Definition
A numerical value representing the mathematical proximity between a query vector and a candidate document embedding in a high-dimensional vector space. In RAG, it serves as the primary filter for context retrieval, where higher scores indicate greater semantic alignment, though architectural trade-offs must be made between strict thresholds (reducing noise) and lenient thresholds (improving recall).
Measures semantic distance in latent space rather than literal keyword frequency or exact character matches.
"A sonar ping measuring the precise distance between a search vessel and various objects on the ocean floor to identify the most likely wreckage."
- Vector Embeddings(Prerequisite)
- Cosine Similarity(Mathematical Method)
- Top-K Retrieval(Dependent Parameter)
- Euclidean Distance (L2)(Alternative Metric)
Conceptual Overview
A numerical value representing the mathematical proximity between a query vector and a candidate document embedding in a high-dimensional vector space. In RAG, it serves as the primary filter for context retrieval, where higher scores indicate greater semantic alignment, though architectural trade-offs must be made between strict thresholds (reducing noise) and lenient thresholds (improving recall).
Disambiguation
Measures semantic distance in latent space rather than literal keyword frequency or exact character matches.
Visual Analog
A sonar ping measuring the precise distance between a search vessel and various objects on the ocean floor to identify the most likely wreckage.