# Graph Design Patterns

# Graph Design Patterns

This page collects practical graph modeling patterns that tend to work well for Curiosity Workspace applications.

# Pattern: Entity-centric navigation

Design around “hub” entities users care about:

• customer, product, ticket, document, case

Ensure each hub has:

• stable key
• meaningful outgoing edges
• a small set of high-signal properties (for display/search)

# Pattern: Relationship as first-class signal

If users frequently ask “show items related to X”, model that as an edge:

• Ticket -> ForCustomer -> Customer
• Document -> Mentions -> Entity
• Case -> HasStatus -> Status

This enables:

• fast traversal and exploration
• graph-constrained search
• related facets

# Pattern: Attributes as nodes (selectively)

Convert a property into a node when you need:

• cross-type filtering (shared status taxonomy)
• navigation (“click status to see all items”)
• metadata on the attribute (owner, definition, lifecycle)

Don’t overuse this: many attributes are best kept as properties.

# Pattern: Mention → resolve → link

When important concepts appear inside text:

1. extract mentions via NLP
2. resolve to canonical entities (or create entities)
3. link into the graph

This pattern turns unstructured text into navigable structure.

# Pattern: Graph + search together

Search retrieves candidates; the graph provides context:

• use search to find starting nodes
• use graph traversal for “next clicks” and constrained neighborhoods
• use facets and related facets to keep results meaningful

# Pattern: Many-to-Many relationships

Model many-to-many relationships using edges with properties:

• User -> AssignedToProject -> Project: A user can be on many projects, and projects have many users.
• Use edge properties to store metadata about the relationship (e.g., role_on_project, date_assigned).

# Pattern: Hierarchical structures

Represent hierarchies (org charts, folder structures) using recursive edges:

• Employee -> ReportsTo -> Employee
• Category -> SubCategoryOf -> Category
• Use path queries to traverse up or down the hierarchy.

# Pattern: Versioned data and time-travel

Handle data that changes over time without losing history:

• Snapshot Pattern: Create a new node for each version, linked to a "canonical" node.
• Validity Dates: Use properties like valid_from and valid_to on nodes or edges.
• Activity Log: Model changes as a series of event nodes linked to the subject entity.

# Next steps

• Apply graph patterns to ingestion: Schema Design
• Optimize graph-constrained retrieval: Search Optimization