1. Towards Generating
Policy-compliant Datasets
Christophe Debruyne, Harshvardhan J. Pandit, Dave Lewis, Declan O’Sullivan
Trinity College Dublin – {first.last}@adaptcentre.ie
2019-01-31 @ ICSC
The ADAPT Centre is funded under the SFI Research Centres Programme (Grant 13/RC/2106) and is co-funded under the European Regional Development Fund.

2. www.adaptcentre.ieMotivation
• Datasets are created and used for a specific purpose, but
such data processing is the subject of various internal and
external regulations – e.g., GDPR. People must give their
informed consent for that purpose.
• Research question: Can we generate datasets for a specific
purpose “just in time” that complies with informed
consent?
• Unlike other initiatives (which we will discuss later), we
focus on consent information that has been stored, not the
context (e.g., via a form) nor the process.
1/31/19 2

3. www.adaptcentre.ieMethod
• R2DQB*, pronounced
R-2-D-cube, proposes
a method to annotate
RDF Data Cube data
structure definitions to
generate R2RML
mappings that will create
a RDF Data Cube dataset.
• How can we extend or adopt this method for generating
compliant datasets?
1/31/19 3
*Christophe Debruyne, Dave Lewis, Declan O'Sullivan: Generating Executable Mappings
from RDF Data Cube Data Structure Definitions. OTM Conferences (2) 2018: 333-350
• Data Structure Definitions
• Dimensions
• Measures
• Attributes
• References to tables
• References to columns
• Transformation functions
• …
Mapping
Engine
R2RML Mapping
R2RML
Processor
Data Cube Dataset
extended with
according
to
1
2
3
Validation 4
Provenance
Information
captured with
5

4. www.adaptcentre.ieMethod – The Model
• Modelled using Object Role Modelling (ORM) and translated into OWL
https://w3id.org/consent-mapping-jit
• Policy, Purpose, and Inclusion
• Consenting Party, Consent, Date and Times
• External uniqueness constraints indicate how inclusion and consent are
identified
• Integration of PROV-O allow for revisions of inclusions, consent, and policy
• Consenting Party ≡ prov:Person ⊔ prov:Organization
1/31/19 4

5. www.adaptcentre.ieMethod – Using the ontology
DESCRIBE ?consent WHERE {
?consent ont:forInclusion ?inclusion .
{ # GET THE LATEST INCLUSION OF PURPOSE FOR A POLICY
SELECT ?inclusion WHERE {
?inclusion ont:ofPurpose <.../purpose> .
?inclusion ont:ofPolicy <.../policy> .
<.../policy> dcterms:created ?dt . }
ORDER BY DESC(?dt) LIMIT 1 }
?consent ont:givenBy ?user .
?consent ont:registeredOn ?datetime .
# GET THE LATEST CONSENT INFORMATION FOR EACH USER
FILTER NOT EXISTS {
[ ont:forInclusion ?inclusion ;
ont:givenBy ?user ;
ont:registeredOn ?datetime2 ]
FILTER(?datetime2 > ?datetime)
}
}
1/31/19 5
Retrieving the
latest consent
information for a
specific purpose
of the latest
version of a
policy.
DESCRIBE query
returns a graph
which we will use
to manipulate the
dataset.

6. www.adaptcentre.ieExtending R2DQB
1. # PREFIXES OMMITED FOR BREVITY
2. @base <http://www.example.org/>
3. dct:identifier a rdf:Property, qb:DimensionProperty ;
4. rr:template "http://data.example.com/user/{id}" ;
5. rdfs:label "user id"@en ;
6. rdfs:subPropertyOf sdmx-dimension:refPeriod ;
7. rdfs:range owl:Thing .
8. foaf:mbox a rdf:Property, qb:MeasureProperty ;
9. rr:template "mailto:{email}" ;
10. rdfs:label "email address"@en ;
11. rdfs:subPropertyOf sdmx-measure:obsValue ;
12. rdfs:range owl:Thing .
13. <#dsd-le> a qb:DataStructureDefinition;
14. rr:tableName "user";
15. ont:forPurpose <http://data.example.com/purpose/8> ;
16. ont:forPolicy <http://data.example.com/policy/10> ;
17. qb:component [ qb:dimension dct:identifier ];
18. qb:component [ qb:measure foaf:mbox ] .
1/31/19 6
A simple data
structure
definition (DSD).
Yellow à
Annotations for
the generation of
an R2RML
mapping
Cyan à Linking
the DSD to a
purpose of a
policy

