FRESCO benchmark reveals re-rankers consistently favor older Wikipedia documents even when factually obsolete, and instruction optimization yields up to 27% gains on evolving-knowledge tasks while preserving non-evolving performance.
In addition, each passage is summarized with a LLM (LLaMA-3.3-70B-Inst here)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
FRESCO: Benchmarking and Optimizing Re-rankers for Evolving Semantic Conflict in Retrieval-Augmented Generation
FRESCO benchmark reveals re-rankers consistently favor older Wikipedia documents even when factually obsolete, and instruction optimization yields up to 27% gains on evolving-knowledge tasks while preserving non-evolving performance.