A knowledge graph stores facts as a network: things are nodes, and relationships between them are labelled edges, such as "Ada wrote Notes" or "Paris is the capital of France". Because meaning is explicit and links can be followed, it joins data from many sources and supports questions that tables answer awkwardly.
Key concepts
- Triples: a fact as subject, predicate and object, the basic unit of RDF.
- Property graphs: nodes and edges that carry their own key-value properties, queried with graph query languages.
- Ontologies and schemas: the types of things and relationships allowed, and the rules between them.
- Identity: giving each entity one stable identifier so the same thing from two sources is merged.
- Reasoning: deriving new facts from rules, such as a subclass inheriting what its parent has.
Vocabulary
- RDF and SPARQL: the standard triple data model, and its query language.
- Entity resolution: deciding that two records describe the same real-world thing.
- Linked data: publishing data with global identifiers that point to other datasets.
- Taxonomy: a hierarchy of categories, simpler than an ontology.
What to study, in order
- Graph basics: nodes, edges, paths and traversals.
- Write twenty facts about a topic you know as triples and draw them as a graph.
- Load them into a triple store or graph database and query them with SPARQL or a property-graph language.
- Model a small domain as an ontology with classes, properties and constraints.
- Extract entities and relations from a few pages of text and add them to the graph, resolving duplicates.
- Add a rule or reasoner that infers new facts, and check them for mistakes.