Reproduce “Doc-Researcher: A Unified System for Multimodal Document Parsing and Deep Research” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kuicai Dong, Shurui Huang, Fangda Ye, Wei Han, Zhi Zhang, Dexun Li, Wenjun Li, Qu Yang and 4 more
Huawei Technologies' Doc-Researcher is a unified system that integrates deep multimodal document parsing with an iterative multi-agent framework, enhancing AI's ability to conduct complex, multi-hop research on multimodal documents. The system achieves 50.6% accuracy on the newly introduced M4DocBench benchmark, significantly outperforming existing baselines.
Key insights
- Effective deep research on multimodal document collections fundamentally requires deep parsing that preserves multimodal integrity and layout, rather than relying on simplistic text extraction or screenshot processing.
- Adaptive and heterogeneous multimodal retrieval, combining both text and vision and adjusting retrieval granularity dynamically, is crucial for precisely locating relevant evidence.
- Iterative, multi-agent workflows are essential for handling the complexity of multi-hop reasoning, multi-document synthesis, and multi-turn interactions inherent in human-like research processes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.