Scaling Beyond Context: A Survey of Multimodal Retrieval-Augmented Generation for Document Understanding

ACL 2026 Main Conference

Sensen Gao^1*, Shanshan Zhao^2†, Xu Jiang³, Lunhao Duan^4*, Yong Xien Chng^3*,
Qing-Guo Chen², Weihua Luo², Kaifu Zhang², Jia-Wang Bian¹, Mingming Gong^1,5†

¹MBZUAI   ²Alibaba International Digital Commerce Group   ³Tsinghua University   ⁴Wuhan University   ⁵University of Melbourne

_{* Work done during an internship at Alibaba International Digital Commerce Group. † Corresponding authors.}

---

This repository maintains a curated list of methods, datasets, and benchmarks for Multimodal Retrieval-Augmented Generation (RAG) in Document Understanding, based on our ACL 2026 survey. We will keep updating this list. Feel free to open an issue or PR if we missed any relevant work!

Updates

• Ongoing - We keep this list actively maintained! If you have a recent paper, or notice any relevant work we missed, feel free to contact us or open an issue/PR — we'll include it in a future update.
• 2026.04 - Paper accepted at ACL 2026 Main Conference!
• 2025.10 - Paper available on arXiv.

Table of Contents

• Overview
• Methods
  • Open-Domain Methods
  • Closed-Domain Methods
  • Graph-based Methods
  • Agent-based Methods
• Datasets & Benchmarks
  • Document Understanding Datasets
  • Multimodal RAG Benchmarks
• Citation

Overview

A taxonomy of multimodal RAG for document understanding: retrieval domain (open vs. closed), retrieval modality (image vs. image+text), retrieval granularity (page- vs. element-level), and hybrid graph-/agent-based enhancements.

Methods

Open-Domain Methods

Closed-Domain Methods

Graph-based Methods

Agent-based Methods

Datasets & Benchmarks

Document Understanding Datasets

Multimodal RAG Benchmarks

Citation

If you find this survey useful, please cite our paper:

@article{gao2025scaling,
      title={Scaling Beyond Context: A Survey of Multimodal Retrieval-Augmented Generation for Document Understanding},
      author={Sensen Gao and Shanshan Zhao and Xu Jiang and Lunhao Duan and Yong Xien Chng and Qing-Guo Chen and Weihua Luo and Kaifu Zhang and Jia-Wang Bian and Mingming Gong},
      year={2025},
      eprint={2510.15253},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2510.15253},
}

Contact

For questions or suggestions, feel free to open an issue or contact Sensen Gao.