Dot Product Search
Also called Inner Product Search
Ranking vectors by their dot product — fast, and equal to cosine when vectors are normalized.
Think of it like
Scoring how aligned two arrows are, rewarding both same-direction and longer arrows.
Example
A vector DB configured for inner product returns matches slightly faster than cosine because it skips the normalization step.
How it actually works
The dot product combines angle and magnitude, so longer vectors score higher even at the same direction. On unit-normalized embeddings it is mathematically identical to cosine similarity but cheaper to compute, which is why many indexes prefer it. On un-normalized vectors it behaves differently, and using it by accident is a subtle source of bad rankings.
For product teams
A performance-friendly default when embeddings are normalized; otherwise a footgun.
For engineers
Maximum inner product search; equivalent to cosine under L2 normalization, cheaper without the divide.
Related
- Vector Similarity — A specific vector similarity metric.
- Cosine similarity — Equals cosine on normalized vectors.
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