A metagraph is an intermediate graph layer that stores unmerged entity mentions, their extracted relationships, and links back to the pages where they originated. It separates extraction from normalization and entity resolution before constructing a final knowledge graph. section 6.1.1 section 6.1.2
A practical metagraph pipeline:
- Chunk long documents, using page-level splitting as a simple strategy. section 6.1.2
- Create an Entity node for every extracted mention instead of merging mentions immediately. section 6.1.2
- Link entities to their originating pages and connect related mentions with typed relationships. section 6.1.1 section 6.1.2
- Use graph structure, string similarity, and domain rules to support later entity resolution. section 6.1.4
Diary ingestion
A 150-page subset of Warren Weaver’s 1939 diary is split by page. Each page produces unmerged entity mentions and relationships, while OCR errors remain visible for later normalization and resolution. section 6.1.2
Figure 6.2 Simplified KG schema of the Rockefeller Archive Center project
Use the schema view to trace the progression from files and pages, through the metagraph, to normalized knowledge-graph entities and relationships. section 6.1.1
