Glossary
semantic search
Search that matches by meaning rather than literal terms, using embeddings to rank results by similarity to the query's intent rather than its surface tokens.
The retrieval pattern where “what is the capital of France” can find a
document that contains the word “Paris” even if the document never uses
the word “capital.” Each document and query is embedded; ranking is
cosine similarity. Compared to lexical search (BM25, TF-IDF) it
handles synonyms, paraphrase, and cross-lingual cases far better.
The cost is precision on exact-term queries. Searching for a specific function name or a product SKU often degrades under pure semantic search because the embeddingretrieval-memoryA fixed-size vector representation of a piece of text learned so semantically similar texts land near each other in the vector space, the basis for vector search and most RAG. Open full entry model summarizes meaning at a higher level than tokens. The standard fix is hybrid search: run lexical and semantic in parallel, fuse the scores (reciprocal rank fusion is the common choice).
Modern RAGretrieval-memoryA pattern where a model retrieves relevant documents from an external store at query time and conditions its answer on them, instead of relying only on parametric knowledge. Open full entry stacks default to hybrid plus a reranker on top: cheap recall from BM25retrieval-memoryA classical lexical ranking function for information retrieval, based on term frequency and inverse document frequency with saturation, still the strong lexical baseline for hybrid search. Open full entry and semantic search, then a cross-encoder reranker picks the top few from a larger candidate pool.