Modular Pipeline Converts Historical Tables to Knowledge Graphs
A new modular, provenance-aware pipeline transforms handwritten archival tables into Knowledge Graphs (KGs) through three stages: table reconstruction, information extraction, and KG construction. The approach exposes intermediate representations for human inspection, evaluation, and correction, addressing opacity in end-to-end AI systems. Data provenance is systematically integrated at every stage, ensuring traceability of all extracted entities and literals. The pipeline supports human-AI collaboration by enabling oversight and trust in the conversion of complex multimodal historical data.
Key facts
- Pipeline converts handwritten tabular images into Knowledge Graphs
- Three stages: table reconstruction, information extraction, KG construction
- Provenance-aware design integrates data provenance at every stage
- Exposes intermediate representations for human inspection and correction
- Addresses opacity of end-to-end AI implementations
- Supports human-AI collaboration and trust
- Focuses on historical archival tables with rich information
- arXiv preprint arXiv:2605.08222v1
Entities
Institutions
- arXiv