Research map / Graph learning / Knowledge and temporal graphs
DocTr: Document Transformer for Structured Information Extraction in Documents
Haofu Liao, Aruni RoyChowdhury, Weijian Li, Ankan Bansal, Yuting Zhang, Zhuowen Tu, Ravi Kumar Satzoda, R. Manmatha, Vijay Mahadevan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use anchor word detection formulation instead of IOB tagging (raster scan text) (ablation) | DocTr backbone · CORD receipt parsing | F1: 93.2 -> 94.4 | Improved |
| use anchor word detection formulation instead of SPADE graph decoding (ablation) | DocTr backbone · CORD receipt parsing | F1: 93.0 -> 94.4 | Improved |
| switch from oracle to raster scan text order (ablation) | anchor word formulation vs IOB/SPADE · CORD receipt parsing | F1 drop: 0.6 drop vs 0.9 drop | Improved |
| fine-tune general-purpose models with the anchor word formulation instead of IOB tagging | LayoutLMv3 · CORD receipt parsing | F1: 92.2 -> 93.6 | Improved |
| use first and last words as anchors instead of first word only (ablation) | DocTr · CORD receipt parsing | F1: 94.2 -> 94.4 | Improved |
| use last word instead of first word as anchor (ablation) | DocTr · CORD receipt parsing | F1: 94.2 -> 94.1 | No clear effect |
| use name anchor as primary anchor instead of first entity of line item (ablation) | DocTr · CORD receipt parsing | F1: 94.0 -> 94.4 | Improved |
| add masked detection modeling pre-training instead of none (ablation) | DocTr · CORD receipt parsing | F1: 82.3 -> 94.4 | Improved |
| use MDM instead of MVLM pre-training (ablation) | DocTr · CORD receipt parsing | F1: 90.9 -> 94.4 | Improved |
| use MDM instead of MVLM pre-training (ablation) | DocTr · FUNSD entity linking | F1: 73.0 -> 73.9 | Improved |
| add vision encoder and VL-decoder to language encoder (ablation) | DocTr · CORD receipt parsing | F1: 92.3 -> 94.4 | Improved |
| add VL-decoder without vision encoder (ablation) | DocTr · CORD receipt parsing | F1: 92.3 -> 92.6 | No clear effect |
Closest papers by meaning
- Relation-Rich Visual Document Generator for Visual Information Extraction (CVPR 2025 · similarity 0.87)
- Span Graph Transformer for Document-Level Named Entity Recognition (AAAI 2024 · similarity 0.87)
- DocLayLLM: An Efficient Multi-modal Extension of Large Language Models for Text-rich Document Understanding (CVPR 2025 · similarity 0.85)
- An Autoregressive Text-to-Graph Framework for Joint Entity and Relation Extraction (AAAI 2024 · similarity 0.85)
- SAIL: Sample-Centric In-Context Learning for Document Information Extraction (AAAI 2025 · similarity 0.85)
- Revisiting Document-Level Relation Extraction with Context-Guided Link Prediction (AAAI 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
