Introduction to Labeled Property Graphs

Property Graphs and RDF

Jakob Voß

2024-11-22

Background

RDF

RDF data model

• subject: IRI or blank node
• predicate: IRI
• object: IRI or blank node or literal:
  Unicode string with either data type (IRI) or language tag

RDF data model: named graphs

• graph: IRI or blank node
• subject: IRI or blank node
• predicate: IRI
• object: IRI or blank node or literal:
  Unicode string with either data type (IRI) or language tag

RDF* data model

• Subject and object can be triples as well!
• Useful in theory, but complicated

Motivation: Semantic Web

I have a dream for the Web [in which computers] become capable of analyzing all the data on the Web – the content, links, and transactions between people and computers.
– Tim Berners Lee

• Information integration
• Common Ontologies
• Linked Open Data!

Motivation: Property Graphs

• SQL ↔︎ NoSQL
• Local databases
• Enterprise systems
• Efficient data storage and retrieval

Practice

Theory again

• Several academic papers how to map
  RDF graphs and property graphs
• Everything can be mapped to everything anyway

Example

<Padmé>  a <person> ; <gender> "female" .
<Anakin> a <person> ; <gender> "male" .
<Luke>   a <person> ; <gender> "male" .
<R2D2>   a <robot> .
<C3PO>   a <robot> .

<Padmé>  <owns> <R2D2> .
<Anakin> <owns> <R2D2> .
<Anakin> <child> <Luke> .
<Padmé>  <child> <Luke> .
<Luke>   <owns> <R2D2> .
<Luke>   <owns> <C3PO> .

Example graph

flowchart LR
   male([male])
   female([female])
   Padmé -- a --> person
   Anakin -- a --> person
   Luke -- a --> person
   R2D2 -- a --> robot
   C3PO -- a --> robot 
   Padmé -- owns --> R2D2
   Anakin -- owns --> R2D2
   Luke -- owns --> R2D2   
   Luke -- owns --> C3PO   
   Anakin -- child --> Luke
   Padmé -- child --> Luke   
   Padmé -- gender --> female
   Luke -- gender --> male
   Anakin -- gender --> male

Example graph without rdf:type

flowchart LR
   male([male])
   female([female])
   Padmé -- owns --> R2D2
   Anakin -- owns --> R2D2
   Luke -- owns --> R2D2   
   Luke -- owns --> C3PO   
   Anakin -- child --> Luke
   Padmé -- child --> Luke   
   Padmé -- gender --> female
   Luke -- gender --> male
   Anakin -- gender --> male

Some limitations of RDF

• No local properties but IRIs
• Special property rdf:type (= a) instead of node labels
• Statements can’t have statements (unless RDF*)

Example graph in RDF*

<Padmé>  <owns> <R2D2>   {| <episode> 1 |} .
<Anakin> <owns> <R2D2>   {| <episode> 2 |} .   
<Anakin> <parent> <Luke> {| <episode> 3 |} .
<Padmé> <parent> <Luke>  {| <episode> 3 |} .
<Luke>  <owns> <R2D2>    {| <episode> 4 |} .

SPARQL query

SELECT ?p {  
  BIND( << ?p <owns> <R2D2> >> AS ?e )
  FILTER ( e.episode >= 2 )
}

SPARQL RDF* vs Cypher

SELECT ?p {  
  BIND( << ?p <owns> <R2D2> >> AS ?e )
  FILTER ( e.episode >= 2 )
}

MATCH (p)-[e:owns]->({name:"R2D2"})
WHERE e.episode >= 2 RETURN p.name

Comparison

RDF	Property Graphen
nodes and edges	nodes, edges, properties
standard (W3C), but…	standard (ISO), but…
global IRIs	local references
either schema-less…	either schema-less…
…or shared ontologies	…or local schema
optional inference and rules	strict schemas and custom rules
SPARQL	Cypher

Discussion

• What do you think?
• Questions?

When to use which?

↑ Performant & Scalable

• Structured Property Graphs
• Schema-less Property Graphs
• RDF Triple Stores

↓ Flexible

But RDF also supports data integration!

Summary

• Property graphs are popular in enterprise data management
• Try out new stuff
• All data modeling starts with “boxes with arrows and stuff”