Word Embedding Space Simulator

Word Embedding Space

Selected word

Cluster spread 1.0

Rotation speed 0.3

Selected

–

Nearest word

–

Cosine similarity

–

Words plotted

–

How it works

A word embedding model (Word2Vec, GloVe, fastText) assigns every word a dense vector such that words used in similar contexts end up close together in the vector space. This scene projects each word's high-dimensional embedding down to 3D and plots it as a point, so semantic clusters — animals, royalty, numbers, cities — become visible as literal clumps in space. The "closeness" between two words is measured with cosine similarity, the angle between their vectors, not raw Euclidean distance — direction, not vector length, carries the meaning.

cos_sim(a,b) = (a · b) / (‖a‖ ‖b‖)
analogy: vec(king) − vec(man) + vec(woman) ≈ vec(queen)

Selected word — pick any word to see it highlighted and its embedding vector drawn from the origin.
Nearest neighbors — toggle lines connecting the selected word to its closest neighbors by cosine similarity.
Analogy — runs the classic word-vector arithmetic and draws the resulting point, showing why embeddings support analogical reasoning, not just lookup.
Cluster spread — exaggerates or compresses inter-cluster distance to make the semantic groupings easier or harder to see.

Real-world relevance: this is exactly the representation that sits underneath every modern NLP pipeline — search engines use embedding similarity for semantic search, recommendation systems for "related items," and transformer models start every layer by looking up a token's embedding before applying attention.

Selected word Nearest neighbor Analogy result Other words

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.