Meaning as Coordinates

From text to numbers

An embedding model converts text into an embedding: a fixed-length ordered list of numbers called a vector. You can picture every number as one coordinate on a map with hundreds of directions. Text with related meaning tends to receive nearby coordinates, even when it uses different words.

Cosine similarity is a calculation that compares the direction of two vectors instead of their raw length. A higher score usually means the texts are more related. The vector helps rank text; it cannot be turned back into the original sentence.

Analogy: A library assigns every book coordinates on a giant map of topics. A question gets coordinates too; nearby books are candidates even when they use different words.