Definition
A numerical value representing the semantic or statistical alignment between a user's query and a retrieved document chunk, typically calculated using cosine similarity or BM25. In RAG pipelines, it acts as a gatekeeper to determine which context fragments are high-quality enough to be injected into the LLM's context window, balancing the trade-off between comprehensive context and noise reduction.
Measures mathematical proximity in vector space, not the factual accuracy or 'truthfulness' of the information.
"A Metal Detector's Beep Intensity: the louder the signal (higher score), the closer the retrieved fragment is to the user's specific informational need."
- Cosine Similarity(Mathematical Method)
- Vector Embedding(Prerequisite)
- Thresholding(Optimization Process)
- Cross-Encoder(Refinement Component)
Conceptual Overview
A numerical value representing the semantic or statistical alignment between a user's query and a retrieved document chunk, typically calculated using cosine similarity or BM25. In RAG pipelines, it acts as a gatekeeper to determine which context fragments are high-quality enough to be injected into the LLM's context window, balancing the trade-off between comprehensive context and noise reduction.
Disambiguation
Measures mathematical proximity in vector space, not the factual accuracy or 'truthfulness' of the information.
Visual Analog
A Metal Detector's Beep Intensity: the louder the signal (higher score), the closer the retrieved fragment is to the user's specific informational need.