跳转至

被遗忘的历史还是经受住时间考验?从信息检索视角看 RAG 的回顾与展望

文章背景与核心概要

尽管检索增强生成(RAG)通常被认为是为了解决大语言模型(LLM)局限性而诞生的最新范式,但本文指出,这种观点在历史脉络上并不完整。RAG 的核心机制——包括检索与生成的集成、知识增强、答案验证以及迭代式查询优化,早在 2000 年代初期就已经在信息检索(IR)和问答(QA)研究中得到了广泛研究。

作者将现代 RAG 的思想渊源追溯到经典的信息检索和问答基础,认为其之所以表现出“新颖性”,主要是由于社区割裂、术语变迁以及近因偏见。通过将大语言模型重新定位为一个新的界面层,而不是检索增强智能的起点,本文凸显了许多未被充分利用的历史框架(如用户建模和查询优化),这些框架可以指导下一代 RAG 系统的开发,并促进不同技术社区之间的深度融合。


📌 Summary (摘要与总结)

While Retrieval-Augmented Generation (RAG) is commonly perceived as a recent paradigm created to address the limitations of Large Language Models (LLMs), this paper argues that such a view is historically incomplete. The core mechanics of RAG—including retrieval-generation integration, knowledge augmentation, answer validation, and iterative query refinement—were extensively studied in Information Retrieval (IR) and Question Answering (QA) research as early as the 2000s.

The authors trace the intellectual lineage of modern RAG back to classical IR and QA foundations, attributing its apparent novelty to community fragmentation, changing terminology, and recency bias. By reframing LLMs as a new interface layer rather than the origin point of retrieval-augmented intelligence, the paper highlights underutilized historical frameworks (such as user modeling and query refinement) that can guide the development of next-generation RAG systems and foster cross-community integration.

尽管检索增强生成(RAG)普遍被视为一种为解决大语言模型(LLM)局限性而生的新兴范式——一种将其输出扎根于外部知识的机制,但从更广阔的历史背景来看,这种观点并不完整。在本文中,我们认为 RAG 背后的核心理念并非全新事物:诸如检索与语言生成的集成、知识增强、答案验证以及迭代式查询(或提示词)优化等基础概念,早在 2000 年代初期(远在大语言模型出现之前)就已经在信息检索(IR)和问答(QA)研究中得到了研究和实现。

我们通过系统地将现代 RAG 和智能体 RAG(Agentic RAG)的思想渊源追溯到其经典的信息检索和问答前身,并探讨了为什么这种历史延续性被忽视——这是社区割裂、术语变迁以及快节奏领域特有的近因偏见共同作用的结果。与其将大语言模型视为检索增强智能的起点,我们建议将其视为数十年前问答架构之上的一个新界面层。这种重新定位不仅具有历史意义:通过将 RAG 置于更长远的信息检索研究轨迹中,我们挖掘了先前未被充分利用的前期工作(涉及用户建模、答案验证和查询优化),这些工作可以直接为下一代 RAG 的设计提供参考,减少无意的重复发明,并促进真正的跨社区融合。



📖 Abstract (论文摘要)

Retrieval-Augmented Generation (RAG) is widely regarded as a novel paradigm born from the limitations of large language models (LLMs)—a mechanism to ground their outputs in external knowledge. This view, however, is incomplete when considered within a broader historical context. In this paper, we argue that the core ideas underlying RAG are not new: foundational concepts such as integrating retrieval and language generation, knowledge augmentation, answer verification, and iterative query (or prompt) refinement had already been studied and instantiated in information retrieval (IR) and question answering (QA) research dating back to the early 2000s, well before the emergence of LLMs.

We make this case by systematically tracing the intellectual lineage of modern RAG and Agentic RAG back to their classical IR and QA antecedents, and examining why this continuity has gone under-recognized — a consequence of community fragmentation, shifting terminology, and the recency bias endemic to fast-moving fields. Rather than treating LLMs as the origin point of retrieval-augmented intelligence, we propose viewing them as a new interface layer atop a decades-old QA architecture. This reframing is not merely historical: by situating RAG within the longer trajectory of IR research, we surface underutilized prior work — on user modeling, answer validation, and query refinement — that can directly inform next-generation RAG design, reducing unintentional rediscovery and fostering genuine cross-community integration.

检索增强生成(RAG)被普遍认为是由大语言模型(LLM)的局限性孕育出的一种新颖范式——一种将其输出根植于外部知识的机制。然而,若将其置于更广泛的历史背景中来看,这种观点是不完整的。在本文中,我们认为 RAG 背后的核心思想并非新事物:诸如检索与语言生成相集成、知识增强、答案验证以及迭代查询(或提示词)优化等基础概念,早在 2000 年代初期、即 LLM 出现之前很久,就已经在信息检索(IR)和问答(QA)研究中得到了深入研究和实例实现。

我们通过系统地将现代 RAG 和智能体 RAG 的知识谱系追溯到其古典 IR 和 QA 的前身,并探讨了这种连续性为何未被充分认识——这是社区碎片化、术语演变以及快节奏领域特有的近因偏见所导致的结果。与其将 LLM 视为检索增强智能的起点,我们建议将其视为数十年历史的 QA 架构之上的一个新型接口层。这种重新审视不仅仅具有历史意义:通过将 RAG 置于更长期的 IR 研究轨迹中,我们浮现出了那些未被充分利用的前期成果(关于用户建模、答案验证和查询优化),这些成果可以直接为下一代 RAG 的设计提供信息,减少无意识的重复发现,并促进真正的跨社群融合。