⌕
Words 5,000
Dimensions 50d → 3d
Model GloVe 6B
Projection UMAP

Word Embeddings

Each dot is a word represented as a 50-dimensional vector, trained on 6 billion tokens of text. Words used in similar contexts end up with similar vectors — so "king" and "queen" are close neighbors.

Dimensionality Reduction

UMAP projects the 50D vectors into 3D space while preserving local neighborhoods. Clusters that form naturally reveal semantic categories — nature words group together, emotions cluster nearby.

Vector Arithmetic

Embeddings encode meaning as direction. The classic example: king − man + woman ≈ queen. The "royalty" direction is preserved when you swap gender. Try the analogies in the side panel.

Selected Word
Click a word or search
Nearest Neighbors
Select a word to see its nearest neighbors in embedding space
Click a dot or type in the search box
Categories