Cosine Similarity

Interactive example of how similarity works in a vector space. Click around and see the results in the table.

3D projection. Nearby points have similar embeddings.

Normalized similarity

0.995

Very close

The demo normalizes embedding cosine around a neutral baseline. Related words are below 90°, unrelated words are above 90°, and curated opposites are 180°.

Nearest to Tiger

Leopard0.995

Lion0.995

Mouse0.289

Computer mouse-0.041

Computer-0.044

Nearest to Leopard

Tiger0.995

Lion0.995

Mouse0.216

Computer mouse-0.138

Keyboard-0.303