Course outline for Knowledge Graph Engineering

Module 1: Establish structure and identity

• Knowledge graph modelling: Map relational entities and keys into one local graph model while preserving identifiers, source evidence, and relationship meanings.
• Graph schemas: Define entity and relationship structure with stable keys, distinguishing schema rules from labels and individual accepted assertions.
• Graph databases: Select a local graph representation for supplied relationship queries and map its storage responsibilities to the relational source.

Practical outcome: Select one local graph model and map relational entities and keys to its schema.

Module 2: Resolve entities and inspect relationships

• Entity resolution: Apply explicit duplicate-matching criteria and hold ambiguous matches for human review instead of silently merging similarly named entities.
• Relationship modelling: Distinguish relationship, temporal, dependency, and causal edges so a connection or chronological sequence cannot silently imply causation.
• Querying: Answer multi-hop questions with source references and separate current accepted assertions from conflicting or retired historical evidence.

Practical outcome: Apply reviewed identity rules and query multi-hop relationships while distinguishing relationship, temporal, dependency, and causal edge meanings.

Module 3: Ingest and maintain reviewable assertions

• Knowledge graph construction: Validate candidate relationships before ingestion and verify that repeated source loading leaves entity and accepted-assertion counts unchanged.
• Knowledge graph maintenance: Supersede outdated assertions while retaining source and validity history, excluding retired claims from the current view.
• Source-qualified assertions: Attach source and validity fields to reviewed facts so conflicting evidence remains queryable without presenting every claim as accepted.

Practical outcome: Ingest accepted candidate facts with source and validity fields, then supersede an assertion without deleting its history.

Practical exercise

• Produce a local graph, repeatable ingestion script, identity-review report, and source-qualified assertion queries.
• Load valid relational records; reject an unsupported relationship and hold an ambiguous entity match for review.
• Rerun ingestion and verify unchanged entity and accepted-assertion counts; answer three supplied multi-hop queries.
• Query conflicting sources and retire an outdated assertion, preserving historical evidence while excluding it from the current view.