> For the complete documentation index, see [llms.txt](https://yeasy.gitbook.io/context_engineering_guide/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://yeasy.gitbook.io/context_engineering_guide/fu-lu/references.md).

# 附录 D：参考文献

本文档列出了书中引用的学术论文和关键技术报告，供读者深入研读。

## 学术论文

### 基础架构

* **Attention Is All You Need**: Vaswani, A., et al. (2017). [Attention Is All You Need](https://arxiv.org/abs/1706.03762). NIPS.
  * *解读：Transformer 架构的奠基之作，理解 Self-Attention 机制的必读文献。*

### 检索增强生成

* **RAG**: Lewis, P., et al. (2020). [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401). NeurIPS.
  * *解读：RAG 概念的开山之作，提出了将检索器与生成器端到端联合训练的范式。*
* **Self-RAG**: Asai, A., et al. (2023). [Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection](https://arxiv.org/abs/2310.11511). ICLR 2024.
  * *解读：让模型自主决定何时检索、并对自身生成结果进行反思与评估，是 Agentic RAG 的重要基础。*
* **CRAG**: Yan, S., et al. (2024). [Corrective Retrieval Augmented Generation](https://arxiv.org/abs/2401.15884).
  * *解读：引入检索评估器与纠正机制，当检索结果不可靠时触发查询改写或网页搜索，提升 RAG 鲁棒性。*
* **REPLUG**: Shi, W., et al. (2023). [REPLUG: Retrieval-Augmented Black-Box Language Models](https://arxiv.org/abs/2301.12652).
  * *解读：探讨了如何将检索器视为黑盒模型的插件进行优化。*

### 长上下文与注意力

* **Lost in the Middle**: Liu, N. F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F., & Liang, P. (2023). [Lost in the Middle: How Language Models Use Long Contexts](https://arxiv.org/abs/2307.03172). arXiv preprint arXiv:2307.03172.
  * *解读：揭示了部分模型在处理长上下文时存在位置偏置；现代系统仍需用任务级评测验证，而不能把该结论机械套用到所有模型。*
* **RULER**: Hsieh, C. P., et al. (2024). [RULER: What's the Real Context Size of Your Long-Context Language Models?](https://arxiv.org/abs/2404.06654).
  * *解读：指出单针 NIAH 只能覆盖较浅层的长上下文检索，扩展到多针、聚合和追踪等更难任务。*
* **Found in the Middle**: Hsieh, C.-Y., et al. (2024). [Found in the Middle: Calibrating Positional Attention Bias Improves Long Context Utilization](https://research.google/pubs/found-in-the-middle-calibrating-positional-attention-bias-improves-long-context-utilization/).
  * *解读：从位置注意力偏置角度分析并缓解长上下文中间位置利用不稳定的问题。*

### 知识图谱与 Graph RAG

* **Graph RAG**: Edge, D., et al. (2024). [From Local to Global: A Graph RAG Approach to Query-Focused Summarization](https://arxiv.org/abs/2404.16130).
  * *解读：微软提出的 Graph RAG 方法，通过构建社区层级的知识图谱来支持全局性摘要问答，突破了传统 RAG 在全局查询上的局限。*

### 智能体与记忆

* **MemGPT**: Packer, C., et al. (2023). [MemGPT: Towards LLMs as Operating Systems](https://arxiv.org/abs/2310.08560).
  * *解读：提出将上下文窗口视为“主内存”、外部存储视为“磁盘”的操作系统隐喻，通过虚拟上下文管理实现无限对话记忆。*
* **MemOS**: Li, Z., et al. (2025). [MemOS: A Memory OS for AI System](https://arxiv.org/abs/2507.03724).
  * *解读：以 MemCube（记忆立方体）为基本单元封装记忆内容与元数据，统一管理明文、激活、参数三类记忆的表示、调度与演化，并采用接口层、操作层、基础设施层的三层架构。*
* **Generative Agents**: Park, J. S., et al. (2023). [Generative Agents: Interactive Simulacra of Human Behavior](https://arxiv.org/abs/2304.03442). UIST 2023.
  * *解读：斯坦福的生成式智能体研究，展示了如何通过反思（Reflection）和记忆检索构建具有可信行为的智能体。*
* **Reflexion**: Shinn, N., et al. (2023). [Reflexion: Language Agents with Verbal Reinforcement Learning](https://arxiv.org/abs/2303.11366). NeurIPS 2023.
  * *解读：让智能体通过语言化的自我反思进行学习，将失败经验转化为长期记忆，提升后续任务的成功率。*
* **ReAct**: Yao, S., et al. (2023). [ReAct: Synergizing Reasoning and Acting in Language Models](https://arxiv.org/abs/2210.03629). ICLR 2023.
  * *解读：提出“思考-行动-观察”的交替循环模式，是现代智能体架构的核心范式。*
* **Agentic Reasoning**: Wei, T., et al. (2026). [Agentic Reasoning for Large Language Models](https://arxiv.org/abs/2601.12538). arXiv preprint arXiv:2601.12538. DOI: 10.48550/arXiv.2601.12538.
  * *解读：系统梳理基础智能体推理、自我进化推理和集体多智能体推理三层框架，适合作为第九章高级智能体上下文管理的理论背景。*
* **OpenAI Dreaming**: OpenAI. (2026). [Dreaming: Better memory for a more helpful ChatGPT](https://openai.com/index/chatgpt-memory-dreaming/).
  * *解读：ChatGPT 记忆从“保存的记忆”（显式写入）演进到后台自动综合的工程实践；提出“带上下文前进 / 遵循偏好 / 随时间保持时效”三条记忆评测目标，是第九章“做梦”机制的真实产品对照。*

### 工具使用

* **Toolformer**: Schick, T., et al. (2023). [Toolformer: Language Models Can Teach Themselves to Use Tools](https://arxiv.org/abs/2302.04761).
  * *解读：展示了语言模型如何自主学习调用外部工具（计算器、搜索引擎等），是理解工具调用机制的重要文献。*

## 技术报告与基准测试

* **LongBench**: Bai, Y., et al. (2023). [LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding](https://arxiv.org/abs/2308.14508).
  * *解读：专门针对长上下文能力的评测基准。*
* **RAGAS**: Es, S., et al. (2023). [RAGAS: Automated Evaluation of Retrieval Augmented Generation](https://arxiv.org/abs/2309.15217).
  * *解读：RAG 系统的自动化评估框架，提出了忠实度、答案相关性、上下文精确度等核心指标。*
* **MTEB**: Muennighoff, N., et al. (2023). [MTEB: Massive Text Embedding Benchmark](https://arxiv.org/abs/2210.07316). EACL 2023.
  * *解读：大规模文本嵌入基准，覆盖 8 类任务、58 个数据集，是嵌入模型选型的权威排行榜。*
* **ColBERT**: Khattab, O., & Zaharia, M. (2020). [ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT](https://arxiv.org/abs/2004.12832).
  * *解读：late interaction 检索的代表工作，通过查询和文档的细粒度交互提升检索质量。*
* **Late Chunking**: Günther, M., Mohr, I., Williams, D. J., Wang, B., & Xiao, H. (2024). [Late Chunking: Contextual Chunk Embeddings Using Long-Context Embedding Models](https://arxiv.org/abs/2409.04701).
  * *解读：先用长上下文嵌入模型编码文档，再对片段池化，缓解先切块带来的上下文丢失。*


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://yeasy.gitbook.io/context_engineering_guide/fu-lu/references.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.