AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Attributing Response to Context: A Jensen–Shannon Divergence Driven Mechanistic Study of Context Attribution in Retrieval-Augmented Generation

Ruizhe Li, Chen Chen, Yuchen Hu, Yanjun Gao, Xi Wang, Emine Yilmaz

ICLR 2026 · Large language models › Retrieval and knowledge · approach: RAG robustness and evaluation · 15 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
attribute with ARC-JSD instead of ContextCite, ALTI-Logit and MIRAGE baselinesRAG LLMs (Qwen2, Gemma2) · MuSiQuetop-1 attribution accuracy: about 10.7% average improvementImproved
use ARC-JSD instead of gradient or surrogate-based attributionRAG attribution · compute-accuracy trade-offcompute: up to 3-fold speedupImproved
gate top-5 JSD-located attention heads and MLPs with semantic gain maskQwen2-7B-IT · Hotpot QA (200 failed samples)hallucination rate: 13.4% -> 8.2%Improved
gate top-5 JSD-located componentsQwen2-7B-IT · Hotpot QA (200 failed samples)Pass@1 Factual F1: 76.1 -> 75.9No clear effect
gate random 5 attention and MLP components instead of JSD-located ones (ablation)Qwen2-7B-IT · Hotpot QA (200 failed samples)Pass@1 Factual F1: 76.1 -> 69.4Hurt
mask top-10 JSD-located attention heads rather than random heads (ablation)ARC-JSD · all RAG models, all datasetsaverage JSD: 1.53 -> 2.23Improved
use JSD instead of KL, Wasserstein, TV, MMD as divergence (ablation)ARC-JSD · MuSiQuecompute-accuracy trade-offImproved
ablate attention heads or MLP layers with higher JSD score (ablation)Qwen2-1.5B-IT · TyDi QAattribution accuracy dropHurt
compare ContextCite with 32 forward calls vs ARC-JSDContextCite-32 · MuSiQueattribution accuracyHurt
apply ARC-JSD to LLaMA-3.1-8B-IT and Qwen3-Next-80B-A3B-ITARC-JSD · MuSiQuecompute-accuracy trade-offImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy