1 of 15

Exploring the Potential of Knowledge Graphs in DPIs

Milan Markovic

University of Aberdeen, UK

2 of 15

Semantic Web (Web 3.0)

  • A vision of extending World Wide Web to an internet of data rather than just an internet of documents
  • Key characteristics
    • Information is described in machine-understandable format
    • Information is linked to enable data discovery thus forming a large knowledge graph
    • Standardisation is achieved through common open data standards
    • Data representation supports deductive and inductive reasoning

​

3 of 15

Knowledge graphs (KGs)

  • We are still far from the Web 3.0 vision, however, semantic web technologies have been increasingly adopted in domain specific applications
    • Google, Amazon, Airbnb, eBay, IBM, Microsoft …
  • Different types of knowledge graphs exist
    • General Knowledge KGs (Wikidata, Dbpedia, Google Knowledge Graph)
    • Domain-Specific KGs (Gene Ontology, SNOMED CT)
    • Social KGs (Facebook Graph)
    • Enterprise Knowledge KGs

​

​

​

4 of 15

Building Knowledge Graphs

  • KGs may represent knowledge using different semantics
  • We will focus on KGs where semantics are formalized by ontologies using W3C OWL 2 and W3C RDF standards

​

+

=

Data

Ontology

Knowledge Graph

5 of 15

Exercise (50 mins)��https://tinyurl.com/KGsession

6 of 15

Resource Description Framework (RDF)

  • RDF is a W3C standard for representing information about resources on the Web in a directed, labeled graph data format
  • Data is described using a series of RDF statements - “triples”
  • Structure
    • Subject - The resource being described
    • Predicate – A relationship between the subject and the object
    • Object - Can be another resource or a data value (literal)
  • All resources and relationships are identified by a unique IRI
  • Literals specification is based on the XML Schema Datatypes

7 of 15

RDF Example

http://example.org/NationalDigitalIDSystem

http://example.org/BiometricAuthenticationModule

http://example.org/usesComponent

Prefix: <http://example.org/>

ex:usesComponent

ex: BiometricAuthenticationModule

ex: NationalDigitalIDSystem

“National Digital ID System”^^xsd:String

rdfs:label

ex:Aadhaar

rdf:type

8 of 15

Web Ontology Language (OWL)

  • W3C standard
  • Provides additional semantics for describing ontologies such as:
    • Object and data properties
    • Classes
    • Named Individuals
  • Supported by different tools (e.g., Protégé Ontology Editor), programming frameworks (e.g., Jena API), reasoners (e.g., HermiT), graph databases (e.g., GraphDB), etc.
  • Useful for data integration (e.g., using ontology alignment techniques)

9 of 15

Example OWL

ex: NationalDigitalIDSystem

ex: BiometricAuthenticationModule

ex:Aadhaar

rdf:type

owl:NamedIndividual

rdf:type

owl:Class

rdf:type

rdf:type

owl:ObjectProperty

rdf:type

ex:usesComponent

rdfs:range

rdfs:domain

ex:Module1

ex:usesComponent

rdf:type

10 of 15

Querying KGs

  • Semantic Query Languages can be used to query and manipulate KGs
  • W3C SPARQL is a standard query language for RDF

​

PREFIX ex: <http://example.org/>

​

SELECT ?system

​

WHERE {

?system a ex:NationalDigitalIDSystem.

?system ex:usesModule ?module.

?module a ex:BiometricAuthenticationModule.

}

?system

<ex:Adhaar>

11 of 15

Inferring New Knowledge

  • SPARQL can be also used to define inference rules and add new triples to the graph

PREFIX ex: <http://example.org/>

​

INSERT {

ex:Adhaar ex:hasEnhancedSecurity "true"^^xsd:boolean .

}

WHERE {

ex:Adhaar a ex:NationalDigitalIDSystem .

ex:Adhaar ex:usesComponent ?module .

?module a ex:BiometricAuthenticationModule .

}

12 of 15

Deductive Reasoning

  • Automatic entailment over KGs can be used to enhance query answering, consistency checking, etc.
  • OWL 2 enables formalisation of additional information to support deductive reasoning:
    • Class and Individual disjointness and equivalence
    • Class constraints (e.g., cardinality, domain and range)
    • Property features (e.g., symmetricity)
  • Users can also define their own inference rules
  • Different reasoners (HermiT, FACT++, Pellet) and rule languages (SHACL, SWRL, Datalog, etc.) are available to perform logic-based reasoning

​

13 of 15

Inductive Reasoning

  • Extracts generalised patterns from KGs
  • Used for KG completion and generating data insights
  • Popular techniques:
    • Graph Analytics (e.g., centrality, connectivity)
    • Knowledge Graph Embeddings
    • Graph Neural Networks
    • Symbolic Learning (e.g., rule mining)

14 of 15

Open Discussion (20 mins)

Form groups of four people and discuss how could KGs be potentially used to enhance DPIs.

Spend 10 mins on discussing the ideas and try to write at least three ideas on a post-it note.

We will discuss the ideas of each group during the last 10 mins.

15 of 15

Further Resources

Books

  • Knowledge Graphs - https://kgbook.org/

Web Standards

Conference Proceedings

  • International Semantic Web Conference
  • Extended Semantic Web Conference
  • Most web and AI conferences now include Semantic Web track, e.g., World Wide Web conference

Tools

  • Protégé Ontology Editor - https://protege.stanford.edu/

Public KG Endpoints

Ontology Repositories

  • Linked Open Vocabularies - https://lov.linkeddata.es/dataset/lov/
  • OBO Foundry - https://obofoundry.org/

​

​