Retrieval-Augmented Generation: Research Digest

Literature Digest: Retrieval‑Augmented Generation

Retrieval‑augmented generation (RAG) is a paradigm that couples a generative language model with an external retrieval system, enabling the model to condition its outputs on dynamically retrieved documents rather than relying solely on parametric knowledge. This approach has become central for knowledge‑intensive NLP tasks, where access to up‑to‑date, factual evidence is critical.

Foundations and Formulations

The seminal work “Retrieval‑augmented generation for knowledge‑intensive NLP tasks” introduces RAG models that combine a pre‑trained seq2seq generator with a dense vector index of Wikipedia, accessed via a pre‑trained neural retriever. It proposes two variants: one that conditions on the same retrieved passages for the entire output sequence, and another that can switch passages per token. Joint fine‑tuning of retriever and generator is shown to improve both retrieval and generation quality, yielding state‑of‑the‑art results on open‑domain QA and more factual, diverse language generation compared to parametric‑only baselines.

Survey and Design Landscapes

Subsequent surveys, such as “Retrieval‑augmented generation for large language models: A survey” and “Retrieval‑augmented generation for AI‑generated content: A survey,” systematize RAG across retrieval, generation, and integration components, highlighting how RAG improves accuracy, credibility, and controllability of AI‑generated content. Works like “Searching for best practices in retrieval‑augmented generation” and “Graph retrieval‑augmented generation: A survey” further explore design choices, including retrieval strategies, indexing schemes, and graph‑structured knowledge bases, underscoring that RAG can be adapted to domains ranging from technical support to legal reasoning.

Open Problems

• Dynamic and active retrieval: Methods such as “Active retrieval augmented generation” and “Chain‑of‑retrieval augmented generation” point to the need for principled policies that decide when and what to retrieve during generation, rather than relying on static pre‑retrieval.
• Evaluation and metrics: “Evaluation of retrieval‑augmented generation: A survey” stresses that current metrics for retrieval and generation quality are fragmented and often misaligned with downstream utility, calling for more holistic, task‑aware evaluation frameworks.
• Global and structured knowledge: “Graph retrieval‑augmented generation: A survey” notes that standard RAG typically retrieves isolated passages, missing global structure; integrating richer graph‑based knowledge while preserving efficiency remains an open challenge.

Key papers

1. Retrieval-augmented generation for knowledge-intensive nlp tasks — P Lewis,E Perez,A Piktus,F Petroni…
2. Retrieval-augmented generation for large language models: A survey — Y Gao,Y Xiong,X Gao,K Jia,J Pan,Y Bi,Y Dai…
3. Searching for best practices in retrieval-augmented generation — X Wang,Z Wang,X Gao,F Zhang,Y Wu…
4. Active retrieval augmented generation — Z Jiang,FF Xu,L Gao,Z Sun,Q Liu…
5. Chain-of-retrieval augmented generation — L Wang,H Chen,N Yang,X Huang…
6. Retrieval-augmented generation for ai-generated content: A survey — P Zhao,H Zhang,Q Yu,Z Wang,Y Geng,F Fu…
7. Graph retrieval-augmented generation: A survey — B Peng,Y Zhu,Y Liu,X Bo,H Shi,C Hong…
8. Evaluation of retrieval-augmented generation: A survey — H Yu,A Gan,K Zhang,S Tong,Q Liu,Z Liu
9. Retrieval-augmented generation for natural language processing: A survey — S Wu,Y Xiong,Y Cui,H Wu,C Chen,Y Yuan…
10. Longrag: Enhancing retrieval-augmented generation with long-context llms — Z Jiang,X Ma,W Chen

Papers via the AISA Scholar API; synthesis by the AISA LLM layer. 2026-06-23.