Knowledge Graph

Nodes represent the discrete real-world objects, concepts, or events that form the graph's vocabulary. Unlike text strings, these are unique, disambiguated identifiers (URIs) that resolve to a single, specific thing.

• Person Node: A specific individual, not just a name string.
• Place Node: A canonical location with geo-coordinates.
• Event Node: A specific conference or historical occurrence.
• CreativeWork Node: A specific book, article, or video.

Each node acts as a deterministic anchor point, preventing the ambiguity that plagues keyword-based systems.

Edges define the directed, labeled relationships between nodes, forming the graph's intelligence layer. These predicates are not arbitrary links; they are typed connections from a formal ontology that a machine can traverse and reason over.

• bornIn: Connects a Person to a Place.
• worksFor: Connects a Person to an Organization.
• hasPart: Connects a CreativeWork to another CreativeWork.
• reviewedBy: Connects a Product to a Review.

By traversing these edges, a system can infer implicit knowledge, such as 'this person is a colleague of that person' through shared organizational affiliation.

A knowledge graph is governed by a strict ontology—a formal specification of the types of entities and relationships allowed within a domain. This schema enforces logical consistency and enables automated reasoning.

• Domain Constraints: An edge hasCapital can only originate from a Country node.
• Range Constraints: An edge hasCapital can only point to a City node.
• Cardinality: A Person has exactly one dateOfBirth.
• Transitive Properties: If A isLocatedIn B, and B isLocatedIn C, the system infers A isLocatedIn C.

This rigidity is what distinguishes a knowledge graph from a loose property graph, making it suitable for deterministic AI grounding.

The power of a graph is its ability to collapse multiple surface-form mentions into a single, canonical node. This process, known as entity reconciliation, resolves 'New York', 'NYC', and 'The Big Apple' to the same unique identifier.

• sameAs Linking: Explicitly connects a local entity to its authoritative Wikidata or DBpedia URI.
• Deduplication: Prevents the creation of duplicate nodes for the same real-world object.
• Contextual Disambiguation: Uses surrounding relationships to determine if 'Mercury' refers to the planet, the element, or the singer.

This ensures that all facts about a single entity are consolidated, creating a single source of truth for AI models.

Knowledge graphs are serialized into formats designed for machine consumption and interchange, not just human readability. These formats encode the triples (subject-predicate-object) that form the graph's atomic units.

• JSON-LD: The W3C standard for embedding graph data in web pages, using @context to map terms to IRIs.
• RDF/XML: A foundational, verbose serialization of the Resource Description Framework.
• Turtle (TTL): A compact, human-friendly syntax for writing RDF triples.
• N-Triples: A line-based, plain-text format for representing a single triple per line.

These serializations allow search engines and AI agents to ingest the graph as a set of deterministic facts without parsing unstructured text.

A knowledge graph's ultimate value lies in its capacity for logical inference. By applying rules to the explicit facts, new, implicit knowledge is derived. This is the engine of semantic reasoning.

• Rule-Based Reasoning: If X isMarriedTo Y, and Y isParentOf Z, infer X isStepparentOf Z.
• Path Queries: Traverse multiple edges to answer complex questions like 'Which companies have board members who are also professors at MIT?'
• Subsumption: If an ontology states 'Dog is a subclass of Mammal', any instance of Dog is automatically inferred to be an instance of Mammal.

This capability allows a graph to answer questions it was never explicitly programmed to answer, providing a rich, dynamic context for grounding large language models.

A formal, explicit specification of a shared conceptualization. Unlike a simple taxonomy, an ontology defines the types, properties, and interrelationships of entities within a domain, enabling logical inference. In knowledge graph engineering, ontologies allow systems to derive new facts—if a 'Person' has a 'worksFor' relationship with an 'Organization', the inverse 'employee' relationship is automatically inferred.

The process of matching a named entity reference in a dataset to a unique, canonical identifier in a target knowledge base, such as a Wikidata Q-ID or a Google Knowledge Graph MID. This resolves ambiguity—for example, determining whether 'Mercury' refers to the planet, the element, or the car brand—and consolidates disparate authority signals into a single, definitive node.

A critical Schema.org property used to perform explicit entity reconciliation. By linking an entity to its corresponding canonical URI on an external authoritative knowledge base (e.g., a Wikipedia page or Wikidata entry), the sameAs property provides an unambiguous signal to search engines that two references describe the identical real-world entity, strengthening the entity's identity in the knowledge graph.

A Schema.org property used to explicitly indicate the primary entity that a webpage is about. For a product page, the mainEntity would be the Product schema. This provides a strong, direct signal to search engine parsers for entity extraction, helping them disambiguate the page's core topic from secondary content like navigation, sidebars, or related articles.