Definition
The algorithmic process of determining the relevance between a user's input query and indexed document segments, typically achieved through semantic vector similarity, lexical overlap, or hybrid scoring methods to identify the optimal context for an LLM response. It involves mapping disparate text lengths into a shared embedding space where the proximity of the query vector to a passage vector denotes informational alignment.
Distinct from 'Query Expansion'; it focuses on the retrieval score calculation rather than modifying the query string itself.
"A high-precision magnet tuned to a specific frequency (query) that only attracts metallic filings (passages) tuned to that same frequency from a massive pile of debris."
Conceptual Overview
The algorithmic process of determining the relevance between a user's input query and indexed document segments, typically achieved through semantic vector similarity, lexical overlap, or hybrid scoring methods to identify the optimal context for an LLM response. It involves mapping disparate text lengths into a shared embedding space where the proximity of the query vector to a passage vector denotes informational alignment.
Disambiguation
Distinct from 'Query Expansion'; it focuses on the retrieval score calculation rather than modifying the query string itself.
Visual Analog
A high-precision magnet tuned to a specific frequency (query) that only attracts metallic filings (passages) tuned to that same frequency from a massive pile of debris.