// the find
hymie122/RAG-Survey
Collecting awesome papers of RAG for AIGC. We propose a taxonomy of RAG foundations, enhancements, and applications in paper "Retrieval-Augmented Generation for AI-Generated Content: A Survey".
A link-dump companion repo for the paper "Retrieval-Augmented Generation for AI-Generated Content: A Survey" (arXiv 2402.19473) — it's not a library, there's no code, just a huge taxonomy of paper titles and arXiv/ACL/IEEE links organized by RAG technique and application modality. Useful for someone doing a literature review across RAG for text, code, audio, image, video, or 3D generation who wants a pre-built map of the space rather than a chatbot or retrieval framework to run.
The taxonomy itself is the value: papers are split into foundations (query-based, latent, logit-based, speculative RAG), enhancements (input/retriever/generator/result-level), and applications by modality, mirroring the survey's structure, which makes it far easier to navigate than a flat awesome-list. Coverage is genuinely wide — it pulls in RAG work from NLP, code generation, audio/image/video synthesis, and 3D generation, so it surfaces cross-domain papers you wouldn't find searching any single venue. It's tied to a peer-reviewed survey rather than being an ad-hoc collection, so the categorization has some rigor behind it.
Every entry is just a title and a link — no one-line summary of what each paper contributes or why it's placed in that subcategory, so you still have to open dozens of papers to figure out relevance. Several papers are duplicated across categories (REPLUG, EditSum, BashExplainer, Retrieval-Augmented Score Distillation all appear 2+ times) with no cross-referencing, which suggests the list was assembled by hand without dedup tooling. The README has broken/malformed image embeds (`<div aligncenter>` instead of proper HTML, missing closing tags) that render badly on GitHub. Last commit is August 2024, so it's missing well over a year of RAG research — for a fast-moving field like this, that's a significant and growing gap, and there's no changelog or version marker to tell you how stale a given section is.