Typed Property Graph Model

• Typed property graphs are formal data models with explicitly defined schema types and constrained properties, ensuring enhanced data quality and expressiveness.
• They employ static semantics, inheritance, and subtyping to enforce rigorous schema validation and maintain type safety in complex data relationships.
• Utilizing DDL constructs and algebraic formalisms, typed property graphs integrate relational, NoSQL, and knowledge graph paradigms for robust, scalable implementations.

A typed property graph is a formal data model in which each node and edge is explicitly associated with a type defined in a schema, and whose properties are constrained by type-specific rules, cardinalities, and potentially a type hierarchy. This model strengthens data quality, enables reasoning about structure and constraints, and provides a foundation for both expressive queries and robust data integration. Typed property graphs subsume or generalize untyped property graph models by adding schema binding, static type-checking, inheritance, and other features drawn from database, programming language, and knowledge representation traditions. Multiple formalizations exist: algebraic, categorical, functional, and DDL-based, many of which have influenced current standards and industrial practice.

1. Formal Definitions and Core Structure

The typed property graph model (TPG) is typically defined by a pair of levels: schema and instance.

• Schema (Typed Graph Schema, TGS):
  • $N_S$ — a finite set of node types, each with label and data-type;
  • $E_S$ — a finite set of edge types, each with label, property type, and incidence constraints;
  • $T$ — set of allowed data types (scalar or structured);
  • $\rho : E_S \to \wp(N_S) \times \wp(N_S)$ — domain/range specification for edge types;
  • $$\tau : E_S \times N_S \to \mathbb{N}_0 \times \mathbb{N}$$ — min/max multiplicities for node/edge incidences;
  • $C$ — additional integrity constraints.
• Instance (Typed Property Graph, TPG or TGM):
  • $N$ — set of instance-nodes;
  • $E$ — set of instance-edges;
  • $N_S$0 — homomorphism assigning each graph element a type, preserving incidence and multiplicity as per $N_S$1 and $N_S$2;
  • Properties, labels, and type assignments are enforced as declared in the schema (Laux, 2021, Laux, 2021).

This formalization supports hyper-nodes and hyper-edges: edge types may be $N_S$3-ary (connecting sets of node types) and node types may encapsulate subgraphs as (typed) values.

2. Typing Mechanisms: Static Semantics, Subtyping, and Inheritance

Typed property graphs provide static semantic guarantees by associating schema-defined types to graph elements and enforcing:

• Type Safety: Every node and edge has exactly one (possibly multi-inherited) type, with all property values conforming to declared data types.
• Domain/Range & Multiplicity: Edges may only connect node types as allowed by $N_S$4, and the number of such incidences per node is bounded as specified by $N_S$5.
• Structured and Container Types: Properties may be scalars (e.g. INT, STRING), records, sets, lists, etc., admitting strict schema-based validation (Wu, 2018).
• Inheritance and Multi-inheritance:

Type systems such as that of PG-Schema allow unions, intersections, optional labels, and open/closed specification, producing a subtyping lattice with intersection and union operations as algebraic type combinators (Angles et al., 2022).

Subtyping is generally set-theoretic: if $N_S$6, then an element typed $N_S$7 also conforms to $N_S$8's constraints.

3. Data Definition Languages and Formal Syntax

Typed property graphs are commonly specified and manipulated via declarative DDLs, which provide constructs for nodes, edges, graphs, types, and labels. Examples:

• BNF/EBNF Statements:
  • Vertex type:
  • $E_S$7
  • Edge type:
  • $E_S$8
  • Inheritance and abstract types:
  • $E_S$9
• Schema Validation and Constraints:

Constraints cover primary keys, edge source/target closure, property nullability, and complex participation or key constraints (e.g., via PG-Keys: uniqueness, mandatory, singleton/foreign key-style patterns) (Angles et al., 2022, Wu, 2018).

• Schema Evolution and Instances:

Multiple graphs may share or have private type-containers, supporting both global typing and instance-specific subgraphs, with backward-compatible evolution supported via ALTER operations (Wu, 2018).

