What is the difference between RDF and property graphs?

RDF (Resource Description Framework) and property graphs are two approaches to modeling graph data, each with distinct structures and use cases. RDF is a W3C standard designed for representing interconnected data on the web using triples (subject-predicate-object), such as <http://example.org/Alice> <http://example.org/knows> <http://example.org/Bob>. These triples use URIs for global uniqueness, enabling data integration across sources. Property graphs, used in databases like Neo4j, model data as nodes and relationships, both of which can have key-value properties. For example, a node labeled Person might have {name: "Alice", age: 30}, connected via a KNOWS relationship with {since: 2020}. This flexibility makes property graphs intuitive for many application developers.

The data models differ significantly. RDF focuses on open-world semantics, where missing information doesn’t imply falsity (e.g., if no data states Alice’s age, it’s treated as unknown). Property graphs typically assume a closed-world model, where only stored data is considered true. RDF relies on standardized query languages like SPARQL and ontologies (e.g., RDFS, OWL) for reasoning, while property graphs use traversal-optimized query languages like Cypher or Gremlin. For example, SPARQL can infer that ex:Student is a subclass of ex:Person, whereas property graphs require explicit modeling of such hierarchies. Additionally, RDF edges (predicates) are global and reusable, while property graph relationships are often specific to node pairs.

Use cases and tooling also vary. RDF excels in scenarios requiring interoperability, linked data, or semantic reasoning, such as integrating biomedical datasets using shared vocabularies. Tools like Apache Jena or Stardog handle RDF storage and inference. Property graphs are preferred for applications needing high-performance traversal, such as recommendation engines or fraud detection, where relationships and properties are queried efficiently. Databases like Neo4j or Amazon Neptune optimize for these workloads. While RDF’s standardized approach suits the semantic web, property graphs prioritize developer-friendly, application-specific modeling. Choosing between them depends on whether global data integration or localized, performant querying is more critical.