What is Hybrid search?

Traditional keyword search finds documents that contain the exact words a user typed. Semantic search finds documents that match the meaning behind a query, even when the wording differs. Hybrid search runs both approaches in parallel and merges their scores, so a query like 'comfortable running shoes for bad knees' can surface results that mention 'joint support' or 'cushioned trainers' alongside exact keyword hits. The result is a search layer that handles precise lookups and exploratory, conversational queries alike.

How does hybrid search work?

• A lexical engine (such as BM25) scores documents by term frequency and exact match.
• A vector engine converts the query and documents into numerical embeddings and measures meaning similarity.
• A fusion step - often Reciprocal Rank Fusion or a weighted blend - combines both score lists into one ranked output.
• The merged list is returned to the user or downstream component.

Why does it matter?

Shoppers and site visitors rarely type perfectly. They abbreviate, misspell, or describe what they want in their own words. Pure keyword search misses those queries; pure semantic search can rank vague results too highly when an exact match exists. Hybrid search narrows that gap, which typically lifts click-through rates and reduces 'no results' pages. For automotive dealerships, it also means inventory searches for partial VINs or colloquial trim names still surface the right vehicles.

Nobi uses hybrid relevance ranking - semantic understanding plus behavioral signals - so results reflect what shoppers mean, not just what they typed.

Frequently asked questions

How is hybrid search different from semantic search? Semantic search relies entirely on meaning-based vector similarity. Hybrid search adds a keyword matching layer on top, so exact-term queries (like a product SKU or a specific model name) are still retrieved with high precision even when the semantic model is uncertain.

Does hybrid search require more infrastructure than keyword search? Yes, it requires both a traditional inverted index and a vector index, plus a fusion layer. Most modern search platforms bundle all three, so the operational overhead is lower than building each piece separately, but storage and compute costs are higher than keyword-only setups.

What is Reciprocal Rank Fusion (RRF)? RRF is a common method for combining two ranked lists without needing to calibrate their raw scores against each other. Each document earns a score based on its position in each list - higher rank means higher score - and those position-based scores are summed to produce the final order.