Kernel SHAP and ContextCite approximate exact Shapley attribution for RAG documents accurately at low sample counts, but all methods systematically undervalue the document needed for multi-hop synthesis.
In: In- ternational Conference on Learning Representations (2021),https://openreview
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Source Attribution in Retrieval-Augmented Generation
Kernel SHAP and ContextCite approximate exact Shapley attribution for RAG documents accurately at low sample counts, but all methods systematically undervalue the document needed for multi-hop synthesis.