How the recommendation engine works
The Similar attributes recommendation engine analyzes shared catalog attributes between a reference item (such as a product currently viewed by a customer) and other items in the product catalog. The system uses these shared attributes to find the closest matching products.
The recommendation engine looks for catalog properties (such as category, brand, size, price, and color) that are common between the reference item and other products in the catalog.
For example, if a customer sees a red T-shirt, the engine will search for other products with similar properties, such as the same color, category (for example, "T-shirts"), brand, size, or price range.
A few examples of shared attributes that could be used:
- Category: Helps the engine find products in the same product group or category.
- Brand: Ensures the engine looks for products from the same brand.
- Price: A similar price range helps ensure the recommendations are relevant in those terms.
- Color, Size, and more: Other specific attributes that match across products (like color or size for clothing).
For more details about this recommendation engine, see Similar attributes documentation.
Troubleshooting steps
If you encounter the "No items could be recommended" message, the engine couldn’t find eligible catalog items for the reference item and its configured filters. Follow these steps to debug the issue:
1. Check the reference item
Confirm that the recommendation request includes the intended reference item and that the item exists in the catalog and its ID matches the item ID in the catalog. Check that the item has values for the attributes used by the template. Missing or incorrectly mapped data can limit the number of similar items found.
2. Verify catalog attributes
Review the Exact match and Optional attributes configured for the template (for example, category, brand, color). Make sure these attributes are populated consistently on the reference item and other catalog items. If only a few items share the configured values, the recommendation may return few or no results.
3. Check the item-context dynamic filters
Similar attributes support an item-context dynamic filter. The filter narrows results by matching a catalog attribute to the corresponding attribute on the reference item. For example, a Brand filter returns items with the same brand. A restrictive filter can remove all eligible items.
Check that the reference item has a value for the selected attribute and that other catalog items have matching values. If the reference item or selected property is missing or null, the filter is skipped, and the unfiltered result set is used. Temporarily disable the filter in the preview to see whether results change.
4. Review the block list configuration
You can configure up to three blocking events per recommendation. They use OR logic: an item is blocked if it matches any one of the events. A broad event selection or time window can remove all eligible items.
Check each event, item ID property, and time window. Confirm that the event property contains IDs that match the catalog item IDs. If the property contains a list of item IDs, all items in the list are blocked. In the preview, disable or narrow one blocking event at a time to check whether results change.
5. Understand the fallback mechanism
If Similar attributes finds fewer items than requested, fallback logic fills the remaining results with randomly selected items that match the configured catalog filters. These items may be less similar to the reference item. If fallback is disabled, or no eligible items match the filters, the result may contain fewer items than requested, or no items.
6. Compare results in the preview
Preview the recommendation with a reference item known to have several similar catalog items, using the same settings as the failing item. If the known-good item returns results, compare the two items' catalog values and filter matches. If both return no results, adjust one setting at a time, for example, the item-context dynamic filter or a block list event, to identify what is limiting the result set. Restore the intended settings after testing.