Attribute Weighting in Recommendations Criteria w/o a filter value? | Community
Skip to main content
D
DavidLu17
Level 3
April 8, 2026
Solved

Attribute Weighting in Recommendations Criteria w/o a filter value?

  • April 8, 2026
  • 1 reply
  • 40 views

I’d like to use attribute weighting in an item-based criterion in a way that makes sense to me but apparently didn’t occur to Adobe devs. The current setup has a weighting % slider, then you choose the attribute, then an operator, then you must enter a static value, so this is really just a leaky filter.

What I want I want from your AI is this:

  • Look at the attributes of the current item
  • Look at which attribute(s) I’ve selected to weight
  • Then look at the catalog items that also have this/these attribute(s)
  • Recommend items with rank based first on similarity in these attribute values
  • Ideally, you could also apply an operator here to tailor matching vs. anti-matching, e.g., highly recommend items that are more or less different in these attribute values
Best answer by PrasanthV

Hi ​@DavidLu17,

 

I believe it will be helpful to you.

 

1. Use the "Content Similarity" Algorithm

The user is struggling because they are likely using a behavioral algorithm (like "People Who Viewed This") and trying to force similarity via filters. To achieve what they described, the Algorithm Type must be set to Content Similarity. This tells Target to use the current item’s metadata as the baseline for the "AI" to find matches.

2. Leverage Dynamic Attribute Matching

Instead of entering a static value (which creates that "leaky filter" feeling), use the Current Item match.

  • To Match: Set a rule where Attribute X | Matches | Current Item's Attribute X.

  • To Anti-Match: Set a rule where Attribute Y | Does Not Match | Current Item's Attribute Y.

This allows the engine to look at the "Blue" shirt the user is viewing and dynamically boost other "Blue" items without you ever typing the word "Blue."

3. Adjust Weighting for "Fuzzy" Ranking

Once the "Matches" rules are set, the Weighting % slider acts as a ranking booster rather than a binary "on/off" switch.

  • If you weight Color at 100% and Brand at 20%, Target will prioritize items that match the color first, then sort by brand similarity within those results.

  • This avoids the "empty results" problem because it’s a boost to the score, not a requirement for the item to show up.

4. The "Secret" Solution: Custom Weights via Feed

If the UI is still too restrictive, the best architectural move is to calculate a "Product Affinity Score" in the product feed itself.

  • Pass a comma-separated list of "Related IDs" or a "Style Group ID" into a custom entity attribute.

  • In Target, weight based on that pre-calculated logic. This bypasses the UI limitations entirely and lets your own data science logic drive the ranking.

1 reply

P
PrasanthVAccepted solution
Level 3
April 10, 2026

Hi ​@DavidLu17,

 

I believe it will be helpful to you.

 

1. Use the "Content Similarity" Algorithm

The user is struggling because they are likely using a behavioral algorithm (like "People Who Viewed This") and trying to force similarity via filters. To achieve what they described, the Algorithm Type must be set to Content Similarity. This tells Target to use the current item’s metadata as the baseline for the "AI" to find matches.

2. Leverage Dynamic Attribute Matching

Instead of entering a static value (which creates that "leaky filter" feeling), use the Current Item match.

  • To Match: Set a rule where Attribute X | Matches | Current Item's Attribute X.

  • To Anti-Match: Set a rule where Attribute Y | Does Not Match | Current Item's Attribute Y.

This allows the engine to look at the "Blue" shirt the user is viewing and dynamically boost other "Blue" items without you ever typing the word "Blue."

3. Adjust Weighting for "Fuzzy" Ranking

Once the "Matches" rules are set, the Weighting % slider acts as a ranking booster rather than a binary "on/off" switch.

  • If you weight Color at 100% and Brand at 20%, Target will prioritize items that match the color first, then sort by brand similarity within those results.

  • This avoids the "empty results" problem because it’s a boost to the score, not a requirement for the item to show up.

4. The "Secret" Solution: Custom Weights via Feed

If the UI is still too restrictive, the best architectural move is to calculate a "Product Affinity Score" in the product feed itself.

  • Pass a comma-separated list of "Related IDs" or a "Style Group ID" into a custom entity attribute.

  • In Target, weight based on that pre-calculated logic. This bypasses the UI limitations entirely and lets your own data science logic drive the ranking.

D
DavidLu17Author
Level 3
April 16, 2026

Thanks, Prasanth, for the detailed reply.
"Content Similarity" is not present in our version but I see “Items with similar attributes” which must surely be the same thing.
As to #4, that’s great also; we figured out something similar already. Some of our products have up to 50 variants (different entity.ids) but share a common base entity code so we added that code as an attribute. We use it to filter out entities that are too similar to the current item. But thanks for the tip, I see that we can further extend this with additional attributes.

P
PrasanthV
Level 3
April 17, 2026

Thanks for the update!

Terms & ConditionsAccessibility statement

Loading footer...