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

0.844

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

Leopard0.844

Lion0.794

Mouse0.076

Cheerful-0.476

Music-0.511

Nearest to Leopard

Tiger0.844

Lion0.823

Mouse0.054

Computer-0.441

Computer mouse-0.448