An embedding converts a passage into a list of numbers positioned in a high-dimensional space so that semantically similar text occupies nearby positions. Comparing positions measures similarity of meaning.
Documents are typically split into chunks before embedding, since a representation of an entire long document is too coarse to support precise retrieval.
Alternative Names:
Vector Embedding, Text Embedding
Why it Matters?
Chunking strategy has more practical effect on legal retrieval quality than most buyers realize. Splitting a deposition mid-answer or separating a medical entry from its date can make relevant material unfindable. Vendors that have thought carefully about document structure produce noticeably better retrieval on litigation records than those applying uniform character-count splits.
Frequently Confused with
Related terms
Frequently asked questions
Why are documents split before embedding?
Does chunking affect accuracy?
