Definition
A dynamic pruning algorithm for efficient top-k retrieval that uses upper-bound score estimates to skip candidate documents that cannot mathematically exceed the current minimum score of the top results. In RAG pipelines, it accelerates the initial retrieval phase by bypassing the exhaustive evaluation of low-relevance documents in large-scale indices.
Not a logical 'AND' operator; it is a performance optimization for weighted term retrieval.
"A high-jump bar that is raised every time a better athlete jumps; if a new athlete's recorded personal best is lower than the current bar height, they are disqualified before even attempting the jump."
- BM25(Prerequisite)
- Inverted Index(Prerequisite)
- Top-K Retrieval(Component)
- MaxScore Algorithm(Alternative)
Conceptual Overview
A dynamic pruning algorithm for efficient top-k retrieval that uses upper-bound score estimates to skip candidate documents that cannot mathematically exceed the current minimum score of the top results. In RAG pipelines, it accelerates the initial retrieval phase by bypassing the exhaustive evaluation of low-relevance documents in large-scale indices.
Disambiguation
Not a logical 'AND' operator; it is a performance optimization for weighted term retrieval.
Visual Analog
A high-jump bar that is raised every time a better athlete jumps; if a new athlete's recorded personal best is lower than the current bar height, they are disqualified before even attempting the jump.