Complete the dot product as cosine identity on unit vectors

Attempt it

For vectors u and v with ||u|| = ||v|| = 1, the dot product u·v equals between them, which ranges from for opposite-direction vectors to for same-direction vectors.

Quick reference

TL;DR

On unit vectors, the dot product equals the cosine similarity, with -1 meaning opposite directions and +1 meaning the same direction.

Memory aid

Sign in for the mnemonic that makes this stick. See Pro pricing.

Easy to grasp

Picture two arrows of equal length pointing on a clock face. If they both point at 12 they are pointing the same way and their dot product is the maximum, 1. If one points at 12 and the other at 6 they are exactly opposite and their dot product is -1. Halfway in between, at 3 o'clock, the answer is 0. The number tells you 'how much do they agree' on a tidy scale from minus one to plus one.

Key concepts

Embeddings

Detailed answer & concept explanation

~3 min read

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.

Sign in for 5 free unlocks/day or go Pro for unlimited. See Pro pricing.

Quick scan · 2–4 min

Everything important, quickly.

5-7 min: state the identity + derive the bound + interpret +1, 0, -1 + empirical distribution in trained encoders + calibration considerations.

Real-world examples

Real products, models, and research that use this idea.

OpenAI text-embedding-3 returns vectors whose pairwise cosines rarely fall below 0 in real corpora.
Pinecone Hybrid Search documentation notes that cosine scores below 0.4 typically mean 'unrelated' in trained encoder spaces.
BGE-M3 model card shows score distributions clustered in [0.0, 0.9] for typical retrieval tasks.
Cross-encoder rerankers like BGE-reranker-v2 produce calibrated probabilities, unlike raw cosine.

Interviewer follow-ups

What an interviewer would ask next. Try answering before peeking at the approach.

QWhy are truly negative cosines so rare in trained embedding spaces?

Sign in or go Pro to see the approach for this follow-up.

QHow would you map cosine scores to a calibrated probability for a downstream UI?

Sign in or go Pro to see the approach for this follow-up.

QIf the empirical distribution sits in [0, 0.9], should you normalize the score range to [0, 1] for display?

Sign in or go Pro to see the approach for this follow-up.

Don't say this

Red flags and common mistakes that signal junior thinking. Click to expand.

Most common mistake

Stating the cosine range as [0, 1] instead of [-1, 1]; cosine includes negative values when vectors point in opposing directions.

Sign in or go Pro to see all red flags and common mistakes. See Pro pricing.

60-second cheat sheet

The night-before-the-interview bullets. Scan these on the way to the call.

Sign in for 5 unlocks/day or go Pro for unlimited revision sheets. See Pro pricing.

References

Primary sources. Skim if you want the original framing.

Sentence Transformers documentation on score interpretation