kg_ontology

What problem does it solve?

Resolves noisy entity and predicate surface forms into stable canonical node identities so your knowledge graph doesn’t fragment into duplicates and mismatched concepts across documents.

Core Features & Use Cases

• Canonical identity resolution for DKG: Collapses surface spans (e.g., SOW / “Statement of Work” / “sow”) into a single canonical entity/predicate identity suitable for downstream triplets and graph storage.
• Synset + hypernym driven disambiguation: Uses a multi-stage approach (synset augmentation + LLM selection) to pick best synsets and preserve hierarchical hypernym context for retrieval without exposing ontology scaffolding in the visible graph.
• BM25 vertical alignment without graph traversal: Injects synset and first-level hypernym tokens into bm25_text so cross-entity matching works via lexical retrieval rather than expensive topology traversal.
• Layer separation & provenance-only debug: Keeps evidence (surface phrases, candidate lists) separate from ontology winners, hiding debug structures (e.g., alias links, subClassOf, candidate senses) from the default user-facing graph.

Quick Start

Use the kg_ontology skill to convert extracted triplets into canonical (subject, predicate, object) node identifiers using WordNet synsets and hidden hypernym scaffolding for BM25 vertical alignment.