Knowledge graph glossary: 40 terms explained

A business knowledge graph connects the systems, records, and rules a company runs on into one queryable model, and the vocabulary around it gets used loosely. This glossary defines the 40 terms that come up most often, each in one plain, standalone sentence, so you can look one up and move on.

How to use this glossary

Terms are grouped alphabetically under letter headings, and each has its own anchor link so you can share or bookmark a single definition. Every entry is written to stand alone: you should be able to quote the sentence out of context and still have it make sense.

Where a term connects to a longer explanation, the definition links out to the relevant deep-dive post. If you are new to the topic, start with what a knowledge graph actually is before working through the list, or jump straight to how it works.

A

Audit trail {#audit-trail}

A chronological, unchangeable record of who changed what data, when, and through which system, used to reconstruct exactly how a reported number was produced. Read more on how a knowledge graph supports audit and lineage.

B

Bi-temporal modeling {#bi-temporal-modeling}

A data modeling approach that tracks two separate timelines for every fact: when something was true in the real world and when the system recorded it, letting you correct history without losing the original entry.

Business rule {#business-rule}

A stated policy, such as a discount threshold or an approval limit, that is stored as structured logic in the graph so every query applies it the same way instead of relying on memory or spreadsheet formulas. See how a knowledge graph stores business rules.

C

Confidence score {#confidence-score}

A numeric estimate attached to an answer that reflects how complete, fresh, and well-matched the underlying data is, so a reader can judge how much weight to put on the result.

D

Data quality {#data-quality}

The measurable condition of a dataset across completeness, accuracy, consistency, and timeliness, judged against what the business actually needs the data to do rather than an abstract ideal. See what good knowledge graph data quality looks like.

Data steward {#data-steward}

The person, often a department lead rather than a dedicated data role, responsible for deciding what a term means and approving changes to it within their area of the business.

Decision brief {#decision-brief}

A packaged answer to a business question that includes the finding, the supporting evidence with links back to source records, a confidence score, and a recommended next action.

E

Edge {#edge}

The named connection between two nodes in a knowledge graph, such as "invoiced to" or "reports to," that carries the meaning of the relationship rather than just its existence.

Embedding {#embedding}

A numeric vector representation of text or data that captures semantic similarity, allowing a system to find related items even when the wording differs, commonly used alongside a knowledge graph in retrieval systems.

Entity {#entity}

A distinct, identifiable thing the business tracks, such as a customer, a product, or an invoice, represented as a single node even when it appears differently across source systems.

Entity resolution {#entity-resolution}

The process of recognizing that records from different systems, such as "Acme Inc." and "Acme, Incorporated," refer to the same real-world entity and merging them into one node. Full explanation: entity resolution for business data.

F

Freshness lag {#freshness-lag}

The time gap between when a fact changes in a source system and when that change is reflected in the knowledge graph, a key measure of how current an answer actually is. Covered in knowledge graph maintenance.

G

Golden record {#golden-record}

The single, authoritative version of an entity's data that results after conflicting or duplicate records from multiple systems have been compared and reconciled into one accepted set of values.

Governance {#governance}

The set of decisions and accountable people that determine who can define a metric, approve a business rule, or grant access to a dataset within the graph. See who owns the knowledge graph in a smaller company.

Graph traversal {#graph-traversal}

The act of following edges from one node to another to answer a question, such as walking from a customer to its invoices to its payment history in a single connected path.

I

Intercompany elimination {#intercompany-elimination}

The accounting process of removing transactions between related entities inside the same group, such as one subsidiary billing another, before producing consolidated financial results.

K

Knowledge graph {#knowledge-graph}

A model of a business built from connected entities and the relationships between them, used to answer questions with traceable evidence instead of static reports. See the concepts glossary for related terms.

L

Lineage {#lineage}

The documented path a piece of data took from its original source, through any transformations, to the final answer it appears in, making it possible to trace a number back to where it came from.

M

Master data management {#master-data-management}

A discipline and set of tools focused on creating one consistent, trusted version of core reference data, such as customers or products, across an organization's systems. Compared in knowledge graph vs master data management, including when to build a system versus adopt one, a tradeoff also covered in the build vs buy cost breakdown.

Metric definition {#metric-definition}

The precise, agreed-upon formula and set of filters that determine how a business term, such as "active customer" or "revenue," is calculated, so different teams produce the same number. See how a knowledge graph handles conflicting definitions.

Multi-entity consolidation {#multi-entity-consolidation}

The process of combining financial and operational data from multiple legal entities or subsidiaries into a single, comparable view while preserving each entity's own detail. Explored in knowledge graphs for multi-entity and holding companies.

N

Node {#node}

A single point in a knowledge graph representing one entity, such as a customer, an order, or an employee, that can be connected to other nodes through edges.

