Definition
Re-ranking in the context of RAG involves taking an initial set of retrieved documents or chunks and applying a secondary, more sophisticated relevance scoring mechanism. This refines the selection of the most pertinent information before it’s passed to the LLM.
Why it matters (in Poovi’s context)
Improves the precision of RAG by filtering out less relevant results from the initial retrieval, leading to more accurate and focused responses from the AI agent.
Key properties or components
- Secondary relevance scoring
- Refines initial retrieval
- Improves accuracy
- Often uses more complex models than initial retrieval
Contradictions or debates
None.