Fusion Retrieval
Also called Fusion Retrieval
Run several searches at once and blend their results into one ranked list instead of trusting a single method.
Think of it like
Asking three librarians the same question and merging their piles — the book everyone points to floats to the top.
Example
A support bot fires a keyword search and a meaning-based search for the same question, then fuses both lists so a good hit from either one still surfaces.
How it actually works
No single retrieval method is best for every query — keyword search nails exact terms, vector search catches paraphrases. Fusion runs multiple retrievers in parallel and combines their rankings, usually with a scoring rule like reciprocal rank fusion. The payoff is coverage; the cost is running and merging more searches per query.
For product teams
Fewer "I couldn’t find that" misses, because a weak result from one method is rescued by another.
For engineers
Query N retrievers concurrently, then combine their ranked lists via a fusion function (commonly RRF) into a single ordering.
Related
- Reciprocal Rank Fusion — The scoring rule most often used to merge the lists.
- Hybrid Search — Running keyword and vector search together is the classic case.
- BM25 — Blends the strengths of keyword-based BM25.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome