Knowledge Graph (KG)

In a KG, real-world objects or abstract concepts are represented as entities (nodes). Unlike rows in a table, each entity has a unique, resolvable identifier (URI/IRI). This allows the graph to distinguish between the string 'Paris' and the entity representing the capital of France. Every entity is a container for properties and the anchor point for relationships, enabling fine-grained entity typing and disambiguation at the data model level.

The connections between entities are not just lines; they are typed, directed edges that form the predicate of a triple (Subject-Predicate-Object). Instead of a generic 'related_to' link, a KG uses specific predicates like foundedBy, headquarteredIn, or isA. This semantic rigor allows machines to traverse the graph with logical precision, enabling complex reasoning and inference.

A KG's intelligence is driven by its ontology (T-Box), which defines the classes, properties, and constraints governing the data (A-Box). Using standards like RDFS and OWL, the ontology enforces logical consistency. For example, an ontology can declare that the property hasCEO can only connect an instance of Organization to an instance of Person, preventing nonsensical relationships and enabling automated validation.

KGs are built on the principle of explicit context. The meaning of data is not hidden in application logic or column headers; it's part of the graph itself. This is achieved through serialization formats like RDF (Resource Description Framework), which structures all information as unambiguous triples. This self-describing nature makes KGs inherently interoperable, allowing disparate systems to consume and understand the data without prior coordination.

A defining characteristic is the ability to derive new, implicit knowledge from asserted facts using a reasoner. By applying ontological rules (e.g., hasParent + hasBrother -> hasUncle), the graph can automatically materialize transitive, symmetric, or inverse relationships. This deductive power transforms a KG from a static store into a dynamic system that uncovers hidden connections and validates logical consistency at scale.

Unlike relational databases that simulate joins at query time, KGs are stored in graph-native databases (triplestores) optimized for interconnected data. They are accessed via declarative graph query languages like SPARQL or Cypher, which are designed for pattern matching across vast networks. This architecture avoids the performance cliff of expensive table joins, making deep traversals—such as finding all connections within six degrees of a target entity—computationally efficient.