📚 Word Embeddings Visualizer
See how words become numbers that capture meaning
â‘ Select Words to Compare
â‘¡ The Embedding Vectors (Simplified to 6 Dimensions)
Select words above to see their vectors
What you're seeing: Real embeddings have 768+ dimensions.
I've simplified to 6 interpretable dimensions so you can see patterns.
Each bar shows how much that word relates to a concept.
Similar words have similar patterns.
â‘¢ Words in 2D Space (Projected from 6D)
living → non-living
↑ power/status
Notice: Similar words cluster together!
Animals are near animals, royalty near royalty.
This is how neural networks "understand" that words are related.
â‘£ Similarity Scores (Cosine Similarity)
How similarity works: We measure the angle between two vectors.
If they point the same direction → similar meaning (score near 1.0).
If they point opposite directions → opposite meaning (score near -1.0).
⑤ The Famous Word Math
king
−
man
+
woman
≈
queen
Why this works: The vector for "king" minus "man" gives you
the concept of "royalty without maleness". Add "woman" and you get
"royalty with femaleness" — which lands near "queen" in the vector space.
Similarly:
cat − small + big ≈ lion