DOSA: A Tree-Guided Framework for Document Structure Analysis
A new framework called DOSA (Document Structure Analyzer) has been proposed for analyzing long document structures. It processes documents chunk-by-chunk, fusing visual, textual, and layout features to predict hierarchical and ordering relations among page objects. These predictions incrementally build a semantic tree that guides inference on subsequent chunks, addressing challenges in multi-page documents with long-range dependencies and heterogeneous layouts. The framework is described in a paper on arXiv (2607.22679).
Key facts
- DOSA stands for Document Structure Analyzer
- It is a tree-guided, self-regressive framework
- Processes documents chunk-by-chunk
- Fuses visual, textual, and layout features
- Predicts hierarchical and ordering relations
- Constructs a semantic tree incrementally
- The tree guides inference on subsequent chunks
- Paper available on arXiv: 2607.22679
Entities
Institutions
- arXiv