Embeddings & Search

Similarity Metrics

Similarity metrics help computers decide which item is closer or more related.

Known words use preset semantic positions that group similar meanings together. Unknown words get a character-based estimate.

Interactive Playground

Query

Compare against

Item A

Item B

Statistics

Query

Best Cosine

Best Euclidean

Best Dot Product

How Cosine Similarity Works

            cos(θ) = (A · B) / (|A| × |B|)
          

angle between two vectors in space

Score scale

1.0

Identical meaning, vectors point the same direction

0.5

Related but different topics

0.0

Completely unrelated, vectors are perpendicular

Live Visualization

🎓 Tips 2 tips

A score above 0.85 is usually strong similarity, exact thresholds depend on your model, but 0.85+ typically means very similar meaning.

Cosine ignores length, it measures angle, not distance. A short tweet and a long article about the same topic can both score 0.9+.

💡

Key Takeaway

Different similarity methods can rank results differently picking the right one depends on your data.