A configurable benchmark with four retrieval-noise types shows RAG accuracy drops sharply beyond 50% noise and that noise type, not just quantity, determines failure patterns.
Self-rag: Learn- ing to retrieve, generate, and critique through self-reflection
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Magic Mushroom: A Customizable Benchmark for Fine-grained Analysis of Retrieval Noise Erosion in RAG Systems
A configurable benchmark with four retrieval-noise types shows RAG accuracy drops sharply beyond 50% noise and that noise type, not just quantity, determines failure patterns.