Knowledge Graph

Data is structured as a network of nodes (entities) and edges (relationships), not rigid tables. This model prioritizes the connections between data points as first-class citizens.

• Nodes represent real-world objects, concepts, or events (e.g., a person, a drug, a location).
• Edges define the semantic relationships between nodes (e.g., EMPLOYED_BY, TREATS, LOCATED_IN).
• Properties are key-value pairs attached to both nodes and edges, storing attributes like a person's birth date or the start date of employment.

Edges carry a formally defined meaning or type, creating a network of qualified, directional connections. This allows for precise, multi-hop reasoning.

• Unlike a simple link, a relationship like HAS_SYMPTOM is a defined predicate with a specific domain and range.
• This enables complex pathfinding queries, such as: "Find all drugs that treat a disease which has a symptom of chest pain."
• The meaning is often defined by an external ontology, ensuring shared understanding across systems.

The graph is organized around unique, identifiable entities, not documents or tables. Each entity is a single, canonical node that consolidates all information about that thing.

• Entity Resolution is the critical process of merging data from multiple sources into a single, authoritative node.
• This eliminates data silos and provides a unified 360-degree view of a concept.
• A single node for "Ada Lovelace" would connect to her birth date, her education, her published works, and her familial relationships, all in one place.

The structure and meaning of the graph are defined in a formal, machine-interpretable language, often an ontology. This allows software agents to perform automated reasoning.

• RDF Schema (RDFS) and the Web Ontology Language (OWL) are W3C standards for defining classes, properties, and logical constraints.
• An ontology can declare that the relationship parentOf is the inverse of childOf, enabling an AI to infer new facts.
• This formal logic enables inference engines to derive implicit knowledge from explicit facts, a process called materialization.

Knowledge graphs are queried using precise, logic-based languages like SPARQL and Cypher, which return exact, verifiable results. This is a cornerstone of factual grounding for AI.

• A SPARQL query is a declarative statement of graph patterns to match, returning a set of bindings that satisfy the pattern.
• This contrasts with the probabilistic output of a large language model, providing a source of absolute truth.
• Graph RAG architectures use these deterministic queries to retrieve verified facts and inject them into an LLM's context window, eliminating hallucination.

Beyond explicit facts, a knowledge graph can derive new, implicit knowledge through logical rules. This is a key differentiator from a simple graph database.

• Rule-based reasoning applies IF-THEN logic. For example: IF a person was born in a city, AND that city is in a country, THEN the person is a citizen of that country.
• This allows a graph to answer questions based on logic, even if the final fact was never explicitly stored.
• Graph Neural Networks (GNNs) perform a form of learned, probabilistic reasoning over the graph structure to predict missing links or classify nodes.