跳转至

ISO-RAG：用于检索增强生成等周噪声控制方法

文章背景与核心概要

检索增强 generation (RAG) 能够有效减少大语言模型（LLMs）的幻觉现象，但在面对多跳问答（QA）所需的复杂推理路径时，传统的稠密检索方法往往表现乏力。尽管基于图的 RAG 方法能够捕获多步关联，但由于存在充满噪声的全局图遍历，它们常常会遭遇语义漂移和高在线延迟的问题。

为了克服这些局限性，该论文引入了一种几何感知 RAG 框架——ISO-RAG（等周检索增强生成，ISOperimetric Retrieval-Augmented Generation）。通过将底层知识图谱投影到双曲庞加莱球（Poincaré ball）中并预计算节点级的等周轮廓（isoperimetric profiles），ISO-RAG 能够在检索过程中修剪伪边，并将搜索空间限制在严格局部的子图中。这种拓扑净化机制调节了个性化 PageRank (PPR) 扩散过程，从而实现了精确且低延迟的收敛。

ISO-RAG: Isoperimetric Noise Control for Retrieval-Augmented Generation

Summary

Retrieval-Augmented Generation (RAG) helps reduce hallucinations in Large Language Models (LLMs), but traditional dense retrieval methods struggle with the complex reasoning paths required for multi-hop Question Answering (QA). While graph-based RAG approaches can capture multi-step relationships, they often suffer from semantic drift and high online latency caused by noisy global graph traversals.

To overcome these limitations, the paper introduces ISO-RAG (ISOperimetric Retrieval-Augmented Generation), a geometry-aware RAG framework. By projecting the underlying knowledge graph into a hyperbolic Poincaré ball and precomputing node-wise isoperimetric profiles, ISO-RAG prunes spurious edges during retrieval and constrains the search space to a strictly localized subgraph. This topological purification regulates the Personalized PageRank (PPR) diffusion process, resulting in exact, low-latency convergence.

论文元数据 (Paper Metadata)

  • arXiv ID: arXiv:2609.00513 [cs.AI]
  • Primary Subject: Artificial Intelligence (cs.AI)
  • Submission Date: September 1, 2026
  • Authors:
  • Siyuan Zhang
  • Hanchen Wang
  • Dong Wen
  • Ying Zhang
  • Wenjie Zhang

摘要 (Abstract)

Retrieval-Augmented Generation (RAG) mitigates large language models (LLMs) hallucinations, yet conventional dense retrieval struggles with the complex reasoning paths of multi-hop question answering (QA). Graph-based RAG captures multi-step relationships but suffers from severe semantic drift and high online latency due to noisy global graph traversals. Thus, we propose ISO-RAG (ISOperimetric Retrieval-Augmented Generation), a geometry-aware RAG framework. By projecting the underlying knowledge graph into a hyperbolic Poincare ball to precompute node-wise isoperimetric profiles, ISO-RAG prunes spurious edges during retrieval, restricting the search space to a strictly localized subgraph. This topological purification regulates Personalized PageRank (PPR) diffusion driving the retrieval process, ensuring exact and low-latency convergence. Experiments on multi-hop QA benchmarks demonstrate that ISO-RAG outperforms state-of-the-art baselines by average absolute gains of 10.0% in retrieval recall and 4.3% in downstream exact match, achieving a superior accuracy-efficiency trade-off by fundamentally eliminating the latency bottleneck of global traversals.

核心亮点与性能表现 (Key Highlights & Performance)

  • Novel Geometry-Aware Approach: Projects knowledge graphs into a hyperbolic Poincaré ball to isolate meaningful subgraphs and precompute node-wise isoperimetric profiles.
  • Topological Purification: Prunes spurious edges and controls Personalized PageRank (PPR) diffusion to prevent semantic drift and remove the latency bottlenecks of global graph traversals.
  • Performance Gains: Achieves average absolute improvements of 10.0% in retrieval recall and 4.3% in downstream exact match across multi-hop QA benchmarks.