Reproduce “HRDoc: Dataset and Baseline Method Toward Hierarchical Reconstruction of Document Structures” 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
Jiefeng Ma, Jun Du, Pengfei Hu, Zhenrong Zhang, Jianshu Zhang, Huihui Zhu, Cong Liu
The HRDoc paper defines the task of Hierarchical Reconstruction of Document Structures, providing a new large-scale dataset (HRDoc) for this purpose. It also introduces the Document Structure Parsing System (DSPS) model, which leverages multi-modal features and a structure-aware decoder to reconstruct multi-page document hierarchies, outperforming baselines in both element classification and hierarchical reconstruction.
Key insights
- Modeling multi-page documents requires considering both element content and their complex, hierarchical inter-relationships beyond single-page analysis.
- Integrating multi-modal features (textual, visual, layout) significantly improves the understanding and reconstruction of document structures.
- Domain knowledge, such as valid parent-child relationships between element types, can be effectively incorporated (via a soft mask) to guide the hierarchical reconstruction process.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.