O

Ontology {#ontology}

The formal set of entity types, relationship types, and rules that defines what kinds of things can exist in a knowledge graph and how they are allowed to connect. Full explanation: what is an ontology in a business knowledge graph.

Orphan node {#orphan-node}

An entity in the graph that has few or no connections to other entities, which usually signals a missing integration, a broken match, or incomplete source data.

P

Property {#property}

An attribute attached directly to a node or edge, such as a customer's signup date or an invoice's amount, that describes the entity without being a relationship to another entity.

Provenance {#provenance}

The record of where a specific fact originated, including the source system, the record identifier, and the time it was captured, distinct from lineage in that it points to a single origin rather than a full transformation path.

Q

Query {#query}

A structured question asked of the knowledge graph, expressed either directly or generated by an AI agent, that traverses nodes and edges to return a specific answer. See how AI agents use knowledge graphs to answer business questions.

R

RAG {#rag}

Retrieval-augmented generation, a technique where a language model pulls relevant text passages before generating an answer, distinct from a knowledge graph approach that pulls structured facts and relationships instead. Compared in knowledge graph vs RAG.

Read-only connection {#read-only-connection}

An integration method that allows a system to read data from a source application without any ability to write, edit, or delete records in that source. More in how it works.

Relationship coverage {#relationship-coverage}

A measure of how many of an entity's expected connections, such as a customer having at least one order or contact, are actually present in the graph.

Row-level permission {#row-level-permission}

An access control that restricts which individual records a user can see within a dataset, rather than granting or denying access to an entire table or system. Detailed in knowledge graph security and row-level permissions.

S

Schema {#schema}

The defined structure of tables, fields, and data types in a database or warehouse, narrower in scope than an ontology because it describes storage format rather than business meaning. Compared in knowledge graph vs data warehouse.

Semantic layer {#semantic-layer}

A layer that maps raw data fields to consistent business terms and calculations, so the same metric name returns the same result regardless of which report or tool queries it.

Survivorship {#survivorship}

The set of rules used to decide which value wins when multiple source records disagree about the same fact, such as preferring the most recently updated record or the most trusted source system.

T

Taxonomy {#taxonomy}

A hierarchical classification system, such as product categories nested under departments, that organizes entities into groups without necessarily defining how those groups relate to each other.

Temporal knowledge graph {#temporal-knowledge-graph}

A knowledge graph that stores not just current facts but the full history of when each fact was true, allowing a query to be answered as of any past date. Full explanation: temporal knowledge graphs.

Tenant isolation {#tenant-isolation}

An architectural guarantee that one customer's data, queries, and AI processing are kept completely separate from every other customer's, even on shared infrastructure. See security.

Triple {#triple}

The basic unit of a knowledge graph, made up of a subject, a predicate, and an object, such as "Customer A - purchased - Product B," that together form one fact.

V

Valid time {#valid-time}

The period during which a fact was actually true in the real world, as opposed to when it was entered into a system, a distinction that matters for correcting backdated transactions.

Vector database {#vector-database}

A database optimized for storing and searching embeddings by similarity rather than exact match, often paired with a knowledge graph so a system can find related text and confirm it against structured facts. See knowledge graph and private LLM deployment on Bedrock.

Key takeaways

Terms are grouped alphabetically under letter headings, and each has its own anchor link so you can share or bookmark a single definition.
The process of recognizing that records from different systems, such as \"Acme Inc.\" and \"Acme, Incorporated,\" refer to the same real-world entity and merging them into one node.
The process of combining financial and operational data from multiple legal entities or subsidiaries into a single, comparable view while preserving each entity's own detail.
The defined structure of tables, fields, and data types in a database or warehouse, narrower in scope than an ontology because it describes storage format rather than business meaning.
Start with what a knowledge graph is for your business, then review data quality practices, governance ownership, and security controls before comparing specific products.

FAQ

Is a knowledge graph the same thing as a database?

No. A database stores records in tables defined by a schema, while a knowledge graph stores entities and the named relationships between them, which makes it better suited to questions that span multiple systems.

Do I need to learn query languages to use these terms?

No. These definitions describe concepts a business user needs to evaluate or govern a knowledge graph, not the syntax used to write queries against one.

What is the difference between an ontology and a taxonomy?

A taxonomy organizes entities into a hierarchy of categories, while an ontology also defines the relationship types and rules connecting different kinds of entities, which is what a knowledge graph runs on.

Where should I start if I am evaluating a knowledge graph vendor?

Start with what a knowledge graph is for your business, then review data quality practices, governance ownership, and security controls before comparing specific products.

If you want to see these concepts applied to your own data, Try SIGNLD free or see how it works.

Knowledge graph glossary: 40 terms explained in plain English