What Makes a Station a Station: Node Properties and Labels

Opening Scene

A station on the map isn’t just an undifferentiated dot. It has a name, a category (major hub versus local stop), an accessibility rating, an opening year. All of that descriptive detail travels with the station itself, available the instant a rider taps on it, without needing to consult a separate reference somewhere else. The station’s identity and its description are inseparable — you can’t meaningfully talk about “the station” without also knowing what actually makes it that specific station.

Node properties and labels in a graph database carry this exact same descriptive richness.

In Plain English

A node label categorizes what kind of entity a node represents — Person, Product, Station — roughly analogous to a table name in a relational database. Node properties are the actual descriptive attributes attached directly to that node — a name, a category, a rating — stored right on the node itself, immediately available whenever that node is retrieved, without a separate lookup.

The Old Way

Designing genuinely useful node properties and labels has always required the same kind of thoughtful attribute design covered in relational and document modeling elsewhere on this site, applied specifically to a graph’s node-centric structure:

A node label groups similar entities together conceptually, supporting queries like “find all Person nodes” without needing to inspect every node in the graph individually.
Node properties should hold genuinely intrinsic attributes of the entity itself — a person’s name, a product’s price — rather than information that’s actually about a relationship, which belongs on an edge instead, a distinction covered more deeply in Article 4.
A node can carry multiple labels simultaneously in many graph databases, useful for entities that genuinely belong to more than one category — a person who is also an Employee, for instance — without needing separate, disconnected node representations.

Getting this right has always meant distinguishing genuinely intrinsic node attributes from relationship-specific attributes, a distinction that shapes a graph’s real clarity and query performance just as much as the choice of which entities to model as nodes in the first place.

What’s Changing (and Why AI Is the Reason)

AI-assisted entity extraction can identify genuine entities and their intrinsic attributes directly from unstructured text or existing structured data, proposing appropriate node labels and properties automatically. Rather than a modeler manually deciding which attributes belong on a node, AI-assisted analysis of source data can distinguish genuinely intrinsic entity attributes from relationship-specific information, informing a well-structured initial node design.
AI-assisted label consistency checking can identify inconsistent or redundant node labeling across a large graph, catching a genuine source of confusion where similar entities have accidentally been labeled differently by different contributors over time. This closes a practical gap specific to graphs built incrementally by multiple teams or automated processes, where labeling drift can quietly accumulate.
AI agents querying a graph rely on consistent, well-designed node labels and properties to correctly interpret what kind of entity they’re working with and what its genuine attributes actually mean, making this foundational design decision directly consequential for reliable agent-generated queries. An agent encountering inconsistently labeled or poorly attributed nodes has a genuinely harder time generating a correct query than one working against a clean, consistently structured graph.

The Metaphor, Fully Extended

Subway Element	Node Property and Label Concept
A station’s category — major hub versus local stop	A node label, categorizing what kind of entity a node represents
A station’s name, accessibility rating, and opening year, available the instant it’s selected	Node properties, descriptive attributes stored directly on the node
A station that’s both a subway stop and a regional rail connection point simultaneously	A node carrying multiple labels for an entity that genuinely belongs to more than one category
A cartographer distinguishing a station’s own attributes from details that are actually about a specific line passing through it	Distinguishing genuinely intrinsic node attributes from relationship-specific attributes that belong on an edge
A city planning office auditing station labels across the whole network for inconsistencies introduced over decades	AI-assisted label consistency checking catching inconsistent or redundant node labeling across a graph

For Beginners: What to Actually Do

Practice distinguishing a node label (what kind of entity this is) from node properties (the entity’s actual descriptive attributes).
Get comfortable with the idea that a node’s properties should be genuinely intrinsic to that entity, not information that actually describes a relationship to something else.
Before adding a property to a node, ask whether it describes the entity itself or actually describes a specific relationship that entity participates in.
Notice that consistent labeling across a graph matters more than it might first seem, since inconsistent labels make querying “all entities of this kind” genuinely unreliable.

For Practitioners and Leaders: The Deeper Layer

Use AI-assisted entity extraction to propose well-structured initial node labels and properties directly from unstructured or existing structured source data.
Use AI-assisted label consistency checking to catch labeling drift across a large graph built incrementally by multiple teams or automated processes.
Maintain consistent, well-documented node labeling conventions, since AI agents depend on this consistency to correctly interpret and query your graph.
Treat the distinction between node properties and edge properties as a genuine design discipline worth getting right early, since correcting it later across a large graph is real, non-trivial work.

Quick Recap

A node label categorizes what kind of entity a node represents, while node properties are the entity’s actual descriptive attributes, stored directly on the node.
Genuinely intrinsic attributes belong on a node; relationship-specific information belongs on an edge instead.
AI-assisted entity extraction can propose well-structured node labels and properties from source data, and AI-assisted label consistency checking can catch labeling drift across a large graph.
AI agents depend on consistent, well-designed node labels and properties to correctly interpret entities and generate reliable queries.

Where This Fits in the Series

Article 1 introduced the fundamental building blocks of a graph. This article covered what actually makes a station a station. Article 3 looks at why not every line runs both ways — directed versus undirected edges.