Definition
In RAG architectures, keyword search (sparse retrieval) identifies relevant documents by matching exact tokens or lexemes using algorithms like BM25. It is technically precise for retrieving unique identifiers, jargon, or proper nouns, though it lacks the semantic flexibility of vector-based retrieval.
Exact literal matching vs. semantic meaning matching.
"A physical index at the back of a textbook mapping specific keywords to exact page numbers."
- Hybrid Search(Composition (Combining keyword and vector search))
- Vector Search(Complementary Retrieval Method)
- BM25(Underlying Ranking Algorithm)
- Sparse Embeddings(Mathematical Representation)
Conceptual Overview
In RAG architectures, keyword search (sparse retrieval) identifies relevant documents by matching exact tokens or lexemes using algorithms like BM25. It is technically precise for retrieving unique identifiers, jargon, or proper nouns, though it lacks the semantic flexibility of vector-based retrieval.
Disambiguation
Exact literal matching vs. semantic meaning matching.
Visual Analog
A physical index at the back of a textbook mapping specific keywords to exact page numbers.