Standardized re-evaluation of CRS methods on ReDial shows nearly 50% of reported accuracy stems from repetition shortcuts absent in novelty-focused tests, with gains driven more by LLM backbone than architectures and recall overstating effectiveness.
Title resolution pending
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
-
A Standardized Re-evaluation of Conversational Recommender Systems on the ReDial Dataset
Standardized re-evaluation of CRS methods on ReDial shows nearly 50% of reported accuracy stems from repetition shortcuts absent in novelty-focused tests, with gains driven more by LLM backbone than architectures and recall overstating effectiveness.