Skip to main content

Help Center

< Back
You are here:
Print

Semantic Search

Created On
Last Updated On
by

What is Semantic Search?

Semantic Search uses vector embeddings and sentence transformers to understand the meaning behind search queries, not just the literal keywords. A search for “comfortable office chair” will find results about “ergonomic desk seating” — even when those exact words don’t appear in the product title or description.

This is powered by the all-MiniLM-L6-v2 sentence transformer, which maps both queries and products into a shared high-dimensional vector space where semantically similar items sit close to one another.

Three Search Modes

ExpertRec offers three search modes so you can pick the right balance of accuracy and latency for your store.

Pure AI Search (~3 seconds)

  • Highest semantic relevance.
  • Best for natural language queries and discovery-focused experiences.
  • Understands intent and context even when wording is very different from your catalog.
  • Higher latency — most suited to slower, deliberate browsing experiences.

Hybrid Search (~0.3 seconds)

  • Combines AI semantic search with traditional keyword search.
  • Balanced accuracy and speed.
  • Recommended for most use cases — the sweet spot for typical ecommerce stores.
  • Captures both exact keyword matches and semantically related products in a single result set.

Traditional Text Search (~0.02 seconds)

  • Fast keyword-based search with exact-match capability.
  • Lowest latency — ideal for instant search, autocomplete, and high-QPS scenarios.
  • Best for catalog lookups where shoppers already know the product name or SKU.

When to Use Each Mode

  • Pure AI — discovery-focused experiences, editorial storefronts, long-tail catalogs, and natural-language search bars.
  • Hybrid — general ecommerce stores where you need good relevance and fast response times.
  • Traditional — catalog lookups, SKU search, and latency-critical autocomplete.

How to Enable Semantic Search

  1. Log in to your ExpertRec control panel.
  2. Navigate to Search Settings > Search Mode.
  3. Select one of the three modes: Pure AI Search, Hybrid Search, or Traditional Text Search.
  4. Save your changes. The new mode will be applied immediately to subsequent queries.

Performance Trade-offs

  • Higher semantic accuracy generally means higher latency because embedding computation and vector similarity search are more expensive than inverted-index lookups.
  • Hybrid mode is the recommended sweet spot for most stores — it captures the bulk of semantic gains while keeping response times well under half a second.
  • If you’re unsure which mode to choose, start with Hybrid and A/B test against Traditional and Pure AI on your key metrics (click-through rate, conversion, zero-result rate).

Under the Hood

Semantic Search uses the all-MiniLM-L6-v2 sentence transformer to generate 384-dimensional embeddings for both queries and indexed documents. At query time, these embeddings are compared using cosine similarity to retrieve the most semantically relevant products, optionally blended with keyword-based BM25 scores in Hybrid mode.

Table of Contents