RDF goes meta

What (and what not) to expect of RDF-star

Eurecom - 2022-06-15

Pierre-Antoine Champin

Associate professor at Université Lyon 1
Specialized in knowledge representation and exchange on the web
Involved in several standardization groups
Since 2021, W3C fellow

RDF-star in a nutshell

Formerly RDF* [Hartig & Thompson 2014]
Extend RDF to ease the expression of statement-level metadata
Now a CG report

Why it matters

Alternatives suck
Fine-grained provenance
Explainable AI
Property graphs

Alternatives suck are not ideal

Verbosity
Loss of the "original" triple
Performances

Fine-grained provenance

Pierre-Antoine works for W3C *according* to the W3C staff page

Explainable AI

Property Graphs

An example property graph representing a person named 'Pierre-Antoine Champin' working for an organisation named 'World Wide Web Consortium' since February 2021

The RDF stack

The different layers of the RDF standard

RDF-star's abstract syntax

The different layers of the RDF standard, with abstract syntax hilighted

RDF's abstract syntax

Graph := Triple*
Triple := Subject Predicate Object
Subject := IRI | BlankNode
Predicate := IRI
Object := IRI | BlankNode | Literal

RDF-star's abstract syntax

Graph := Triple*
Triple := Subject Predicate Object
Subject := IRI | BlankNode | Triple
Predicate := IRI
Object := IRI | BlankNode | Literal | Triple

→ asserted triple / quoted triple

Turtle-star

The different layers of the RDF standard, with Turtle hilighted

Turtle-star


        

        << <#pa> ex:workingFor <#w3> >>

          prov:wasDerivedFrom

            <https://www.w3.org/People>.

According to some web page, Pierre-Antoine would be working for W3C.

Turtle-star


        <#pa> ex:workingFor <#w3>.

        << <#pa> ex:workingFor <#w3> >>

          prov:wasDerivedFrom

            <https://www.w3.org/People>.

According to some web page, Pierre-Antoine is working for W3C.

Turtle-star


        <#pa> ex:workingFor <#w3> {|

          prov:wasDerivedFrom

            <https://www.w3.org/People>;

        |}.

# Annotation syntax

Turtle-star


        <#pa> a schema:Person;

          schema:name "Pierre-Antoine Champin";

          ex:workingFor <#w3> {|

            schema:startDate "2021-02-01"^^xsd:date;

            prov:wasDerivedFrom

              <https://www.w3.org/People>;

          |}.

Pierre-Antoine is working for W3C, since 2021 and according to W3C.

Turtle-star & co.

TriG-star
N-Triples-star
N-Quads-star

SPARQL-star

SELECT ?s ?o ?src {

      << ?s ex:workingFor ?o >>

        prov:wasDerivedFrom ?src

    }

JSON-LD-star

The different layers of the RDF standard, with JSON-LD hilighted

JSON-LD-star


        {

          "@context": { ... }

          "@id": {

            "@id": "#pa",

            "workingFor": "#w3"

          },

          "from": "https://www.w3.org/People"

        }

JSON-LD-star


        {

          "@context": { ... }

          "@id": "#pa",

          "workingFor": {

            "@id": "#w3",

            "@annotation": {

              "from": "https://www.w3.org/People"

            }

          }

        }

RDF-star semantics

The different layers of the RDF standard, with semantics hilighted

RDF-star semantics

Layered above "plain RDF" semantics

Don't reinvent the wheel
RDF-star can (in theory) be implemented on top of a standard RDF system
Semantics extensions (RDF-S, OWL…) can be directly applied to RDF-star

Asserted triples can not be "cancelled"

Not in RDF, not in RDF-star (mononotic semantics)

Pierre-Antoine was working for ERCIM from 2020 to 2021, and is now working for W3C

Asserted triples can not be "cancelled"

Be careful when you convert PG edges to RDF-star...

Asserted triples can not be "cancelled"

... or accept a weaker vocabulary ("is, has been or will be working for")

Triples are "unique"

They are entierly defined by their subject, predicate and object

John has been working for ACME in 2019 and in 2021

Triples are "unique"

Multiple "occurrences" of the relationship must be reified

Require an explicit relationship between a quoted triple and its reifications

Triples are opaque


      << ex:superman ex:can ex:fly >> dct:source ex:dailyPlanet.

      ex:superman owl:sameAs ex:clark.

does not entail


      << ex:clark ex:can ex:fly >> dct:source ex:dailyPlanet.

Triples are opaque


      << <#pa> ex:workingFor <#w3> >> s:startDate "2021-02-01"^^xsd:date.

      ex:workingFor owl:inverseOf ex:staff.

Should this entail the following?


      << <#w3> ex:staff <#pa> >> s:startDate "2021-02-01"^^xsd:date.

Triples are opaque


      << <#pa> ex:workingFor <#w3> >> s:startDate "2021-02-01"^^xsd:date.

      ex:workingFor rdfs:subPropertyOf ex:involvedIn.

Should this entail the following?


      << <#pa> ex:involvedIn <#w3> >> s:startDate "2021-02-01"^^xsd:date.

Transparency-enabling properties

A property that is "inherited" by all "equivalent" triples
i.e. a property on the situation described by the triple rather than the triple itself

E.g.

dct:source is not transparency-enabling
schema:startDate is transparency-enabling:
schema:startDate a rdf-star:TransparencyEnablingProperty.

Conclusion

RDF-star extends RDF with the ability to talk about itself
“With great power comes great responsibility”
It is not cast in stone:
join the Community Group or the future Working Group