4. Algebraic and Categorical Models

Typed property graphs are given categorical and algebraic semantics to unify database, functional-programming, and logico-mathematical approaches:

• Algebraic Property Graphs (APG):
  • Types are built inductively using sum (union), product (tuple), and primitive types.
  • Schemas are $N_S$9-signatures $E_S$0 where each label $E_S$1 is assigned a type expression.
  • Instances are functors $E_S$2 preserving finite products and coproducts, ensuring set-theoretic and type-theoretic soundness.
  • Schema mappings are functors, inducing data migration via Kan extensions and preserving integrity “by construction” (Shinavier et al., 2019).
• Functional Approach (Simply-Typed λ-Calculus):
  • Vertices, edges, and properties are types or (possibly multi-valued) functions;
  • e.g., $E_S$3, $E_S$4.
  • Queries and transformations are $E_S$5-terms over these types, and type-checking is baked into query compilation (Pokorny, 2018).

5. Integrity Constraints and Validation

Typed property graph schemas support strong constraints at both element and pattern level:

• Property Key Constraints: Enforce uniqueness, existence, and referential integrity for properties and edge patterns (PG-Keys: EXCLUSIVE, SINGLETON, MANDATORY).
• Participation Constraints: Specify mandatory involvement in certain edge types for a node class.
• Multiplicity and Existence-Dependency: Typed edges enforce min–max patterns for relationship incidence, prohibiting orphaned or over-connected entities.
• User-Specified Constraints: Arbitrary logical rules may be added for validity, interpreted during data manipulation or transaction commit (Angles et al., 2022, Laux, 2021).

Validation typically occurs at data load time or upon mutation, with conformance checks ensuring that all graph elements adhere to their declared types and the schema's overall structure and key constraints.

6. Implementation Strategies and Integration with Other Models

Typed property graph models are implemented both natively in specialized systems and as overlays on relational or NoSQL backends:

Implementation Context	Features	Trade-offs
Native graph DBMS (e.g. GQL)	Direct schema binding, property traversals	May lack mature SQL/relational support
Relational DBMS overlay	One table/type per node/edge type; constraints as CHECK/PK/FK	Leverages SQL, but may require recursive CTEs for traversals (Crowe et al., 2024)
Knowledge graph mappings (SPG)	Import RDF/OWL ontologies; ontology-driven validaton	Enables hybrid reasoning, scalable analytics (Purohit et al., 2020)

Typed property graphs can be projected from or to relational, RDF/OWL, or object-oriented data models via computable, semantics- and information-preserving mappings, thus serving as a supermodel for data integration. Hyper-edges and hyper-nodes directly encode relationships and encapsulations that would require auxiliary constructs in other paradigms (Laux, 2021, Laux, 2021).

7. Query Languages, Expressiveness, and Standardization

Pattern-based querying in typed property graph systems is type-driven:

• Type-Safe Pattern Matching: Syntax and semantics of pattern languages (Cypher, GQL, GPC) enforce variable typing, disallow ill-formed queries, and guarantee that atomic variables are only bound to node/edge elements according to schema (Francis et al., 2022).
• Typing Judgments: Each query expression is assigned a type environment, and operations like union, repetition, and filtering have specific typing rules to enforce structural soundness.
• Integration with Relational and Functional Queries: Queries over typed property graphs are compilable into relational algebra (e.g., via recursive CTEs in SQL) or as type-checked $E_S$6-terms, enabling deep integration with traditional database and functional programming analytics (Pokorny, 2018, Crowe et al., 2024).
• Ongoing Standardization: Emerging ISO standards (GQL, SQL/PGQ) are converging towards typed property graph semantics, promoting uniform DDLs, pattern queries, schema validation, and type-driven error reporting (Angles et al., 2022, Francis et al., 2022).

---

References:

• (Pokorny, 2018, Wu, 2018, Shinavier et al., 2019, Purohit et al., 2020, Laux, 2021, Laux, 2021, Francis et al., 2022, Angles et al., 2022, Crowe et al., 2024)