�ݺ�ߣ

Two graph data models
RDF and Property Graphs
Andy Seaborne
Paolo Castagna
andy@a.o, castagna@a.o

Introduction
This talk is about two graph data models
(RDF and Property Graphs), example of a
couple of Apache projects using such data
models, and a few lessons learned along the
way.

Graph Data Models
➢ RDF
● W3C Standard
➢ Property Graphs
● Industry standard

RDF
➢ IRIs (=URIs), literals (strings, numbers, …),
blank nodes
➢ Triple => subject-predicate-object
● Predicate (or property) is the link name : an IRI
➢ Graph => set of triples

prefix : <http://example/myData/>
prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
prefix foaf: <http://xmlns.com/foaf/0.1/>
# foaf:name is a short form of <http://xmlns.com/foaf/0.1/name>
:alice rdf:type foaf:Person ;
foaf:name "Alice Smith" ; # ; means “same subject”
foaf:knows :bob .
:alice
foaf:knows
"Alice Smith"
foaf:name
foaf:Person
rdf:type
:bob

:bob rdf:type foaf:Person ;
foaf:name "Bob Brown" .
"Bob Brown"
foaf:Person
rdf:type
:bob

:alice rdf:type foaf:Person ;
foaf:name "Alice Smith" ;
foaf:knows :bob .
:bob rdf:type foaf:Person ;
foaf:name "Bob Brown" .
:alice
foaf:knows
"Alice Smith"
foaf:name
foaf:Person
rdf:type
"Bob Brown"
foaf:Person
rdf:type
:bob

RDFS
prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#>
foaf:Person rdfs:subClassOf foaf:Agent .
foaf:Person rdfs:subClassOf
<http://www.w3.org/2003/01/geo/wgs84_pos#SpatialThing> .
foaf:skypeID
rdfs:domain foaf:Agent ;
rdfs:label "Skype ID" ;
rdfs:range rdfs:Literal ;
rdfs:subPropertyOf foaf:nick .

RDF : Access
➢ SPARQL : Query language
➢ Protocol : over HTTP
PREFIX : <http://example/myData/>
PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
PREFIX foaf: <http://xmlns.com/foaf/0.1/>
## Names of people Alice knows.
SELECT * {
:alice foaf:knows ?X .
?X foaf:name ?name .
}

RDF : Access
➢ SPARQL : Query language
SELECT ?name ?numFriends {
{ SELECT ?person (count(*) AS ?numFriends) {
?person foaf:knows ?X .
} GROUP BY ?person
}
?person foaf:name ?name .
} ORDER BY ?numFriends

RDF : Access
➢ SPARQL : Update language
PREFIX : <http://example/myData/>
INSERT DATA {
:bob foaf:name "Bob Brown" ;
foaf:knows :alice
} ;
INSERT { :alice knows ?B }
} WHERE {
:bob knows ?B
}

Apache Jena
TLP: April 2012
➢ Involvement in standards
➢ RDF 1.1, SPARQL 1.1
➢ RDF database
➢ SPARQL server
Other RDF@ASF:
➢ Any23, Marmotta, Clerezza, Stanbol, Rya

Property Graph Data Model
A property graph is a set of vertexes and edges with
respective properties (i.e. key / values):
➢ each vertex or edge has a unique identifier
➢ each vertex has a set of outgoing edges and a set of incoming edges
➢ edges are directed: each edge has a start vertex and an end vertex
➢ each edge has a label which denotes the type of relationship
➢ vertexes and edges can have a properties (i.e. key / value pairs)
Directed multigraph with properties
attached to vertexes and edges

Property Graph: Example
id = 1 id = 2
name = “Alice”
surname = “Smith”
age = 32
email = alice@example.com
...
name = “Bob”
surname = “Brown”
age = 45
email = bob@example.com
...
since = 01/01/1970
...
id = 3
knows

Apache Spark: GraphX*
// Creating a Graph
val vertexes: RDD[(VertexId, (String, String))] =
sc.parallelize (Array((1L,("Alice", "alice@example.com")), (2L,("Bob", "bob@example.com"))))
val edges: RDD[Edge[String]] =
sc.parallelize(Array(Edge(1L, 2L, "knows"))
val graph = Graph(vertexes, edges)
...
Example of parallel graph algorithms available:
// Find the triangle count for each vertex
val triCounts = graph.triangleCount().vertices
// Find the connected components
val cc = graph.connectedComponents().vertices
// Run PageRank
val ranks = graph.pageRank(0.0001).vertices
* GraphX is in the alpha stage

Property Graphs @ASF
➢ Apache Tinkerpop (incubating)
➢ Apache Spark > GraphX
➢ Apache Giraph
➢ Apache Flink > Gelly

Use Case for Graphs
➢ Analytics
● Social networks and recommendation engines
● Data center infrastructure management
➢ Knowledge Graphs
● Happenings: people, places, events
● Customer databases / products catalogues

Some Conclusions
➢ Data Graphs are (still) new to many people
➢ RDF emphasizes information modelling
→ Knowledge graphs
→ SQL-like query
➢ Property Graph emphasizes data processing
→ Data capture
→ Graph analytic algorithms
➢ Naive layering of data models leads dissatisfaction
→ Can only mix toolsets by knowing it’s layered
➢ Could share technology
→ Storage, data access, query algebra

�ݺ�ߣ

Two graph data models : RDF and Property Graphs

More Related Content

Two graph data models : RDF and Property Graphs