Raptor recommendation modules, on both your website and in email, expose a much larger set of tuning parameters, letting you shape both which products can be recommended (the candidate set) and the order they come out in (reranking). This article is the complete reference: every one of these additional weights, its API parameter name, its valid range and default, and what it does.

Use it alongside Module adjustment and tuning for the basics, and alongside the specific module's own setup article, such as How to setup GetPIMRelatedItems, when you need more control than the core 6 weights give you.

Prerequisites

• Familiarity with the basics of module tuning: the weight scale, and the difference between candidate-set and rerank weights. See Module adjustment and tuning.
• A live recommendation module already returning results.

Core weights

Covered in full in Module adjustment and tuning: Cross-sell weight, Look-alike weight, Buy history weight, Basket weight, Visit history weight, and Search history weight. Not repeated here to avoid duplication, see that article for definitions and examples.

Candidate-set expansion weights

These weights add new products to the candidate set, they don't just reorder what's already there. Each is disabled at 0 and otherwise set between 1 and 20; 1-5 is low impact, 6-8 moderate, 9-10 high, and 20 strongly boosts matching products to the top.

Products bought together weight (ProductsBoughtTogetherWeight)
Identifies products frequently purchased together with the requested product. Looks at every transaction involving the requested product and finds which other products most often appear in the same transaction. Only products bought together at least 3 times are included.

MasterIds bought together weight (MasterIdsBoughtTogetherWeight)
The same idea as products bought together, but at the MasterId level (grouped products, such as size or colour variants of the same item). Only MasterIds bought together at least 2 times are included, keeping the correlation meaningful.

Top selling together with product category weight (TopSellingTogetherWithProductCategoryWeight)
Looks at every purchase containing a product from the requested product's category, then surfaces the products most often bought alongside that category. Products from the same category are technically included too, but should be deboosted separately using Same category weight, since the point of this weight is cross-category selling.

Categories bought together weight (CategoriesBoughtTogetherWeight)
Identifies which categories are frequently bought together, based on past purchases where at least two customers bought from both categories. Products from these related categories become recommendation candidates.

Brand weight (BrandWeight) Adds the currently most trending products from the same brand as the requested product to the candidate set. Trending here means more product-page views in the last 30 days than that product typically gets.

Target group weight (TargetGroupWeight)
Adds trending products from related target groups (for example, showing trending "children" products alongside a "girls" product). Related target groups must be defined manually via the RelatedTargetGroupSchema input parameter; the product's own target group is always included automatically.

Look-alike products weight (LookAlikeProductsWeight)
Identifies products frequently viewed together with the requested product within the same session or a short window, useful for surfacing complementary products even without a strong purchase correlation. Only products viewed together at least 2 times are included.

Serendipity for categories bought together (SerendipityForCategoriesBoughtTogether)
A modifier for Categories bought together weight, not a standalone candidate-set weight. Popular categories tend to get bought together with almost everything, which can crowd out more relevant categories. This setting suppresses that popularity effect. Test with 0, 30, 100, 300, 1000, or 3000: 0 is disabled, 30 is a very mild suppression, 3000 is a strong suppression that removes nearly all the most common categories.

Rerank-only weights

These weights never add new products to the candidate set, they only move existing candidates up or down. Each uses the standard -20 to 20 scale: negative deboosts, positive boosts, and 0 disables the weight.

Similar items weight (SimilarItemsWeight)
Deboosts products similar to the requested product, since similar products are already covered by the GetSimilarItems module. Similarity is deduced from visit behaviour by default, or can be set explicitly with ExplicitSimilarCategories and a SimilarCategoriesSchema.

Same category weight (SameCategoryWeight)
Boosts or suppresses products in the same category as the requested product. For cross-selling, you generally want to show complementary products from other categories, which is why the default suppresses same-category products.

Similar category weight (SimilarCategoryWeight)
Boosts or suppresses products from categories marked as similar to the requested product's category. Requires ExplicitSimilarCategories enabled and a SimilarCategoriesSchema mapping, for example mapping "Mountaineering Boots" to "Off-trail Heavy Boots" so the two aren't recommended against each other.

Same brand weight (SameBrandWeight)
Boosts or suppresses trending products from the same brand as the requested product. Whether to lean into or away from the same brand depends on the product, a phone case makes sense from the same brand as the phone, but a mouse for a laptop often doesn't need to match brand. A small positive value (1-2) is the recommended starting point.

Similar MasterId weight (SimilarMasterIdWeight)
Boosts or suppresses products from MasterIds considered similar, based on being viewed by the same users. Set negative to avoid recommending near-duplicate product groups.

Associated category weight (AssociatedCategoryWeight)
Boosts or suppresses products from categories associated with the current content, based on what other people who viewed this content also viewed. Considers the depth of the category path, a more specific shared category (for example "sofa" rather than the broader "furniture") gets a stronger boost.

Associated brand weight (AssociatedBrandWeight)
The brand equivalent of Associated category weight: boosts or suppresses products from brands associated with the current content, based on what other viewers of this content also viewed.

Merchandising

Merchandising Boost weight (MerchandisingBoost1Weight)
Controls how strongly the products selected in a Merchandising campaign are boosted or suppressed. This weight is set automatically when you create a campaign in Merchandising, you don't normally need to touch it directly. See Introduction to Merchandising.