Top-k
The number of highest-ranked candidates retained at a retrieval or reranking stage.
What Top-k does not show
Raising it increases coverage and context cost, not necessarily answer quality.
Terms this definition uses
Related terms in Retrieval engineering
How pages become chunks, vectors and candidates inside an answer engine, and why being retrieved is not being cited.
Corpus · Document ingestion · Data connector · Parsing · OCR · Layout-aware parsing · Semantic chunking · Fixed-size chunking · Chunk overlap · Parent-child retrieval · Metadata filtering · Vector database · Vector index · Sparse vector · Dense vector · Cosine similarity · Dot-product similarity · Approximate nearest neighbor · HNSW · Retrieval recall · Retrieval precision · Mean reciprocal rank · Normalized discounted cumulative gain · Relevance score · Candidate generation · Query rewriting · Query expansion · HyDE · Multi-query retrieval · Reciprocal rank fusion · Late interaction · Lexical retrieval · Semantic search · Exact-match retrieval · Freshness boosting · Authority weighting · Source diversity · Deduplication · Near-duplicate detection · Semantic cache · Retrieval latency · Retrieval cutoff · Citation mapping · Answer synthesis
Questions about Top-k
What is Top-k?
The number of highest-ranked candidates retained at a retrieval or reranking stage.
What does Top-k not show or guarantee?
Raising it increases coverage and context cost, not necessarily answer quality.
Which area of the glossary does Top-k belong to?
Retrieval engineering: How pages become chunks, vectors and candidates inside an answer engine, and why being retrieved is not being cited.
← HNSW · Retrieval recall →
See it in the full glossary · 731 terms across 19 areas.