CrAM: Credibility-Aware Attention Modification in LLMs for Combating Misinformation in RAG

Boyi Deng, Wenjie Wang, Fengbin Zhu, Qifan Wang, Fuli Feng
6/17/2024
cs.CL

No official code found. Best implementation: Aatrox103/CrAM (5 stars, last commit 2025-10-12).

Abstract

Retrieval-Augmented Generation (RAG) can alleviate hallucinations of Large Language Models (LLMs) by referencing external documents. However, the misinformation in external documents may mislead LLMs' generation. To address this issue, we explore the task of "credibility-aware RAG", in which LLMs automatically adjust the influence of retrieved documents based on their credibility scores to counteract misinformation. To this end, we introduce a plug-and-play method named $\textbf{Cr}$edibility-aware $\textbf{A}$ttention $\textbf{M}$odification (CrAM). CrAM identifies influential attention heads in LLMs and adjusts their attention weights based on the credibility of the documents, thereby reducing the impact of low-credibility documents. Experiments on Natual Questions and TriviaQA using Llama2-13B, Llama3-8B, and Qwen1.5-7B show that CrAM improves the RAG performance of LLMs against misinformation pollution by over 20%, even surpassing supervised fine-tuning methods.

View on arXivView PDF

Code Implementations(1)

Aatrox103/CrAMCommunity100%
53Shell, PythonJun 14, 20241 years ago

Similar Papers

Not All Contexts Are Equal: Teaching LLMs Credibility-aware Generation

87% match1 repo222024

Enhancing LLM Factual Accuracy with RAG to Counter Hallucinations: A Case Study on Domain-Specific Queries in Private Knowledge-Bases

86% match1 repo452024

Tackling the Inherent Difficulty of Noise Filtering in RAG

85% match2026

Addressing Corpus Knowledge Poisoning Attacks on RAG Using Sparse Attention

85% match2026

Finetune-RAG: Fine-Tuning Language Models to Resist Hallucination in Retrieval-Augmented Generation

85% match1 repo472025

Astute RAG: Overcoming Imperfect Retrieval Augmentation and Knowledge Conflicts for Large Language Models

83% match2024

Cite this paper

@article{deng2024cram,
  title  = {CrAM: Credibility-Aware Attention Modification in LLMs for Combating Misinformation in RAG},
  author = {Boyi Deng and Wenjie Wang and Fengbin Zhu and Qifan Wang and Fuli Feng},
  year   = {2024},
  eprint = {2406.11497},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2406.11497v3}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2406.11497.svg)](https://codeofpaper.com/paper/2406.11497)

Discussion