Knowledge Graph

Nodes represent the fundamental entities in a domain—real-world objects, abstract concepts, or digital assets. Each node is a unique, identifiable instance.

• Examples: A specific person (Marie Curie), a place (Paris), an organization (CERN), or an event (1911 Nobel Prize).
• Identity Resolution: A critical function is linking disparate mentions of the same real-world entity (e.g., 'M. Curie' and 'Marie Skłodowska-Curie') to a single, canonical node.

Edges are the named, directed connections between nodes that give the graph its meaning. Unlike a simple foreign key, an edge explicitly defines the type of relationship.

• Predicates: Edges are often expressed as subject-predicate-object triples, e.g., [Marie Curie] --(discovered)--> [Radium].
• Directionality: Relationships are directional, creating a rich semantic flow: [Paris] --(isCapitalOf)--> [France] is distinct from its inverse.

Labels classify nodes and edges into formal categories, creating a schema that enables logical inference and querying. This is the ontology layer.

• Node Labels: Define the class of an entity, such as :Person, :City, or :ChemicalElement.
• Relationship Types: Define the nature of the connection, such as :BORN_IN, :EMPLOYED_BY, or :HAS_SYMPTOM.
• Inference: A system can infer that if [Marie Curie] has the label :Person and :Person is a subclass of :LivingThing, then Marie Curie is also a :LivingThing.

Both nodes and edges can store structured data as key-value pairs, providing rich context without cluttering the graph with unnecessary nodes.

• Node Properties: A node labeled :Person might have properties like name: "Marie Curie", birthDate: 1867-11-07, and wikidataID: Q7186.
• Edge Properties: An edge :MARRIED_TO between two :Person nodes could have properties like startDate: 1895-07-26 and `location:

The entire graph can be decomposed into atomic units of meaning called triples, conforming to the Resource Description Framework (RDF) standard.

• Structure: Every triple follows the pattern: Subject — Predicate — Object.
• Example: <MarieCurie> <discovered> <Radium> .
• Serialization: Triples can be serialized into formats like JSON-LD, Turtle, and RDF/XML, enabling interchange between systems. This is the foundation for linked data and the Semantic Web.

Knowledge graphs are queried not by joining tables, but by traversing paths through the network. This is the core operational advantage.

• Pathfinding: A query can ask, 'What are the names of all elements discovered by people born in Warsaw?' This traverses [Warsaw] --(BORN_IN)--> [Person] --(DISCOVERED)--> [Element] --(HAS_NAME)--> [Name].
• Languages: Specialized query languages like SPARQL (for RDF) and Cypher (for property graphs) are designed for this pattern-matching traversal.

A formal, explicit specification of a shared conceptualization within a domain. An ontology defines the types of entities, their properties, and the logical relationships between them (e.g., isA, hasPart). Unlike a simple taxonomy, an ontology encodes complex rules and constraints, enabling automated reasoning and inference. For example, an e-commerce ontology might define that a Product hasBrand Organization and that a Laptop isA Product which hasProcessor CPU.

A collaborative, community-driven vocabulary of standardized structured data schemas. It provides a shared collection of types (e.g., Person, Event, Product) and properties (e.g., name, startDate, offers) that webmasters use to mark up their pages. This markup creates a de facto knowledge graph for public web content, directly enabling rich results in search engines like Google's Knowledge Panels.

The natural language processing task of identifying textual mentions of real-world entities and mapping them to their unique identifiers in a knowledge graph. The process involves:

• Named Entity Recognition (NER): Detecting spans of text that refer to an entity.
• Entity Disambiguation: Resolving which specific entity is meant (e.g., distinguishing 'Paris, France' from 'Paris Hilton'). This process is critical for transforming unstructured text into machine-readable, linked data.

A standard model for data interchange on the Web, forming the foundational data structure of a knowledge graph. RDF represents information as triples: a subject, a predicate, and an object (e.g., <DaVinci> <painted> <MonaLisa>). This simple, graph-based data model allows for the merging of data from disparate schemas and enables powerful querying using languages like SPARQL.

A precursor to the modern knowledge graph, representing knowledge as a network of interconnected nodes (concepts) and arcs (semantic relations). Early semantic networks focused on hierarchical inheritance (e.g., a canary can fly because it isA bird). Modern knowledge graphs extend this concept with formal ontologies, logical constraints, and massive scale, moving from a cognitive model to a computational database.

A specialized database designed to index and query high-dimensional vector embeddings, which are numerical representations of data. While a knowledge graph stores explicit, symbolic facts, a vector database captures implicit semantic similarity. A hybrid approach often combines them: using a vector search to find conceptually similar nodes as a starting point, then traversing the explicit graph of relationships for precise, deterministic answers.