How Search Features Interact with Loomi Search+

This guide covers how Loomi Search+ retrieves the recall and rank search results.

Loomi Search+ query processing mechanism

Loomi Search+ processes queries using the mechanism described below.

A. Keyword search processing (existing behavior)

1. Query Side Enhancements
   The search engine enhances the query with Synonyms, canonicalization, and stopwords removal.
2. Text match
   The search engine retrieves products that match all query parts and Synonyms of each query part.
3. Search Recall Precision Modes (Product Type Precision, Category Precision, LLM-based Precision)
   Precision modes that deepen recall relevance for keyword searches. Precision modes act as a filter for noisy products. The Keyword Search recall includes precise product matches.

B. Vector search processing

• Vector Search recalls relevant product matches based on the concepts or intent identified from the query. This is based on the vector embeddings generated from the product catalog. The vector temperature settings determine how precise the product matches are.

Loomi Search+ recall returns the sum of Keyword Search and Vector Search recall (A+B). This allows the Search engine to include all potentially relevant products.

Improving null results with vector temperature relaxation

Loomi Search+ filters the most relevant products using vector temperature. Think of it as a minimum requirement for relevance. Only products with similarity scores higher than the vector temperature are shown to ensure that shoppers see the most meaningful results.

Loomi Search+ offers two temperature thresholds:

1. Standard vector temperature shows a broad range of complementary products, allowing shoppers to explore more of your catalog.

2. High vector temperature shows more precise results that will only attempt to match the exact query intent.

What happens when there are null results?

Loomi Search+ uses a vector temperature relaxation mechanism to ensure that shoppers see the best possible results instead of seeing nothing at all.

Temperature relaxation mechanism

1. When Loomi Search+ finds no results, it lowers the vector temperature to expand the recall. This relaxes the search criteria to find the most relevant results.

2. If there are still no results, Loomi Search+ further relaxes the temperature to its minimum limit and attempts to match up to 200 relevant results.

Refine vector temperature relaxation

Bloomreach products are pre-configured with e-commerce best practices and strive to optimize your revenue by providing the best shopper experience.

Temperature relaxation aims to strike a good balance between returning relevant, exploratory search results and ensuring that shoppers discover products related to their interests by expanding the recall.

If you feel relaxed results lead to a poor shopper experience, please contact your Bloomreach representative to refine your Loomi Search+ algorithm.

Loomi Search+ ranking

Ranking scores

• We compute keyword score for all the products recalled via keyword match.
• We compute vector score for all the products recalled via vector search.
• We then compute a combined score using a mathematical concept that considers both the exact keyword matches and the related semantic context from the vector search.

Ranking mechanism

• If a product appears in the recall of both keyword search and vector search, this product will be ranked higher.
• Additionally, keyword match gets a higher ranking preference over vector search recall.
• The primary signals for search ranking are still performance signals like query-specific revenue, conversion, ATC, and views, as well as sitewide revenue, conversion, ATC, and views.