Chapter 5 · Turning language into numbers
Semantic search
Search by meaning: embed the query and find the closest stored vectors, even with different words.
Chapter 5 · Turning language into numbers
Search by meaning: embed the query and find the closest stored vectors, even with different words.
Query: “affordable holiday”
Every document is already stored as a vector. The query becomes a vector too, and the nearest stored vectors are the best matches.
Classic search checks which documents contain your words. If you type affordable holiday and the page says budget travel, it finds nothing.
Type a query or tap an idea. The left list counts shared words. The right list compares vectors. Try wording that does not appear in any document.
No document contains these words.
A teaching toy: 16 short documents and hand-made vectors over 8 topics. Real semantic search uses learned embeddings with far more dimensions, so it understands many more words and phrases.
Why can semantic search find 'budget flights' for 'affordable holiday'?
When is each document turned into a vector?
Which score ranks the best match first in meaning search?