arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2510.09710cs.CLcs.AI
收藏

SeCon-RAG:一种用于可信RAG的两阶段语义过滤与无冲突框架

SeCon-RAG: A Two-Stage Semantic Filtering and Conflict-Free Framework for Trustworthy RAG

Institute of Software Chinese Academy of Sciences Beijing China(中国科学院软件研究所);Key Laboratory of System Software (Chinese Academy of Sciences) and State Key Laboratory of Computer Science, Institute of Software, Chinese Academy of Sciences, Beijing, China(中国科学院系统软件重点实验室和计算机科学国家重点实验室);University of Chinese Academy of Sciences, Beijing, China(中国科学院大学);Northeast University China(东北大学);Institute of Ai For industries Nanjing China(人工智能产业研究院);Southwest University China(西南大学);Nanyang Technological University Singapore(新加坡南洋理工大学);Alibaba China(阿里巴巴(中国))

Xiaonan Si, Meilin Zhu, Simeng Qin, Lijia Yu, Lijun Zhang, Shuaitong Liu, Xinfeng Li, Ranjie Duan, Yang Liu, Xiaojun Jia

AI总结:

针对RAG系统易受语料库污染的问题,提出SeCon-RAG两阶段语义过滤与无冲突框架,通过EIRE引导的联合过滤和冲突感知过滤保留有用知识,提升生成鲁棒性与输出可信度,实验优于现有防御方法。

摘要:

检索增强生成(RAG)系统通过外部知识增强大语言模型(LLM)的能力,但容易受到语料库投毒和污染攻击,这些攻击可能损害输出完整性。现有的防御方法往往采用激进过滤,导致有价值信息的非必要丢失,并降低生成的可靠性。为解决这一问题,我们提出了一种用于可信RAG的两阶段语义过滤与无冲突框架。在第一阶段,我们执行由实体-意图-关系提取器(EIRE)引导的语义过滤与基于聚类的过滤联合过滤器。EIRE从用户查询和过滤后的文档中提取实体、潜在目标和实体关系,对其语义相关性进行评分,并将有价值的文档有选择地添加到干净的检索数据库中。在第二阶段,我们提出了一个由EIRE引导的冲突感知过滤模块,该模块在最终答案生成之前分析查询、候选答案和检索知识之间的语义一致性,过滤掉可能误导模型的内部和外部矛盾。通过这一两阶段过程,SeCon-RAG在缓解冲突污染的同时有效保留有用知识,在生成鲁棒性和输出可信度方面均取得了显著改进。跨多种LLM和数据集的广泛实验表明,所提出的SeCon-RAG明显优于最先进的防御方法。

英文摘要:

Retrieval-augmented generation (RAG) systems enhance large language models (LLMs) with external knowledge but are vulnerable to corpus poisoning and contamination attacks, which can compromise output integrity. Existing defenses often apply aggressive filtering, leading to unnecessary loss of valuable information and reduced reliability in generation. To address this problem, we propose a two-stage semantic filtering and conflict-free framework for trustworthy RAG. In the first stage, we perform a joint filter with semantic and cluster-based filtering which is guided by the Entity-intent-relation extractor (EIRE). EIRE extracts entities, latent objectives, and entity relations from both the user query and filtered documents, scores their semantic relevance, and selectively adds valuable documents into the clean retrieval database. In the second stage, we proposed an EIRE-guided conflict-aware filtering module, which analyzes semantic consistency between the query, candidate answers, and retrieved knowledge before final answer generation, filtering out internal and external contradictions that could mislead the model. Through this two-stage process, SeCon-RAG effectively preserves useful knowledge while mitigating conflict contamination, achieving significant improvements in both generation robustness and output trustworthiness. Extensive experiments across various LLMs and datasets demonstrate that the proposed SeCon-RAG markedly outperforms state-of-the-art defense methods.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