7. www.adaptcentre.ieExtending R2DQB
• The annotated Data Structure Definition is given to the R2DQB engine to
generate an R2RML mapping
• The R2RML mapping is executed (*) resulting in a graph G1
• We execute the DESCRIBE query introduced earlier, resulting in a graph
G2. This graph is used to create a list of consent instances (URIs) where
the property isGiven is true.
• We then use that list to apply the following query to G1 to create a graph
G1’ only retaining the information of people who have given their consent
DESCRIBE ?obs ?dataset WHERE {
?obs a qb:Observation .
?obs qb:dataSet ?dataset .
?obs dct:identifier ?dim .
VALUES ?dim { <uri1> … <urin> } }
1/31/19 7
We use the R2RML-F (in case the underlying DB technology does not support certain functions): Christophe Debruyne,
Declan O'Sullivan: R2RML-F: Towards Sharing and Executing Domain Logic in R2RML Mappings. LDOW@WWW 2016

8. www.adaptcentre.ieImplementation details
• Both consent information and endpoint are behind the
service, one just needs an annotated DSD. Governance
platforms can be adopted to guide one to identifiers for a
policy and purpose.
• Intermediate graphs that are generated allow for a
posteriori analysis and,
• Both governance problems are outside the scope of this
study and considered future work.
1/31/19 8

9. www.adaptcentre.ieDemonstration
• Due to a lack of data, our method was demonstrated using
a synthetic dataset. This dataset was created by simulating
people giving and withdrawing their consent. Consent can
also expire.
• Paper provides details about the
dataset (*).
• We believe we demonstrated the
viability of our approach, though
more experiments are called for.
1/31/19 9
(*) https://scss.tcd.ie/~debruync/icsc2019/

10. www.adaptcentre.ieRelated Work
• Both SPIRIT (Westphal et al. 2018) and SPECIAL (Kirrane et al. 2018)
studies have similar concepts (data subject, purpose, …), but aim to
analyze compliance a posteriori (both) or a priori (SPECIAL). Our goal was
to generate compliant datasets “just in time”
• Pandit, O’Sullivan and Lewis (2018) proposed an ontology for the
operational representation of informed consent and allows one to analyze
these representations w.r.t. annotated logs and questionnaires.
• Fatema et al (2017) proposed a model for representing the informed
consent an organization has obtained, but there is no explicit notion of
policies or support for revisions.
• Given the overlap in terminology, it is clear there is an opportunity in
aligning the vocabularies.
1/31/19 10

11. www.adaptcentre.ieSummary and Conclusion
• We argued that datasets are used for a purpose and that datasets should
be built suitable for a purpose, including any policies it should comply
with. Can we generate datasets for a specific purpose “just in time” that
complies with informed consent? The answer is yes. How?
1. By adopting a method for generating R2RML mappings from
annotated Data Structure Definitions that in turn generate datasets
2. By creating an ontology for informed consent (and revisions thereof)
and link aforementioned DSDs to purposes in policies
3. We use a knowledge base using that ontology to manipulate the
generated dataset – from (1) – retaining the data that is compliant
4. All intermediate graphs allow one to trace the various steps
1/31/19 11

12. www.adaptcentre.ieFuture work
A current limitation is a lack of evaluation beyond the
synthetic dataset created for the study.
We furthermore recognize the opportunities in aligning or
integrating our models and approach with related work
(ongoing).
See http://openscience.adaptcentre.ie/ for more of our
projects.
1/31/19 12