用图结构重排序，防伪造文档误导大模型。
GRADA: Graph-based Reranking against Adversarial Documents Attack
- 构建文档相似性图，识别并剔除伪装成查询的恶意文档。
- 在Natural Questions数据集上，攻击成功率降低80%且准确率损失极小。
- 适合防御大模型检索增强系统被恶意干扰的研究者使用。
检索增强生成（RAG）框架通过整合外部文档知识提升大语言模型（LLM）的准确性，克服模型静态知识的局限。然而，这类系统易受对抗攻击，攻击者通过引入语义上与查询相似但与正常文档不相关的恶意文档，操纵检索过程。值得注意的是，这些恶意文档虽与查询相似，但与检索集中良性文档的相似性较弱。为此，我们提出一种简单而有效的图结构重排序方法GRADA，旨在保持检索质量的同时显著降低攻击成功率。实验在五种LLM（GPT-3.5-Turbo、GPT-4o、Llama3.1-8b、Llama3.1-70b、Qwen2.5-7b）上进行，使用三个数据集评估性能，结果表明，在Natural Questions数据集上攻击成功率最高可降低80%，同时准确率损失微乎其微。
原文摘要 · Abstract (English)
Retrieval Augmented Generation (RAG) frameworks improve the accuracy of large language models (LLMs) by integrating external knowledge from retrieved documents, thereby overcoming the limitations of models' static intrinsic knowledge. However, these systems are susceptible to adversarial attacks that manipulate the retrieval process by introducing documents that are adversarial yet semantically similar to the query. Notably, while these adversarial documents resemble the query, they exhibit weak similarity to benign documents in the retrieval set. Thus, we propose a simple yet effective Graph-based Reranking against Adversarial Document Attacks (GRADA) framework aiming at preserving retrieval quality while significantly reducing the success of adversaries. Our study evaluates the effectiveness of our approach through experiments conducted on five LLMs: GPT-3.5-Turbo, GPT-4o, Llama3.1-8b, Llama3.1-70b, and Qwen2.5-7b. We use three datasets to assess performance, with results from the Natural Questions dataset demonstrating up to an 80% reduction in attack success rates while maintaining minimal loss in accuracy.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。