Home▸AI & Machine Learning▸Word Embeddings and the Semantic Vector Space

🧭 Word Embeddings and the Semantic Vector Space

Explore a 3D semantic vector space where words become points, similar meanings cluster together, and analogies like king − man + woman ≈ queen become simple vector arithmetic.

AI & Machine Learning3DAdvanced60 FPS

Thirty-four words rendered as glowing points in a 3D "meaning space," clustered by category, connected by cosine-similarity lines, and combined with real vector arithmetic to reproduce the famous king − man + woman ≈ queen analogy.

🔬 What It Demonstrates

Word embeddings place semantically related words near each other in vector space. This scene shows that geometry directly: same-category words cluster, cosine similarity ranks nearest neighbors, and subtracting/adding word vectors performs meaningful analogical reasoning.

🎮 How to Use

Pick a category to filter the space, choose a query word to see its nearest neighbors, drag the training slider from a random untrained scatter to the organized trained layout, and run the king − man + woman analogy to watch vector arithmetic land near "queen."

💡 Did You Know?

Word2vec's 2013 discovery that analogies work as vector arithmetic was one of the first hints that neural embeddings capture structured, human-interpretable relationships — a property that modern large language models still build on.

⚙ Under the hood

Explore a 3D semantic vector space where words become points, similar meanings cluster together, and analogies like king − man + woman ≈ queen become simple vector arithmetic.

machine learningneural networksvector spacesembeddingssemantic analysisdata representationThree.js

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install