Words, images, and users all become points in the same geometric world. An embedding space maps discrete objects into continuous vectors, placing similar items close together and dissimilar ones far apart. Distance carries meaning.

Word embeddings like Word2Vec and GloVe place related terms near each other. In that space, the vector difference between king and queen resembles the difference between man and woman. Sentence and image embeddings extend the idea to larger units.

Common embedding uses

• Semantic search and retrieval
• Recommendation systems
• Clustering and visualization
• Transfer learning features
• Cross-modal matching

Dimensions range from a few dozen to several thousand. Higher dimensions capture more nuance but demand more data and compute. The geometry is learned, not designed, which makes interpretation difficult even when the results are useful.