Part IV · Ch. 17 — Knowledge Graph + Reasoner

draft

Symbolic · Structured knowledge · Graph + description logic (OWL/RDF). Represents facts as entities and relations and infers new facts by logic. Card: ../02-types-of-ai/01-symbolic.kmd.

🎨 Figure F-IV.17.0Entities and relations. Brief: graph with nodes (people, places, concepts) and labeled edges; a newly inferred edge highlighted (deduction).

Knowledge graph + reasoner

1. Definition and short history

Structured knowledge bases with formal semantics. Lineage: Cyc (1984) → semantic web (RDF/OWL) → Google Knowledge Graph and Wikidata.

2. Foundations

  • Description logic — OWL, inference of classes/relations.
  • Graph theory — structure and traversal.
  • Linguistics / ontology — meaning and taxonomies.
  • Epistemology — explicit and verifiable knowledge.

3. Algorithms and architectures

  • Model — triples (subject, predicate, object) in RDF.
  • Reasoner — inference by description logic; rules (SPARQL/Datalog).
  • KG embeddings — TransE/etc. for link prediction (bridge with ML).
  • Query — SPARQL, graph traversal.

4. Inputs

  • Hardware: CPU; large graphs in triple stores.
  • Data: curated/extracted facts; ontologies.
  • Data structures: triple store, graph indexes, KG embeddings.
  • Systems: graph databases (Neo4j), triple stores (Blazegraph).

5. Specialized lifecycle

Stage Specialization
0 Problem Verifiable knowledge, data integration, semantic search
1 Data Fact extraction/curation; ontology definition
2 EDA Completeness, consistency, link quality
3 Modeling Ontology (OWL), inference rules
4 Construction Populate the graph; entity resolution; (optional) embeddings
5 Evaluation Fact accuracy, coverage, link prediction
5.5 Acceptance Logical consistency, source validation
6 Production Queries/inference; feeds search and RAG (ch. 31)
7 Monitoring Freshness, broken facts, conflicts
8 Maintenance Update facts/ontology
9 Governance Provenance, coverage bias, privacy

6. Capabilities, modes and modalities

Intellectual/factual: semantic search, data integration, factual QA, GraphRAG; auditable knowledge with provenance.

7. Limits, risks and ethics

Curation cost; incompleteness; coverage bias; hard entity resolution. Combines well with LLMs (factual grounding).

8. State of the art and examples

Wikidata, enterprise Knowledge Graphs, GraphRAG; KG embeddings and LLMs that query graphs to reduce hallucination (links to ch. 31 and 32).